About This Guide

Author. Steve Wolf. Expert witness in firearms, pyrotechnics, fire, stunts, rigging, climbing systems, industrial accidents, and entertainment production safety. Thirty-five years of production and forensic work. Columbia University. Eleven patents. Twenty books.

For. Working expert witnesses in every discipline: engineers, physicians, accident reconstructionists, fire investigators, safety professionals, human factors specialists, economists who visit sites, and anyone whose opinion rests on something they saw and now have to prove. It assumes you have never held a camera. It does not assume you are a beginner at anything else.

Contents. Forty chapters in six parts. 602 checklist items across forty field checklists. One hundred forty-five numbered tips. Eight appendices, including templates, a glossary, and a list of the standards that do not exist.

Covers. Camera and lens fundamentals for a non-photographer. Light, color, and scale. Field protocol for scenes, injuries, parts, fire, water, height, darkness, and night. Drone work. Photogrammetry and laser scanning. Video, high-speed video, and surveillance recordings. Audio for witness interviews. Metadata, hashing, chain of custody, and archival storage. Authentication under Federal Rule of Evidence 901 and 902. Discovery, spoliation, and where the law lets you point a camera. Rule 403, the new Rule 107, animation versus simulation, courtroom display, and testifying about your own images.

Does not cover. Writing the report, negotiating the engagement, or surviving cross examination generally. Those belong to Being a Great Expert Witness. Building the underlying investigation belongs to Investigating Personal Injury Cases. This book stops at the moment the image goes into the report.

Currency. Revision 2026-07. Federal Rule of Evidence 702 current through the December 1, 2023 amendment. Federal Rule of Evidence 107 current through its December 1, 2024 effective date. Standards and technical guidance current through July 2026.

Disclaimer

This book is a guide for expert witnesses, attorneys, and claims professionals. It is not legal advice, and it is not a substitute for the judgment of retaining counsel. Rules of evidence, procedure, consent, privacy, and access vary by jurisdiction and change over time. Confirm every rule, statute, and standard against a primary source before you rely on it in a matter. Admissibility is for the court. Weight is for the jury.

Introduction: The Photograph You Did Not Take

Every imaging problem in litigation starts the same way. Somebody stood in a place where the evidence still existed, raised a camera, and made a set of choices. Two years later a room full of people who were not there has to decide what happened, using nothing but what that person chose to capture. The questions that follow are always the same four. What did the camera record? What did it miss? What did it distort? And can you prove which is which?

Most experts are hired for what they know. They are judged, in the end, on what they can show.

I have testified in cases decided by a single frame, and I have watched cases collapse over a frame nobody took. The pattern does not vary. The expert arrives at the scene with the right credentials and the wrong instrument, photographs what looks important, and leaves. Months later the case turns on the thing that did not look important: the wear pattern at the edge of the frame, the light fixture nobody framed, the fifteen feet of floor between the two photographs that were taken. The expert cannot go back. The scene has been repaired, the part has been scrapped, the surveillance recorder has overwritten itself, and the only version of the truth that still exists is the version that person happened to capture.

That is the whole problem. The scene decays and the file does not grow.

Four ideas run through everything in this book.

First, treat the camera as an instrument, not as a convenience. You would not report a torque value from a wrench you had never calibrated or a temperature from a probe you could not identify. A camera is a measuring device with a documented configuration, a known error, and a method that can be written down and repeated. Experts who treat it as a phone in their pocket get taken apart on that point alone.

Second, shoot everything to the higher standard, because you do not get to decide what your images are for. That decision gets made months later by lawyers looking at what survived. The frame you captured as a memory aid will be asked to carry a measurement, and it will either hold or it will not.

Third, capture is only the first third of the job. An image that cannot be authenticated never reaches the jury, no matter how good it is. An image that cannot survive discovery becomes a liability instead of an exhibit. An image that cannot be read by a juror in the back row does not exist. Capture, preserve, present. Fail any one and the other two were wasted.

Fourth, there is no certificate that makes your imaging reliable. The standards body that once governed forensic photography is defunct, its successor publishes guidance rather than binding standards, and there is no published American national standard for scene photography as of this writing. That is not a loophole. It is a burden. Under Federal Rule of Evidence 702 you will prove reliability with a documented, reproducible method that you can name and defend, and that is the discipline this book teaches.

The most dangerous photograph in your file is not the one you got wrong. It is the one you never took, because it cannot be attacked, cannot be explained, and cannot be replaced. Opposing counsel will find it, and the question will be short: "You didn't photograph that, did you?"

Take the picture. Take the one you think is pointless. Then read the rest of this book so the ones that matter hold up.

How to Use This Book

Each chapter opens with the reasoning and closes with a working checklist. The checkboxes work on screen and print cleanly on paper. Read the chapters once. Use the checklists every case.

Part One frames the work and tells you the honest state of the standards. Part Two teaches camera craft from zero, for a reader who has never held one. Part Three is field protocol, scene by scene and difficulty by difficulty. Part Four covers video, audio, and testing. Part Five covers preservation, authentication, discovery, and access. Part Six covers presentation and testimony.

If you are on a plane tonight and shooting a scene tomorrow, read Chapters 1, 6, 12, 14, and 15, print the checklists from Chapters 14 and 15, and read the rest when you get home.

Three kinds of callout appear throughout. A Practice Tip gives a concrete action. A Caution flags an exposure in discovery, on the stand, or in the record. A Field Note is my own policy or something I learned the expensive way.

One hundred forty-five numbered tips are scattered through the chapters in chapter-and-tip format, so tip 12.3 is the third tip in Chapter 12. All of them are collected in Appendix E. All forty checklists are consolidated in Appendix A, in the order a case runs, and are also published as a separate downloadable pack you can print, fill out, and put in the file.

Nothing in this book requires you to become a photographer. It requires you to become an expert who can prove what the camera did.

Part One. The Frame

Chapter 1Most Experts Photograph to Remember

Most experts photograph to remember. The good ones photograph to prove.

The difference does not show up on the day you shoot. It shows up eighteen months later, in a conference room, when opposing counsel slides a print across the table and asks you to identify the dark object at the lower left of the frame. If you photographed to remember, you say you believe it was a bracket. If you photographed to prove, you open the frames that show that bracket from three distances with a scale in its plane, and you say what it is, how large it is, and where it sat relative to everything else in the room.

The note you take and the exhibit you build

Two kinds of images come out of an expert's bag. The first is the memory aid: a visual note, taken so that you can reconstruct your own thinking later. It is aimed at you. The second is the exhibit: an image built from the first press of the shutter to be handed to a stranger who was not there, who does not trust you, and who is being paid to find the seam in your work. It is aimed at the record.

The memory aid loses. It loses because it was never asked to survive contact with an adversary, and every property that made it convenient at the scene made it weak in the file. It was shot on automatic, so you cannot state the settings. It has no scale, so nothing in it can be measured. It has no orientation frame, so it floats free of the room it came from. Its clock was never checked. It went to counsel through a text message and arrived with its image metadata stripped, so nothing in the file itself says when it was made or what made it.

What an evidentiary image has to do, in order

An evidentiary image does four jobs, and it does them in that order. It records, which means it fixes a condition at a moment in a form that can be authenticated later. It measures, which means a dimension can be pulled out of it and defended. It explains, which means a person with no technical training can look at it and understand what they are seeing before you say a word. It persuades, which is the only one of the four you are not permitted to optimize for.

That order is not a matter of taste. It is the order in which the image will be attacked. Authentication comes first, under Federal Rule of Evidence 901, and an image that cannot clear that bar never reaches the jury at all, no matter how well it explains. Measurement comes second, because measurement is where a competent opponent will look for your arithmetic. Explanation is third. Persuasion is last, and it is last because an image that persuades before it records is the definition of a misleading exhibit.

The standards body splits every photograph into two classes

The Scientific Working Group on Digital Evidence, the standards body whose current documents govern this work, draws the same line in different words. Its 2025 document on image categories in forensic science divides everything a camera produces into two classes. Documentation images record appearance, condition, or location and are not intended for scientific analysis; the working group accepts a file format with minimal compression for those. Examination quality images are intended for expert scientific analysis and comparison, and for those the working group requires a file format with lossless or no compression, meaning a raw file or the Tagged Image File Format, at the highest resolution the device offers.

Read the two definitions again and you will see the trap. The category is assigned by what the image is later used for, and you do not know at the scene what your images will later be used for. The photograph you took to show where a ladder was standing becomes, four months later, the only image that shows the foot of that ladder against the floor at all, and now somebody wants to measure the angle. It was captured as documentation. It is being asked to serve as examination.

1.1Capture every frame to the examination standard, because the category is assigned later by lawyers, not at the scene by you.

There is only one way out of that trap, and it costs nothing but memory card space. Shoot everything to the higher standard. The cost of a raw file is a few dollars of storage. The cost of a compressed file that turns out to be the only record of a fracture surface is the opinion built on it.

Forty sharp photographs and no method behind them

An expert arrives at deposition with forty photographs taken on a phone. They are sharp, well exposed, and useless. Asked what focal length produced the third one, the expert does not know. Asked whether the camera was level with the surface, the expert says it looked level. Asked whether the color in the sixth image is the color of the stain or the color the phone decided the stain should be, the expert has no answer, because the phone did not record a decision, it recorded a result. Asked to produce the original files, the expert produces images that passed through a cloud photo library and a text thread and now carry a creation date matching the day counsel asked for them.

Nothing in that deposition was about photography. All of it was about method, and the expert had none.

Decide what question the frame answers before you shoot

The habit that fixes all of this is small and unglamorous: decide, before the shutter goes, what question this frame has to answer. If the answer is "so I remember the layout," you are taking a note, and the note is allowed to be a note. If the answer is anything a jury might later be asked to accept, you are building an exhibit, and the exhibit needs a scale, a stated method, a preserved original, and a log entry.

1.2Before each frame, name the question it answers. If you cannot name one, take it anyway and write down that you could not.

1.3Photograph the thing you are certain is irrelevant, because relevance is decided by people who were not standing where you were standing.

Where this book stops and the report begins

What you do with these images after they are made, how they enter your report, and how you testify from them, belongs to a different book in this series, Being a Great Expert Witness. This one stops at the moment the image goes into the report. But the report can only be as good as the frames behind it, and no amount of careful writing rescues a photograph that was taken to jog a memory.

Chapter 1 Checklist: The Two Kinds of Image

Chapter 2The Camera Is a Measuring Instrument

A camera is a measuring instrument that happens to make pictures.

No engineer would accept a torque figure from a colleague who said he tightened the bolt until it felt about right. No chemist would accept a concentration from an instrument nobody had calibrated, run against a known standard, or documented. Yet experts hand over photographs every day with no stated settings, no verification of any kind, and no written procedure, and expect those photographs to carry measurements.

The three questions that make a camera an instrument

Three questions decide whether a court will treat an instrument as an instrument. Does it produce a known quantity? Does it produce the same quantity twice? Can a comparably trained person operate it and reach your answer? Those three questions are calibration, repeatability, and documentation, and a camera answers all three or it answers none.

Your method, not your gear, is what gets admitted

Your method, not your gear, is what gets admitted. A $4,000 camera body operated on automatic with no record of what it did is weaker evidence than a $680 body operated in manual with a written procedure and a log. The expensive camera takes a better picture. The documented camera survives cross-examination.

The oldest rule in this literature is reproducibility

The oldest published statement of this principle in the American forensic imaging literature comes from the Scientific Working Group on Imaging Technology, the body that carried this field before it was absorbed into the current structure. Its Section 11, on documenting image enhancement, rests on four operative principles. The original file must be preserved. The documentation must satisfy the legal requirements for introducing forensic images in a court of law. The documentation must be sufficient to allow other professionals to understand what was done and produce comparable results. And the practices apply regardless of the tools and devices used.

Read the third one again. Reproducibility is the test. Not accuracy, not resolution, not the badge on the front of the camera. Whether somebody else can do what you did and land where you landed.

The current body, the Scientific Working Group on Digital Evidence, says the same thing in operational language. Its image processing guidelines require that steps be recorded "in a manner sufficient to permit a comparably trained person to understand the steps taken, the techniques used, and to extract comparable information from the image." That is the same test, restated for digital work, and it is the sentence I would want read aloud at any hearing about my own imagery.

Work on duplicates and leave the originals alone

The working group's 2025 document on maintaining the integrity of imagery adds the handling rules that make reproducibility possible. Images are stored in an unaltered state. All examination and processing happens on duplicates, never on originals. The duplicate is verified against the original by hash comparison before any processing begins, and processed imagery is then treated as derivative evidence carrying its own integrity requirements. Hashing is performed before and after any copy operation and again before storage. Storage means physical security, logical security, and redundant copies in more than one location.

Notice what those rules are doing. They are turning a camera from a device that makes pictures into a device that makes records with a provenance you can trace. That is the whole conversion.

2.1Fix your exposure in manual for any frame that will be measured, compared, or color-matched, because a setting you chose is a setting you can describe and a setting the camera chose is not.

2.2Verify the camera clock against a known time reference before every job, record the reference you used, and record the offset if there is one.

What calibration means when it is not about color

Calibration deserves its own paragraph, because photographers hear the word and think of color. Calibration in the evidentiary sense means knowing what your instrument does to the thing it measures. A lens has a distortion signature. A sensor has a sensitivity floor and a saturation point. A scale bought from a forensic supplier has a manufacturing tolerance that may or may not match the specification printed on the package, a point that Chapter 5 takes up with the measured numbers. Every one of those is a property of your instrument that an opponent can look up, and every one is a property you should be able to state before the opponent does.

Compare what exists for the neighboring technology. If you bring a terrestrial laser scanner to a scene, ASTM International publishes test methods that let you say something defensible about that scanner's measured performance rather than repeating the manufacturer's brochure: ASTM E3125 for point-to-point distance measurement performance of spherical coordinate three-dimensional imaging systems in the medium range, and ASTM E2938 for relative-range measurement performance in the medium range. Run either one and you have a number that came from a published test method rather than from a sales page.

Nothing equivalent exists for a camera at a scene. There is no published test method that tells you how to characterize the imaging chain you carried into that warehouse. Chapter 4 deals with that absence directly, because it is the single most important structural fact about this field. For now, take the consequence: where the standard does not supply the verification, you supply it, and you write it down.

Could someone else repeat your shoot and match it

Repeatability is the property that opponents test and that experts most often fail. Repeatability means that if you go back tomorrow, or if the other side's expert goes next month, the same procedure produces the same image within a stated tolerance. It is bought with fixed settings, recorded distances, recorded angles, a tripod, and a scale. It is lost the moment you hand the decisions to the camera.

2.3Record, per frame, the lens and focal length, aperture, shutter, sensitivity setting, camera-to-subject distance, and the position and power of any light you brought.

That list looks tedious. It takes about fifteen seconds per setup with a notebook, and it is the difference between an instrument reading and a snapshot. Under Federal Rule of Evidence 702, as amended effective December 1, 2023, the proponent must show by a preponderance of the evidence that the expert's opinion reflects a reliable application of the principles and methods to the facts of the case. A reliable application is a described application. There is no other kind.

Chapter 2 Checklist: The Instrument

Chapter 3What the Camera Sees and What the Eye Sees

The most dangerous sentence in a visibility case is not "I did not see him." It is "here is a photograph of what he could see."

That sentence is an overreach, and it is the single most common one in this entire field. A photograph is a transform, not a recording. It is the output of an optical system with a fixed aperture, a sensor with a fixed noise floor, a set of processing choices, a display, and a room. A human retina is none of those things. The two systems do not share a response curve, a resolution, an integration time, or a color mechanism, and at night they do not even share a photoreceptor class.

How long the eye takes to adapt to the dark

Start with adaptation, because it is where the argument is usually lost. The human visual system does not have one sensitivity. It has a sensitivity that moves, and it moves slowly. The cone system carries vision above about 0.03 candelas per square meter; below that the rod system takes over and provides scotopic, meaning night, vision. After 5 to 8 minutes in the dark, that second mechanism comes into play. Full dark adaptation, the point at which the rod threshold asymptotes to its minimum of about 10 to the minus 5 candelas per square meter, takes roughly forty minutes.

Bleaching one percent of the rhodopsin in the rods raises the visual threshold by a factor of 10. Bleaching fifty percent of it raises the threshold by 10 log units. A driver who glanced at a lit dashboard, or passed under a sodium lamp, or met an oncoming headlamp, is not the same detector two seconds later that he was two seconds before, and the difference is not subtle. It is orders of magnitude.

What else changes when the light goes away

Resolution and contrast move too. Photopically, meaning in daylight, peak contrast sensitivity reaches roughly one half of one percent contrast, and the acuity cutoff runs 50 to 60 cycles per degree. Scotopically, maximum contrast sensitivity is about eight percent and maximum resolution about 6 cycles per degree. Resolving power falls by roughly a factor of 10 and the contrast threshold worsens by roughly a factor of 16. The night driver is not a daylight driver in a dark room. He is a different instrument.

Then there is where he is looking. In daylight, acuity is foveal, and it degrades within 5 arcminutes of fixation, losing about a quarter of its value by 10 arcminutes off axis. At scotopic levels the picture inverts: maximum acuity sits at roughly 5 to 15 degrees of eccentricity, and peak rod density lies at 15 to 20 degrees. At night the best part of your vision is not where you are pointing your eyes. A photograph, which is uniformly sharp across the frame, represents neither condition.

Color goes away entirely. Rod peak sensitivity is around 507 nanometers, and rod vision is achromatic. A color photograph of a night scene presents chromatic information that the observer's visual system could not have delivered. That is not a subtle exaggeration. It is the addition of a channel that did not exist.

Timing differs as well. Rods integrate light over durations up to about 100 milliseconds; photopic cones integrate over about 10 to 15 milliseconds. A tripod-mounted exposure of several seconds integrates ten to a hundred times longer than any retina can, manufacturing detail no observer could have accumulated. The field's own answer, in the Society of Automotive Engineers literature, is to constrain observer viewing time to approximately one quarter of a second in validation studies. Flicker differs too: critical flicker fusion runs around 60 hertz photopically and around 15 hertz scotopically, which is why frame rates, shutter angles, roadway lighting, and pulsed vehicle lamps interact on video in ways no eye experiences.

3.1Never offer a night photograph as a representation of what a witness could see without stating the adaptation state, the viewing duration, the display, and the calibration method you used to connect them.

Dynamic range in stops is not a sensor property

Practitioners quote "dynamic range in stops," a stop meaning a doubling or halving of light, as though it were a property of a sensor. It is not. It is a property of a sensor plus a choice about how much noise you are willing to call signal.

Bill Claff's published sensor analysis makes this concrete. For a Nikon D300 at the sensitivity setting marked ISO 100, engineering dynamic range, defined at the point where the signal to noise ratio crosses 1, computes to 12.37 stops. Photographic dynamic range for the same sensor at the same setting, defined at a signal to noise ratio of 20 and adjusted for the sensor's circle of confusion, computes to 9.00 stops. Same sensor. Same sensitivity. Same measured saturation point. A difference of 3.37 stops created entirely by the choice of noise criterion.

That is the strongest cross-examination point available in either direction. If an opponent testifies that his camera captured fourteen stops, ask which criterion produced that figure. If you use the number yourself, state the criterion in the same sentence.

Do not put a number on the eye's dynamic range

There is no authoritative published figure for the instantaneous dynamic range of the human eye. The commonly repeated span of fourteen orders of magnitude is not sourced, and the lower bound usually quoted is an order of magnitude below the rod absolute threshold the vision literature reports. Do not put that number in a report. You do not need it. The sourced facts about adaptation, threshold shift, contrast sensitivity, and eccentricity carry the argument by themselves.

3.2Do not state a figure for the dynamic range of the human eye. State the adaptation and threshold numbers instead, which are published and which prove the same point.

The display is part of the evidence chain

The reproduction chain matters as much as the capture. The central methodology paper in the nighttime visibility literature, on depicting night scenes by equivalent contrast, concedes in its own abstract that digital sensors and displays cannot reproduce the full dynamic range visible to human eyes. The foundational calibrated-print method from 1989 was, in its authors' own description, the production of a range of photographic prints from which the correct density level is selected by an observer at the site. Exposure and print density are discretionary choices, and the classical method admits it and builds an observer procedure to constrain them. The 1996 video work went further and required calibrating both recording and playback for the prevailing conditions.

Take that last point seriously, because it has a courtroom consequence most experts miss. If the validated method calibrates playback, then an uncalibrated courtroom monitor sits outside the validated method. The display is part of the evidence chain.

And when the jury looks at your night exhibit on a lit screen in a lit room, they are operating on the photopic branch of the contrast sensitivity function. The driver was operating on the scotopic branch. The juror is a physically different detector than the person whose conduct is at issue.

The founder of the modern nighttime visibility literature wrote in 1987 that while such reconstructions can be helpful, they can also be very misleading. A later paper framed the evidentiary difficulty precisely: the determination of what would have been visible to the reasonably alert person must be made with proper consideration of both inter-person and intra-person variability. Neither of those is modeled by a photograph.

What you may say about a night photograph

None of this means you cannot photograph at night. It means the photograph is evidence of the scene, not evidence of the witness. Under Federal Rule of Evidence 702, general reliability of a calibration method does not establish reliable application to this scene, this headlamp, and this observer's adaptation state. And under Federal Rule of Evidence 403, a night exhibit that overstates what was visible is the textbook posture for exclusion as misleading the jury.

3.3Say what your photograph shows and stop there. It shows the scene under stated conditions through a stated instrument. It does not show what anyone saw.

Chapter 3 Checklist: Eye Versus Sensor

Chapter 4The Standards That Do Not Exist

There is no published American national standard for forensic scene photography. Not a weak one, not an outdated one, not one buried behind a paywall. There is none.

That sentence surprises people, and it should, because the neighboring disciplines have standards and the vocabulary of standards is everywhere in this field. Learn the actual structure, because you will be asked about it under oath, and the expert who answers this question accurately is immediately more credible than the one who waves at a certificate.

The body most practitioners cite stopped publishing

The Scientific Working Group on Imaging Technology produced the foundational consensus documents of American forensic imaging, organized as numbered sections and compiled into a single volume of best practices for the forensic imaging practitioner. Its overview document, Section 1, sits at version 3.3 and is dated June 11, 2010. The working group ceased operations in the middle of the last decade, when its role was absorbed by the newer structure. Its documents are no longer maintained and are not updated. Its old web address no longer belongs to it and now serves unrelated commercial content, so do not send anyone there and do not print that address in a report.

This does not make the imaging working group worthless. It makes it legacy. Cite it for foundational principles, cite it with its date attached, and never let a citation to a 2010 document stand in for current practice.

Where current forensic imaging practice lives now

The live body is the Scientific Working Group on Digital Evidence. It is active, it publishes free versioned documents, it maintains dedicated photography, imaging, and video committees, and it has absorbed the imaging role the older group used to hold. Its photography committee alone maintains current guidance on low light crime scene photography, reflected ultraviolet and infrared imaging, technical aspects of digital cameras, photographic equipment recommendations, macro photography, evidence photography in a controlled setting, and image categories in forensic science. Its imaging committee maintains best practices for image authentication, for maintaining the integrity of imagery, for forensic image analysis, and for compression and file formats. Its video committee maintains the digital video and recorder acquisition documents.

That is where current practice lives. If you take one operational instruction from this chapter, take that one.

4.1Cite the digital evidence working group for current practice and the imaging working group only for principles, with its date attached.

The organization everyone names does not publish standards

The Organization of Scientific Area Committees for Forensic Science, administered by the National Institute of Standards and Technology, was created in 2014 to address a lack of discipline-specific forensic science standards. It runs a Forensic Science Standards Board for governance, a Standards Review Panel, Scientific Area Committees organized by discipline, subcommittees inside each, and cross-cutting resource task groups, with more than 800 volunteer members and affiliates across 19 forensic disciplines.

The organization does not publish standards. It writes drafts and pushes them to accredited standards developing organizations, principally the Academy Standards Board of the American Academy of Forensic Sciences, which is accredited by the American National Standards Institute, and ASTM International. Those bodies publish. The organization then decides whether to place the resulting published standard on its registry, which it describes as a repository of selected published and proposed standards containing minimum requirements, best practices, standard protocols, terminology, or other information. Inclusion on the registry signals that a standard is technically sound and that laboratories should consider adopting it.

So the chain runs in one direction: the organization drafts, the standards board or ASTM publishes, the registry lists. An expert who testifies that his work "complies with the organization's standard" has described something that does not exist. Some working group documents are themselves placed on that registry, which is the strongest single argument for treating the digital evidence working group as the authoritative current source for forensic imaging practice.

4.2Never testify to compliance with a standard published by a body that does not publish standards. Name the publisher, the designation, and the year.

What is published, what is only a draft

Now the negative findings, stated plainly because plain statement is the credibility move. There is no published standard from the Academy Standards Board specifically on forensic photography or forensic imaging. Its Standard 220, on scene documentation, was open for public comment with a deadline of August 31, 2026. That is a draft. Track it. Do not cite it as published, and do not let anyone else cite it as published without correction.

What does exist, and what you can use, is terminology and performance work from ASTM International. ASTM E2916 covers terminology for digital and multimedia evidence examination and sits on the registry in its American National Standards Institute approved form. ASTM E1732 covers terminology relating to forensic science. On the three-dimensional side, ASTM E2544 covers terminology for three-dimensional imaging systems, ASTM E2807 specifies the data exchange format for three-dimensional imaging, and ASTM E2938 and ASTM E3125 are test methods for evaluating measurement performance of three-dimensional imaging systems in the medium range. Those two test methods are why an expert with a laser scanner can say something defensible about instrument performance and an expert with a camera cannot.

Federal Rule of Evidence 702 is what governs

So what governs? Federal Rule of Evidence 702 governs, and it is more demanding than a certificate would be. As amended effective December 1, 2023, the rule requires the proponent to demonstrate to the court by a preponderance of the evidence that the requirements are met, and subsection (d) requires that the expert's opinion reflect a reliable application of the principles and methods to the facts of the case.

Read that as an instruction rather than as a hurdle. The rule does not ask whether a body has blessed your equipment. It asks whether your method is reliable and whether you applied it reliably here. In a field with no governing standard, that means conformance is demonstrated by documented conformance to published best practice plus reproducibility. Not by certification. Not by membership. Not by a laminated card.

The defensible posture is the same one that runs through every document I have named. Capture to examination quality. Preserve originals unaltered. Hash at acquisition. Process only working copies. Document every step well enough that a comparably trained person could repeat it and extract comparable information. That thread runs through the imaging working group's Section 11, the digital evidence working group's image processing guidelines, its compression and file format guidelines, and its integrity guidance. Four documents from two bodies across fifteen years, saying the same thing.

4.3When there is no standard, publish your own method in the report and make it the thing that is reproducible.

4.4Answer "what standard did you follow?" with the documents by name and the honest statement that no American national standard exists, then describe your method. Do not bluff and do not shrug.

Use the missing standard in both directions

The absence cuts both ways, and you should know how to use it in both directions. When your opponent claims you violated a standard, ask him to produce it. When your opponent claims that because no standard exists nothing was required, point him to the rule, which requires reliability regardless of whether anybody has written it into a designation. Same absence. Different verdict.

Chapter 4 Checklist: Standards and Authority

Chapter 5Gear That Earns Its Weight

The most expensive item in a working evidence kit is not the camera body. It is the mistake you make with a phone.

Every piece of gear here answers one question: does it make the image more defensible, or only prettier? Most photography money buys the second. Almost none of what follows does.

Why your phone cannot take the pictures you need

Begin with the phone, the most capable camera most experts own and the most dangerous one they carry. The problem is not resolution; phones out-resolve plenty of dedicated cameras. The problem is that a phone does not photograph the scene. It computes an image a trained model believes the scene should look like, whether or not you asked.

The clearest public demonstration remains the moon test. In March 2023 a tester downscaled a photograph of the moon to 170 by 170 pixels, blurred away the surface detail, displayed it across a room, and photographed it with a flagship phone. The phone returned an image with considerably more surface detail than the doctored source. Samsung's support documentation explains why: its scene optimizer recognizes the moon as an object, synthesizes more than ten images through a super resolution process, then applies a deep learning detail enhancement engine. The manufacturer stated publicly that it does not apply any image overlaying to the photo. That is narrow, true, and beside the point. A model trained on lunar imagery generated detail that was never in the light entering the lens.

5.1Assume every phone image is a composite until you can prove otherwise, and never base a measurement or a color opinion on one.

The same mechanics run on every platform. Apple's deep fusion process fuses nine images across 24 million pixels in about a second, and its trigger conditions are undocumented, so the operator cannot tell whether a file came from one exposure or nine. Google's night mode captures up to 15 frames when handheld, or 6 frames of a full second on a tripod, and applies a learned automatic white balance model. Color in such an image is a neural network's judgment, not a measurement, and for bruise coloration, paint transfer, corrosion, or burn patterns that is disqualifying.

It gets worse for exhibits with people in them. Manufacturers now blend frames so everyone in a group looks their best, and at least one implementation merges facial expressions automatically at capture. The design goal is that the result feel true to the user's memory. Wrong standard.

The reason is physics, not snobbery. Light gathering scales with the square of the pixel pitch. A 24 megapixel full frame sensor has pixels about 6.00 micrometers across; a flagship phone's main sensor has native pixels of 1.22 micrometers, and another runs 200 megapixels at 0.6 micrometers. The full frame pixel gathers roughly 24 times the light of the first and roughly 100 times the light of the second. Phones need computational photography because their pixels cannot produce an acceptable single exposure. A dedicated camera does not.

Where is a phone fine? Canvassing for cameras. Photographing a street sign so you remember the address. Anything where the image is a note and everyone knows it is a note.

Where to spend and where to stop

Every price below is a July 2026 snapshot, several promotional, and bodies are superseded every two to four years. Lenses, tripods, flashes, scales, and color targets change far more slowly. That is where your money should go.

Under $1,000 buys adequate and defensible: a 24 megapixel cropped-sensor body, the format the industry labels APS-C, at around $680, with a kit zoom, a third-party macro lens, a manual flash, and a tripod. With a written procedure behind it, that kit produces admissible imagery.

Between $1,000 and $3,000 is the working expert's kit and this book's default recommendation. The defining feature is not the sensor. It is two memory card slots set to simultaneous backup, which a $1,999 full frame body provides and a $1,449 full frame body from the same manufacturer does not. Add a 90 to 105 millimeter macro lens, a normal prime lens, meaning one that does not zoom, a wide zoom, a radio-triggered flash plus one off-camera unit, a geared tripod head, scales, and a color target. Above $3,000 buys specialization: a 45 megapixel body at around $3,899, a twin macro flash, a tilt-shift lens, and a spare body.

The memory card holds the only copy that exists

For the minutes or hours between the shutter and the offload, the memory card is the single instance of the evidence. Everything else in the bag has a spare. If the camera body dies you rent another one and come back tomorrow. If the card dies you have lost a scene that has already been swept, repaired, or released.

That asymmetry is the whole argument for not economizing here. A card is the cheapest item in the kit and the only one with no backup behind it.

Which memory card your camera takes, and what each format buys

Cameras take one of two families, and the camera body decides for you.

The Secure Digital family, universally shortened to SD on the label, is the common one. The postage-stamp version is called microSD and turns up in action cameras and small drones. It is the same electronics in a smaller body, and it is easier to drop, easier to lose in gravel, and nearly impossible to handle with gloves on. Use the full-size card wherever the camera offers a choice.

Secure Digital cards are sorted by capacity, and the tier decides the file system, which decides whether your computer will read it without a fight. Standard cards stop at 2 gigabytes. The tier marked SDHC runs above 2 up to 32 gigabytes. The tier marked SDXC, which is what you will buy, runs above 32 gigabytes to 2 terabytes. A newer tier marked SDUC extends past that.

CFexpress is the faster family, built on the same signalling a computer uses to talk to its internal drive. It comes in Type A, which uses one data lane, and Type B, which uses two and is therefore roughly twice as quick. Type B in its current generation moves about 4 gigabytes per second. Your camera accepts one or the other and they are not interchangeable, so check the manual before you buy anything.

Read speed is the number on the front, write speed is the number that decides

The large number printed on the package is the read speed, which is how fast the card can hand data to a computer. It is the number you care about least in the field.

The number that decides whether you get the frame is the sustained write speed, because that is what empties the camera's buffer while you are still shooting. The two are not close. One widely sold professional card is rated at up to 200 megabytes per second read and up to 90 megabytes per second write on the same label. The read figure is more than twice the write figure, and only the smaller one governs whether the camera stalls in the middle of a burst.

Buy on the video speed class instead, because it is the only figure on the card that is a guarantee rather than a maximum. A card marked V30 guarantees 30 megabytes per second of sustained write, V60 guarantees 60, and V90 guarantees 90. Note that V60 and V90 require the faster bus, and you can see whether a card has it: those cards carry a second row of contacts along the back edge. A fast card in a slow slot works, at the slow slot's speed.

5.2Buy on the guaranteed sustained write class, never on the read speed printed on the front.

Speed is not the same as reliability

A speed rating tells you how fast a card moves data on a good day. It tells you nothing about the four things that actually lose evidence.

The first is who made it and where you bought it. Counterfeit memory cards are a real and well-documented trade, and the usual forgery reports a large capacity to the camera while containing far less, so it writes happily and then silently discards frames. Buy from the manufacturer or an authorized dealer, keep the receipt, and never buy the cheapest listing of a premium brand.

The second is the controller inside, which is the part that spreads writes across the memory so no cell wears out early. This is where a card from a serious manufacturer and a nameless card of identical printed specification stop resembling each other, and it is not something you can inspect or test at the scene.

The third is endurance. Memory cells wear out with write cycles. A card that has been filled and reformatted every week for three years is not the card you take to the only inspection you will get.

The fourth is physical and environmental tolerance. Fire scenes, dive boats, and industrial floors are hard on hardware. Professional cards are specified for temperature, moisture, shock, and static, and the specification is published. Cheap cards are silent on all four.

5.3Buy from the manufacturer or an authorized dealer, keep the receipt, and never buy the cheapest listing of a premium brand.

Several medium cards beat one large one

The instinct is to buy the largest card that fits the budget so you never have to swap. Resist it. One 512 gigabyte card holding an entire inspection is a single point of failure wrapped around the whole case, and a card that fails takes everything on it.

Two or three cards of moderate capacity spread the risk, and swapping between subjects gives you natural break points that match how the file is organized later. Label each one and log which card holds which sequence.

Handling that keeps the card an original

Format the card in the camera that will use it, never on a computer, because the camera writes the folder structure it expects. Do that at the start of a job, never in the field, and never on a card you have not already offloaded and verified.

When the shooting is done, treat the card as evidence rather than as a supply. Switch the write-protect tab if the format has one, and do not put it back in service until the files are copied, hashed, and verified in two places. Retire cards on a schedule instead of waiting for one to fail, and write the retirement date on the case.

5.4Format in the camera at the start of a job, never in the field, and never on a card you have not offloaded and verified.

Working distance decides which macro lens you buy

Macro is where lens choice becomes an evidentiary question, and the deciding number is one no manufacturer publishes. Minimum focus distance is measured from the sensor plane and includes the whole camera and lens. Working distance is the clear space between the front of the lens and the subject. A 105 millimeter macro lens with a 290 millimeter minimum focus distance delivers only 134 millimeters of working distance. A 50 millimeter macro at the same 1:1 magnification delivers 43 millimeters.

At 43 millimeters you cannot get a light between lens and subject, and a bracket flash is blocked, which means no raking light, which means no tool mark striations and no surface topography. At 110 to 135 millimeters there is room for a twin flash or a side light at a grazing angle, and standoff from a healing injury or a component still in place. Longer is not always better: one 100 millimeter lens reaches 1.4 times magnification, but its working distance collapses to 85 millimeters. Extra magnification is bought with working distance. The optimum is 90 to 105 millimeters.

5.5Buy the macro lens by its working distance at maximum magnification, and measure that distance yourself, because the specification sheet will not give it to you.

Your forensic scale is probably out of specification

Scales carry more evidentiary weight per dollar than anything else in the bag, which makes the next finding uncomfortable. The National Institute of Standards and Technology reviewed commercially sold forensic photography scales against the design specification for the American Board of Forensic Odontology No. 2 scale, developed in 1988 by engineer William Hyzer and Dr. Thomas Krauss. That specification calls for circles of 19.75 millimeters internal and 23.00 millimeters external diameter, 80 millimeters center to center, with the legs perpendicular to within 2 arcminutes.

Across 20 rulers from 4 vendors, plus a survey of 50 rulers from 10 vendors, every vendor met the length requirement. Everything else failed. Circle external diameters ran from 19.176 to 20.396 millimeters against that 23.00 millimeter specification, so no vendor met it. Only one held the 80 millimeter spacing within 0.2 millimeters; five produced 84.9 to 85.2 millimeters. More than half fell outside the perpendicularity requirement, and factory quality control checked only the length scale.

Most readers do not know what those circles are for. They are a perpendicularity check: photographed square to the surface they are circles, and photographed off axis they are ellipses. They are the feature used to verify alignment and, in some workflows, to rectify perspective. A scale whose circles are out of specification is a bad alignment reference and a worse rectification target. Check yours.

5.6Verify your specific scale against a calibrated reference before you use it for anything you will testify to, document the verification, and retain the record.

Carry a color target that no standard requires

The last item I treat as mandatory is a color reference target. No standard requires one inside the frame, and Chapter 11 explains why I carry it anyway. A passport-sized color checker at $119 carries the 24-patch classic target plus white balance, gray balance, and enhancement targets, 52 patches on a card about 109 by 63.5 millimeters, and produces custom camera profiles. Photograph it in the same light as the subject, in the same frame where you can. It converts color from an opinion into a measurement, which is the point of everything in this chapter. Color evidence with no reference target in the file is color evidence you cannot defend.

Buy the body last, and the scale first

Buy the body last. Buy the scale, the color target, the macro lens, and the second card slot first, because those are the four things anyone will ever ask you about.

Chapter 5 Checklist: The Kit

Part Two. The Craft

Chapter 6Exposure Without Guesswork

The camera does not measure light. It measures the light that bounced back off whatever you pointed it at, and those are two different quantities.

Three controls decide how much of that returning light reaches the sensor, and each one charges you for the privilege. Learn what each one costs and you will stop guessing.

Why the aperture numbers run backwards

The first is the aperture, the adjustable hole inside the lens. It is named by the f-number, which is the focal length divided by the diameter of the entrance pupil, the apparent opening you see when you look into the front of the lens. That definition explains why the scale looks strange on the dial: f/1, f/1.4, f/2, f/2.8, f/4, f/5.6, f/8, f/11, f/16, f/22. Every step multiplies the number by the square root of 2, which shrinks the diameter by about thirty percent and the area by half. So every step halves the light. An f/2 lens admits four times the light of an f/4 lens. Smaller number, bigger hole. That inversion trips up every engineer on the first day and nobody on the second.

The shutter sells you time in doublings

The second control is the shutter speed, the length of time the sensor is exposed to the image: 1/1000 second, 1/500, 1/250, 1/125, 1/60, 1/30, 1/15, 1/8, 1/4, 1/2, 1 second. Each step doubles the time and doubles the light.

The sensitivity setting is not a measurement

The third is the sensitivity setting, marked ISO on every camera, and it is not what most people think. The governing standard, ISO 12232, first published in 1998 and revised in 2019, says outright that in a digital camera an arbitrary relationship between exposure and output values is achieved by setting the signal gain of the sensor. The number describes an exposure index chosen by the manufacturer, not a measured physical sensitivity, and two camera bodies both set to ISO 400 are not necessarily receiving the same exposure. Raising the setting collects no additional photons. It amplifies the signal already collected, which shrinks the relative contribution of downstream read noise but does nothing about the randomness in the light itself. Emil Martinec's measurements at the University of Chicago show read noise on one professional body falling from 24 electrons at ISO 100 to 4.2 electrons at ISO 1600. That fall is the entire benefit. What you cannot beat is photon shot noise, which rises as the square root of the signal, so the only real cure for noise is more light.

6.1Shoot at or near the camera's base sensitivity whenever a tripod and thirty seconds are available to you.

One stop is a factor of two, everywhere

Because each step of each control is a factor of two, the three are interchangeable in a common currency called a stop. Open the aperture one stop and you can halve the exposure time, or halve the sensitivity, and the brightness is unchanged. What changes is everything else: depth of field, motion blur, and noise. Exposure value ties it together. It equals the base-two logarithm of the f-number squared divided by the exposure time in seconds, reckoned at ISO 100 by convention.

That gives you a field instrument needing no batteries. Sunny 16 says that on a sunny day, for a subject in direct sunlight, set f/16 and a shutter speed equal to the reciprocal of the sensitivity: at ISO 100, f/16 at 1/100 second. Run the arithmetic and that combination is exposure value 14.6, which rounds to 15, and 15 is precisely the published value for a typical scene in full or slightly hazy sunlight with distinct shadows. The rule and the table are the same statement. Open to f/11 for slight overcast, f/8 for overcast, f/5.6 for heavy overcast, f/4 for open shade or sunset, and add a stop for a backlit subject.

Full sunlight is exposure value 15. Heavy overcast and open shade in clear sun are both 12. A bright street scene is 8. Offices and work areas run 7 to 8. Home interiors run 5 to 7. Night vehicle traffic is 5.

Suppose you want to hold f/5.6 and 1/125 second, a defensible handheld combination for a scene walk. In full sun you would need ISO 12, meaning you have three stops in hand at base sensitivity. At exposure value 12 you need ISO 96. On a bright street you need ISO 1,531. At night traffic levels you need ISO 12,250.

That progression is the exposure triangle made concrete: at exposure value 5 you put the camera on a tripod, you add light, or you accept the noise. There is no fourth option.

The handholding rule was written for small prints

Handholding has its own arithmetic. The reciprocal rule says use a shutter speed numerically closest to the focal length, so a 50 millimeter lens wants 1/60 second and a 200 millimeter lens wants 1/200 second, applied to the 35 millimeter equivalent focal length on a smaller sensor. That rule was derived for six by four inch prints made from 35 millimeter film. Your file will be examined at full magnification on a courtroom monitor. Double the rule. Image stabilization buys back 2 to 5.5 stops, but it corrects camera shake only. It does nothing about a subject that is moving.

Subject motion is where the numbers get uncomfortable. Blur on the sensor equals subject speed times exposure time times magnification. Hold that blur to 0.030 millimeters, the standard sharpness criterion for a full-frame sensor, and a person walking at 1.4 meters per second photographed at 20 meters with a 50 millimeter lens needs 1/117 second. A car at 30 miles per hour in the same geometry needs 1/1117 second. Tighten the criterion to a single pixel on a 24 megapixel full-frame sensor, which is the right criterion when the image will be enlarged or measured, and that same car needs 1/5583 second. That is why photographing motion at a scene is hard, and why an opponent's video frame of a moving vehicle is rarely a measurable object.

6.2Write down the shutter speed the fastest thing in the frame requires before you choose the aperture you want.

What your meter assumes about the world

Your camera's meter reads reflected light, and it is calibrated on the assumption that the world averages out to a middle gray reflecting about eighteen percent of what falls on it. The standard output sensitivity method inside ISO 12232 is defined against exactly that eighteen percent gray. Scenes at issue in litigation are not eighteen percent gray. Snow, a white ceiling, a chrome guard rail, and a stainless housing all read brighter than the assumption, and the meter underexposes, which is how you come home with gray snow. Soot, black asphalt, a dark engine bay, and a matte black firearm read darker, and the meter overexposes, which is how you lose the surface detail you drove four hours to record. The same deception runs your flash: automatic flash units shut off too soon when they see reflected light from tile, white walls, or chrome, and the frame comes out dark.

Bracketing is a method, not a hedge

The answer is not a better meter. The answer is bracketing, and bracketing is not a hedge, it is a method. Shoot the metered exposure, then one and two stops under, then one and two stops over. Five frames costs eight seconds and a few megabytes. It gives you a graded set from which the correct rendering of the evidence can be selected and defended, and it gives you the frames you will need when the item turns out to be both glossy and matte in the same field of view.

6.3Bracket every evidence close-up at plus and minus one and two stops, and keep every frame, including the ones you would never show.

Keep every frame is the part people skip. Merge a bracketed set into a single high dynamic range image and you have performed image enhancement as the Scientific Working Group on Digital Evidence defines it in its Image Processing Guidelines, document 15-M-002. The working group's foundational rule applies: the original image is preserved and processing is performed on a working copy, and the steps must be documented well enough that a comparably trained person can repeat them and extract comparable information. "Combined in Photoshop" does not satisfy that.

6.4Record aperture, shutter speed, and sensitivity for every evidence frame in your photo log, even though the camera writes them into the file, because the log is what you will be handed on the stand.

Opposing counsel will ask what your settings were and why. You want to answer that you chose f/8 to hold depth, 1/250 second because the moving component required it, and base sensitivity because the file would be enlarged, and that your log matches the file. Guessing is not a method. Arithmetic is.

Chapter 6 Checklist: Exposure

Chapter 7Focus, Depth, and the Sharpness You Will Be Asked About

Sharpness is not a property of a photograph. It is an agreement between the photograph, the size you print it, and the distance the viewer stands away.

That sentence is the whole chapter, and it is the reason competent opposing counsel asks about focus at all. Depth of field is the range of subject distances that render acceptably sharp in the final image. The word doing the work is "acceptably." The engineering criterion behind it is the circle of confusion, the diameter of the blur circle a point source is allowed to become before an ordinary viewer calls it unsharp. The industry convention sets it at the frame diagonal divided by 1500. ZEISS derives that number from the eye's resolving limit in its own published whitepaper and confirms it corresponds to the widely used 0.030 millimeter figure for the 35 millimeter format. Divide a 43 millimeter full-frame diagonal by 1500 and you get 0.029 millimeters. For the smaller sensor sizes, published practice runs about 0.018 to 0.019 millimeters for the common crop formats and 0.015 millimeters for Four Thirds. Be careful with the round number 0.020 that many field references use for crop sensors: it sits inside published practice, but it is not the result the diagonal rule gives, and an opposing expert who has read the same references will notice.

Depth of field tables assume a print you never made

The circle of confusion is a viewing assumption, not a property of your camera. Every depth of field table ever printed, and every depth of field scale ever engraved on a lens barrel, silently assumes a modest print viewed at arm's length. In the 1950s the convention for 35 millimeter was 0.05 millimeters, because people looked at contact prints. Blow your file up to a 40 inch courtroom exhibit and the effective circle of confusion shrinks by the enlargement ratio. The depth of field you relied on in the field is no longer there. "Everything in this photograph is in focus" is a claim about a print size you never specified, and it is attackable.

Focus at the hyperfocal distance, not on the subject

The governing equations are old and stable. The hyperfocal distance, the focus distance at which depth of field extends from half that distance out to infinity, is the focal length squared divided by the product of the f-number and the circle of confusion, plus the focal length. Focus a 50 millimeter lens at f/8 with a 0.030 millimeter criterion and the hyperfocal distance is 10.5 meters, so everything from 5.2 meters to infinity is acceptably sharp. On full frame, a 24 millimeter lens at f/11 has a hyperfocal distance of 1.77 meters, and a 35 millimeter lens at f/8 has one of 5.14 meters. Those two settings will cover most scene overview work: the 35 millimeter at f/8 focused at 10 meters is sharp from 3.39 meters to infinity.

7.1For scene overviews, focus at the hyperfocal distance rather than on the subject, and write the distance in the log.

Why depth of field collapses as you move in

Depth of field changes linearly with f-number and with the circle of confusion, but it changes with the square of subject distance and inversely with the square of focal length. A 35 millimeter lens at f/8 focused at 3 meters in a room gives you a sharp zone from 1.90 to 7.16 meters, about five and a quarter meters of working depth. A 105 millimeter lens at f/8 focused at 2 meters gives you 0.17 meters. And a 100 millimeter macro lens at f/16 photographing a fracture surface at life size gives you roughly 10 millimeters of depth of field. Ten millimeters. On a fractured bolt, on a chipped gear tooth, on a tool mark in a door frame, one plane is sharp and everything else is not, and no aperture you can select will change that.

One consequence of the same scaling law is worth committing to memory because it kills a common misconception: for a given size of the subject in the frame, the same f-number on any focal length gives the same depth of field. The long lens does not have less depth of field. It has less depth of field at the same distance, which is not the same claim, because you would not stand at the same distance.

7.2At magnifications approaching life size, assume one plane is sharp and plan to focus stack.

You cannot buy your way out with a smaller aperture

You cannot solve this by closing the aperture further, because the other end of the tradeoff is diffraction. Light passing an aperture spreads, and a point source images as an Airy disk whose diameter is 2.44 times the wavelength times the f-number. At the middle of the visible spectrum, around 550 nanometers, the Airy disk is 7.52 micrometers across at f/5.6, 10.74 micrometers at f/8, 21.47 micrometers at f/16, and 29.52 micrometers at f/22. Compare those against the sharpness criterion you already accepted and the conventional advice falls out on its own: on full frame the Airy disk equals a 0.030 millimeter circle of confusion at about f/22, so the practical floor is f/16. On the common crop formats, taking the round 0.020 millimeter figure, it crosses at about f/15, so the floor is f/11. On Four Thirds it crosses at about f/11, so the floor is f/8. Pushing past those apertures buys you depth and pays for it in resolution, and past a point the trade is losing on both sides.

Aberrations run the other direction: they fall as you stop down while diffraction rises, and the crossing point is the lens's sharpest aperture. For modern lenses of six or seven elements that optimum usually lands around f/5.6 to f/8. For older four-element designs, f/11.

7.3Work at f/8 unless you have a stated reason to be somewhere else, and be able to state the reason.

Write the focus distance down at the scene

None of this survives if the focus itself goes unrecorded. Write the focus distance, the aperture, and the depth of field you calculated into the log for every evidence frame, because the number you will be asked about two years from now is the one you worked out at the scene, not the one you remember. How those numbers are presented in the report belongs to a different book in this series, Being a Great Expert Witness. This one stops at the moment the image goes into the report.

A focus stack is a composite, and you must say so

For anything at high magnification, the answer is focus stacking: a series of frames at incremented focus distances, merged so that the sharp plane from each contributes to a single image with extended depth. It works, it is standard practice for fracture surfaces and tool marks, and it carries a documentation burden you must respect. A stack is a composite. Under the Scientific Working Group on Digital Evidence Image Processing Guidelines, document 15-M-002, the original image is preserved and processes are performed on a working copy, and the steps must be documented in a manner sufficient to permit a comparably trained person to understand what was done and extract comparable information. Retain every source frame. Say in the report that the image is a stack, how many frames, what step size, and what software. Presenting a composite as if it were a single photograph is the failure mode that gets images excluded and experts impeached.

How to prove the defect is not blur

Now the question in the title, because it will be asked in almost exactly these words: "Is that blur, or is that the defect?" You will not talk your way out of it. You defeat it at the scene, three ways. Put the camera on a tripod and use a remote release, so camera shake is not on the table. Focus manually on the feature and confirm at full magnification on the rear screen before you move, because autofocus locks on edges and contrast, and a fracture surface offers it neither. And bracket focus as well as exposure, stepping the focus plane through the feature so that you own a frame with the feature unambiguously sharp and a frame with it unambiguously soft. The contrast between those two frames is the answer to the question. The feature that persists at every focus setting is in the specimen. The feature that appears at one focus setting and vanishes at the next is in the optics.

7.4Photograph the surface at two focus settings on either side of the feature, so you can prove the feature is not an artifact of focus.

Chapter 7 Checklist: Focus and Depth

Chapter 8The Lens Is an Argument

A lens is a projection model with a price tag attached. Choose one and you have chosen where you must stand, and where you stand is the argument.

How much width you cover from where you stand

The angle of view of a rectilinear lens focused at infinity is twice the arctangent of the sensor dimension divided by twice the focal length. On a full-frame sensor of 36 by 24 millimeters, a 24 millimeter lens sees 73.7 degrees horizontally, a 35 millimeter sees 54.4 degrees, a 50 millimeter sees 39.6 degrees, an 85 millimeter sees 23.9 degrees, and a 200 millimeter sees 10.3 degrees. Translate that into the number you need at a scene, which is how much width you cover from where you stand. At 20 meters, a 35 millimeter lens covers 20.6 meters of width, a 50 millimeter covers 14.4 meters, and a 105 millimeter covers 6.9 meters. Decide the lens before you take a step, not after.

Sensor size sets the scale. Full frame is about 36 by 24 millimeters, diagonal 43.3 millimeters. The common crop formats measure 23.6 by 15.6 or 22.3 by 14.9 millimeters, crop factors of about 1.52 and 1.61, and Four Thirds is 17.3 by 13.0 millimeters, a crop factor of exactly 2. Crop factor is the ratio of the full-frame diagonal to yours, so a 50 millimeter lens on a 1.6 crop camera body frames like an 80 millimeter lens on full frame. It is still a 50 millimeter lens, and every calculation in this book uses the real number. Be careful with sensors specified in inches: those fractions come from 1950s video camera tubes, where the number described the outside diameter of the glass envelope, and the real sensor is roughly a third smaller than the fraction implies.

Fifty millimeters is not what the eye sees

You will be asked whether your photograph shows the scene as a person would have seen it, and the honest answer requires separating two claims that everyone else runs together. A "normal" lens is conventionally one whose focal length equals the frame diagonal, 43.3 millimeters on full frame, with anything from 40 to 58 millimeters also called normal. The famous 50 millimeter standard was chosen by Oskar Barnack when he built the Leica: a historical accident, not an optical derivation. And the justification for the diagonal rule is a viewing argument, not a vision argument. Such a lens covers about 53 degrees diagonally, the angle subtended by a print viewed at a distance equal to its own diagonal.

Set that against human vision. A person sees 200 to 220 degrees horizontally with both eyes. The central visual field is conventionally the central 60 degrees. Sharp foveal acuity covers about 1.5 to 2 degrees. So a 50 millimeter lens, at 39.6 degrees horizontal, is not showing what the eye sees: it is far narrower than the visual field and wider than the sharp part of it. What a 43 to 50 millimeter image does reproduce, printed and held at a distance equal to its diagonal, is the geometry of central, attentive looking. Say it that way and it survives cross-examination. Say "50 millimeters is what the eye sees" and it does not.

8.1For any image offered to show what a person could see, use 40 to 50 millimeters equivalent from the measured eye position, and document the eye height and position, because the camera position is the substantive claim.

Perspective comes from distance, not from the lens

Perspective is the next trap, and it is the one that changes verdicts. Linear perspective is caused by distance, not by the lens. Two photographs of the same scene from the same spot have identical perspective geometry no matter what lens is fitted. But focal length dictates where you must stand for the framing you want, and where you stand is what changes perspective. So wide-angle exaggeration and telephoto compression are real effects of the workflow even though neither is a property of the lens. Articulate that distinction and you win the exchange. Fumble it and you look like you are defending a trick.

The numbers make it undeniable. Take two identical objects separated by 5 meters in depth. Photograph them from 2 meters and the near one images three and a half times the size of the far one. From 20 meters, 1.25 to 1. From 100 meters with a long lens, 1.05 to 1: they look identical.

Neither photograph is retouched. Both are accurate. Only one will be admitted without a fight.

The same effect operates on a face: a 24 millimeter lens framing head and shoulders forces you to 0.32 meters, where a near feature images 30.9 percent larger than one 100 millimeters behind it, while a 105 millimeter lens puts you at 1.42 meters and the same error is 7.1 percent. That is an accuracy problem, not a styling problem. Avoid anything longer than about 135 millimeters equivalent for spatial relationships, because a following vehicle shot with a long lens looks dangerously close, and that exhibit is answered from the same position with a normal lens.

8.2Use 90 to 105 millimeters for injury, component, and failure-surface work, and never wider than 24 millimeters equivalent for anything that shows spatial relationships.

Barrel, pincushion, and the kind that will not correct

Barrel distortion means magnification decreases with distance from the optical axis, so straight lines bow outward; it lives at the wide end. Pincushion means magnification increases with distance from the axis, so lines bow inward; it lives at the long end. Mustache distortion is a mixture, barrel near the center turning to pincushion at the edge, and because its radial coefficients change sign, a one-coefficient correction cannot fix it.

Percent distortion is the actual radial distance of an imaged point minus the paraxially predicted distance, divided by the predicted distance, times one hundred. Watch the sign convention, because it is not universal. Edmund Optics calls barrel negative and pincushion positive, and offers the tolerance that 2 to 3 percent goes unnoticed when no measurement algorithm is running. DXOMARK uses the opposite convention and calls 1 percent absolute value the usual acceptable limit. Cite a figure without stating whose convention it follows and a competent cross-examiner will invert your sign for you.

Here are measured magnitudes from LensTip, which publishes figures from uncorrected files to two decimals. A modern prime lens, meaning one that does not zoom, near normal focal length is nearly distortion-free: a current 55 millimeter f/1.8 measures 0.12 percent barrel on full frame. A professional standard zoom swings hard across its range: the Nikkor Z 24 to 70 millimeter f/2.8 measures 3.91 percent barrel at 24 millimeters, 2.18 percent at 28, 0.09 percent pincushion at 35, and 2.38 percent pincushion at 50, a six-point swing inside one lens. An older consumer superzoom measures 5.7 percent barrel at its wide end, and an ultrawide 15 millimeter prime measures 6.12 percent. At 6 percent an edge feature is displaced by six percent of its radial distance, millimeters on the sensor and meters of apparent position error in a scene photograph.

8.3Record the exact focal length used for every frame, because a zoom lens is a different lens at every setting, and the metadata will be asked for.

Distortion correction moves the focal length without telling you

LensTip measured that same 15 millimeter lens by star-field astrometry, fitting 137 stars in the corrected frame and 126 in the uncorrected one. Uncorrected, the lens showed 6.12 percent barrel distortion, a measured horizontal angle of view of 93.53 degrees, and an equivalent rectilinear focal length of 13.3 millimeters. With the camera's distortion correction turned on, the same lens showed 0.09 percent distortion, 89.56 degrees, and an equivalent focal length of 14.3 millimeters. The manufacturer's published angle of view is 86.9 degrees, which matches neither one.

Read that again in the language of a reconstruction. Switching distortion correction on or off moved the field of view by about 4 degrees and the effective focal length by a full millimeter, silently, with no indication anywhere in the camera. Build a photogrammetric solution on the nominal 15 millimeters and you are wrong by 7 percent on a corrected file and wrong by more on an uncorrected one, and the error sits in the most important parameter of the projection model. The nominal focal length is not a measurement. It is marketing.

The photogrammetry standard is silent where the risk is

The standards do not save you here. The Scientific Working Group on Digital Evidence treats geometric correction as image restoration in its Image Processing Guidelines, document 15-M-002, and requires that the original be preserved and that steps be documented well enough for a comparably trained person to reproduce them. But the working group's own Best Practices for the Forensic Use of Photogrammetry, version 1.2, does not address camera or lens calibration, does not address distortion correction, and does not say whether corrected imagery is acceptable input. It is silent exactly where the technical risk lives. Say so plainly in your report. A gap named by you is a gap that cannot be sprung on you.

8.4For any image that will be measured, calibrate that specific body and lens at that specific focal length and focus distance, and never rely on the number printed on the barrel.

Calibration recovers two focal length terms and two principal point coordinates in pixels, plus three radial and two tangential coefficients modeling the misalignment of the elements. Shoot raw, archive the uncorrected original, calibrate, and document the correction status of the file you analyzed. If the work is metrology, leave the zoom in the bag. Two rules sit behind all of it. Federal Rule of Evidence 901 requires evidence sufficient to support a finding that the item is what you say it is, and illustration (b)(9) lets you satisfy that by describing a process or system and showing that it produces an accurate result. Federal Rule of Evidence 403 lets the court exclude relevant evidence when its probative value is substantially outweighed by the danger of misleading the jury. A photograph that exaggerates or compresses depth is not a lie. It is a candidate for that second rule.

Chapter 8 Checklist: The Lens

Chapter 9Light You Find and Light You Bring

Every lighting failure in evidence photography comes from one of three places. The light was in the wrong place, there was not enough of it, or there was far too much of it in one small spot.

When the light you find stops being enough

Light you find is the light already at the scene, and the point at which it stops being enough is defined procedurally, not numerically. The Scientific Working Group on Digital Evidence, in its Guidelines for Low Light Crime Scene Photography, never defines low light with a number. It defines available light photography by its trigger: you use it when there is some existing light and a single flash will not produce a properly exposed photograph. It requires long exposures, a stable tripod, a remote release, and the meter in matrix or evaluative mode. Ambient light is for scene context and spatial relationships, not for texture, because it casts no directional shadow.

That is the idea governing the rest of this chapter. A photograph records luminance differences. A shadow is the only mechanism that converts depth into a luminance difference. Take away the shadow and you have taken away the third dimension.

The flash on top of the camera erases shadow

Which is why the flash on top of your camera is the most destructive tool in the bag. It sits a few centimeters from the lens axis, so light travels out along the optical axis and returns along it, and every shadow falls directly behind the feature that cast it, hidden from the lens. It follows, though I have found no source that says it in these words, that a striation three tenths of a millimeter deep and an ink line three tenths of a millimeter wide photograph nearly identically under axial flash. Reason it out from the geometry rather than citing it. The working group's affirmative rule states the inverse: oblique lighting places the source at a low enough height and angle to cast a shadow into the subject.

The second failure is specular. On glossy, wet, painted, chromed, or oily surfaces a near-axial source produces a mirror image of the flash tube, and those pixels clip to maximum on every channel and hold nothing recoverable. The same reflection fools the exposure the other way: automatic flash units shut off too soon when they see tile, white walls, or chrome, and the whole frame comes out dark. The working group's countermeasures are direct lighting at 45 degrees with two to four lights on each side, diffusion, tenting, and cross-polarization, which puts polarizing filters between the lights and the subject and another over the lens, rotated until the glare extinguishes. The third failure is red-eye, and the in-camera remedy is worse than the problem, because red-eye reduction fires a pre-flash that constricts the pupil, changing the appearance you came to document.

9.1Take the flash off the camera for every evidence photograph, and use the built-in flash for nothing but orientation snapshots.

The fourth failure is arithmetic, and it defeats everybody. A point source spreads its power over a sphere, so illumination falls as the inverse square of distance. Double the distance and you have one quarter of the light, exactly two stops. The distance sequence 1, 1.4, 2, 2.8, and 4 meters costs you 0, 1, 2, 3, and 4 stops, the same sequence as the f-stop scale for the same reason. A foreground at 1.5 meters and a background at 6 meters therefore differ by four full stops. A single flash cannot defeat the inverse-square law. It can only be moved.

Guide numbers turn that physics into a setting: the guide number equals the f-number times the distance. It means nothing without a stated sensitivity, normally ISO 100, the sensitivity setting marked on every camera, and nothing without a stated zoom head position. Manufacturers quote it at the longest zoom, which inflates it: Nikon publishes one professional flash at 34.5 meters with the head at 35 millimeters and 55 meters at 200 millimeters, which is 1.35 stops of apparent extra reach created entirely by narrowing the beam. Cutting manual power one stop multiplies the guide number by 0.707; doubling the sensitivity multiplies it by 1.41.

9.2Use manual flash power for every image that will be examined, compared, or measured, and reserve automatic metering for orientation frames where the exposure is not the evidence.

Manual is not preference. It is repeatability: manual power is a number you set, while automatic metering re-meters every frame and can produce different exposures of the same evidence from the same tripod position. It is documentability: "one flash at one quarter power, 1.0 meter from the surface, 20 degrees from the surface plane, f/11, ISO 100" is a reproducible specification, and "automatic" describes no state at all. Get the flash off the camera with a cord or a radio trigger.

A softbox is not soft, closeness is soft

Softness has one governing variable, and it is not the modifier. Shadow edge softness is set by the angular size of the source as seen from the subject. A one meter softbox at one meter subtends about 45 degrees; at ten meters it subtends about 5.7 degrees and is nearly as hard as a bare flash. A softbox is not soft. Closeness is soft.

The standard names a criterion, not an angle

Oblique or raking light is what makes tool marks, footwear and tire impressions, indented writing, dust prints, abrasions, and dents photographable at all. Note carefully what the standard says, because this is where experts overreach. The working group's Lighting Techniques document specifies a criterion, not an angle: place the source at a low enough height and angle to cast a shadow into the subject, then adjust the angle to produce the appropriate contrast. It prescribes no degree figure anywhere, and deliberately so, because the correct angle is a function of the depth of the relief. An expert who testifies to a fixed universal angle invites impeachment with one question.

What you can give the court are documented working values. Steven Staggs, in the standard practitioner text on crime scene photography, specifies impression photography with an electronic flash at 25 degrees and again at 45 degrees, from each of three directions, six frames minimum. The working group's angle for direct, non-oblique lighting is 45 degrees. Shallow relief such as dust impressions and indented writing calls for a lower, near-grazing angle.

The physics tells you why. For a feature of height h lit at angle θ measured from the surface plane, shadow length is h divided by the tangent of θ. A ridge one tenth of a millimeter high casts a shadow 0.10 millimeters long at 45 degrees, 0.27 millimeters at 20 degrees, 0.57 millimeters at 10 degrees, and 1.14 millimeters at 5 degrees. Your camera cannot resolve a tenth of a millimeter of depth. It has no trouble with a dark bar a millimeter long. Oblique light is a depth-to-contrast transducer.

Shadow length is not a measurement of depth. Oblique light exaggerates relief by design, and the lower the angle, the greater the exaggeration. Record the angle for every frame, and do not let a jury read shadow length as depth.

9.3Record the flash angle from the surface plane, the flash-to-subject distance, the compass or clock direction, and the manual power fraction for every oblique frame.

Light every impression from all four cardinal points

The four cardinal points protocol is binding on your practice. The working group's Guidelines for Digital Imaging of Footwear and Tire Impressions, document 17-P-003 version 2.0, instructs that for each impression you take at least one photograph under ambient light, then photographs using oblique lighting from various angles, including the four cardinal points. Twelve, three, six, and nine o'clock. The reason is geometric and absolute: the shadow relation applies only to the component of relief perpendicular to the light. A striation running along the lighting direction casts no shadow across itself and is invisible no matter how low you drop the angle. A single lighting direction guarantees that part of the evidence goes unrecorded.

The rest of the setup is prescribed and short. Camera on a tripod with the sensor plane parallel to the impression, because parallel is what keeps the geometry measurable. A scale on the same plane, with focus on the bottom of the impression. Fill the frame. Use a lossless format at the camera's highest resolution and a fixed focal length lens of 50 millimeters or longer. Set the aperture about two stops down from the widest opening. Subdue the ambient light and verify it with one flash-disabled frame, which should come out black.

When no flash is big enough, paint the scene

For a large dark scene no flash is big enough, so you stop trying. Painting with light opens the shutter on bulb and builds the exposure out of many individual flash pops fired from inside the scene. The working group's published starting point is bulb, f/8, and ISO 400, with the camera on a stable tripod, a remote release, manual exposure, autofocus off, manual flash at full power, and an opaque cloth held over the lens between pops. The cloth is what makes the technique work, because the sensor accumulates light only during the pops, so your movement, your flashlight, and passing headlights between pops are never recorded. Robert Kramer's published roadway method uses bulb at f/11 with an ISO 100 emulsion, begins at the point furthest from the camera and walks back so you never re-walk a lit zone, fires at an angle away from the camera and low to the ground, and derates the flash to about two thirds of the rated distance.

No standard gives an exposure time. The working group specifies bulb, and bulb means held open until the walk is complete, often several minutes. Pop count follows from the guide number: multiply the working f-number by the distance, divide by the guide number, and square the result. At ISO 400 that professional flash reaches a guide number of 69 meters, so at f/8 a zone 10 meters out needs about 1.3 pops and a zone 20 meters out needs about 5.4. Four times as many for double the distance, exactly as the inverse-square law demands. Walk closer and fire from more positions rather than standing back and firing many. Fire with your back or side to the camera, keep moving so you never register as a figure, count your pops, and note the distances. A counted walk is repeatable. A remembered one is not.

9.4Bracket a painted frame by varying the pop count and angles across several exposures from the same locked camera position, and review on site, because you cannot fix a painted frame later.

Chapter 9 Checklist: Light

Chapter 10Beyond Visible Light

Visible light is a narrow band that happens to match your eyes. The silicon behind your lens was never that fussy, and the manufacturer had to install filters to make it behave.

That is the engine of this chapter. A bare digital sensor responds far outside the roughly 400 to 700 nanometer band the eye uses, and every ordinary camera ships with internal blocking glass to discard the parts nobody asked for. Remove that glass, or buy a camera built without it, and you can photograph two things that were present at your scene the whole time and invisible the whole time: reflected ultraviolet radiation and reflected near infrared radiation. Both are ordinary light. Both obey the same lens, the same tripod, and the same exposure arithmetic you already know. Neither requires you to believe anything mystical.

What ultraviolet shows after white light gives up

Start with ultraviolet, because it is the technique most likely to get you in trouble. The Scientific Working Group on Digital Evidence publishes Guidelines for the Use of Reflected Ultraviolet Radiation in Forensic Photography, document 19-P-001, and it places the ultraviolet band at roughly 10 to 400 nanometers, with two emission wavelengths that matter in the field: 254 nanometers and 365 nanometers. The applications are real and documented. Healing injuries, diminished bruising, latent impressions, surface disturbances on flooring, and text on thermal receipts all photograph under reflected ultraviolet after they have gone invisible under white light.

Shortwave ultraviolet can destroy the evidence on the item

The working group's own limitations section states that exposure to ultraviolet may denature deoxyribonucleic acid, and it puts a threshold on it: wavelengths below 280 nanometers denature deoxyribonucleic acid rapidly during exposure. Read that as an engineer, not as a lawyer. Shortwave ultraviolet photography of an item can destroy the biological evidence sitting on that item. Not degrade. Destroy.

10.1Decide before you open the case whether the item gets swabbed first or imaged first, write that decision in the log with a time, and tell every person on site which order you are in.

The sequencing decision is not always yours to make alone, and it is rarely reversible in either direction. If biological evidence is collected first, the swab may lift the deposit you were going to photograph. If you image first at 254 nanometers, you may hand the laboratory a substrate with nothing left to type. The safe engineering answer, when the choice is yours, is to work at 365 nanometers only and leave the shortwave lamp in the bag. The safe procedural answer is to make somebody else own the call in writing.

A filter that leaks is not making an ultraviolet photograph

The second ultraviolet trap is the filter, and here the working group has done the field an enormous favor by publishing measured numbers. An ultraviolet transmitting filter is supposed to pass the ultraviolet band and block everything else, but several of the classic filters leak badly in the near infrared. The Kodak 18A passes 300 to 400 nanometers and leaks 14 percent at 746 nanometers. The B&W 403 leaks 17 percent at the same wavelength. The Schott UG-1 leaks 9 percent. Against those, the Baader U-Venus passes the same 300 to 400 nanometer band and leaks 0.15 percent at 737 nanometers, and the Kolari Vision ultraviolet filter, passing 320 to 400 nanometers, leaks 0.025 percent.

Translate that into testimony. A filter leaking 17 percent in the near infrared is not producing an ultraviolet photograph. It is producing a blend of two different physical phenomena, weighted by a ratio nobody in the room can state, and any opinion about what the ultraviolet revealed is contaminated by whatever the infrared revealed. That is the kind of methodological error a competent opponent finds in an afternoon.

10.2Name the filter make, model, pass band, and published leakage figure in the report, and buy the filter with the smallest leak you can afford.

Ultraviolet focuses at a different point than visible light, and each lens has a different focus shift, so autofocus achieved in visible light is not focus in ultraviolet. Most ordinary glass absorbs ultraviolet below about 350 nanometers, which is why dedicated work uses lenses with quartz or fluorite elements. And the exposure limits are not decorative: the International Commission on Non-Ionizing Radiation Protection sets the effective radiant exposure limit for unprotected eye and skin at 30 joules per square meter over any 8 hour period, with the hazard weighting peaking at 270 nanometers, and an unweighted limit of 10,000 joules per square meter for the 315 to 400 nanometer band. Wear ultraviolet blocking eyewear. Ordinary sunglasses are not that.

Near infrared separates materials the eye reads as identical

Infrared is the gentler sibling. The working group's companion document, 19-P-002, identifies near infrared from 700 to 900 nanometers as the practical forensic range and notes that a silicon sensor may reach only about 1,100 nanometers regardless of what you put in front of it. Infrared also focuses at a different point than visible light, which is why the guidance directs you to focus using live view through the installed filter rather than trusting the visible-light focus. Infrared penetrates the upper layers of skin, which makes it useful for tattoo and bruise documentation, and it discriminates materials that look identical to the eye, which makes it useful for blood on dark fabric, gunshot residue on clothing, ink comparison, and questioned documents. The guidance is blunt about the output: infrared images carry no color data and should be converted to black and white after capture. Do not hand a jury a false-color infrared frame and let them read hue as meaning.

A thermal image is not a temperature measurement

Thermal imaging is a different technology, and it does not do what most people in a conference room assume. Reflected ultraviolet and reflected infrared photography record light that bounced off the subject. A thermal camera records long-wave radiation the subject emits, typically across 7.5 to 13 micrometers, and it does not care whether the lights are on. That difference has consequences that surprise people. Glass is opaque at those wavelengths, so you cannot thermally image through a window, while some thin plastics are transparent. Resolution is a fraction of what your camera delivers: a representative handheld unit such as the Teledyne FLIR E8 Pro carries a 320 by 240 detector, which is 76,800 pixels for the entire scene. And a thermal image is not a temperature measurement unless you have done more work than pressing the button. That same unit publishes an accuracy of plus or minus 2 degrees Celsius or 2 percent of reading, before any error from emissivity, and emissivity is the whole problem: the camera measures radiance and converts to temperature only by assuming a surface emissivity and a reflected background temperature. Point one at bare shiny metal and it will lie to you confidently.

10.3Never call a thermal image an infrared photograph, and never state a temperature from one without stating the assumed emissivity in the same sentence.

Thermal work also carries a constitutional footnote worth knowing even in civil practice. In Kyllo v. United States, 533 U.S. 27 (2001), the Supreme Court held that where the government uses a device that is not in general public use to explore details of a private home that would previously have been unknowable without physical intrusion, the surveillance is a search and is presumptively unreasonable without a warrant. The technology has become far more common since 2001, but the instinct to point a thermal camera at somebody's house from the street should still trigger a phone call to counsel.

How fluorescence works, and what you must record

Alternate light sources sit between these techniques. A narrowband source excites fluorescence, the subject re-emits at a longer wavelength, and a barrier filter on the camera and on your goggles blocks the excitation so you can see the faint re-emission. Yellow pairs with blue excitation, orange with blue-green, red with green. A commercial unit such as the Foster and Freeman Crime-lite 82S PRO offers bands from 350 to 380 nanometers through 800 to 900 nanometers, and the working group imposes exactly one documentation duty in its low-light guidance: record the wavelength used and the filter combinations used.

Be honest about what fluorescence proves. A randomized controlled trial by Scafide and colleagues, published in the Journal of Forensic Sciences in 2020 across 157 subjects balanced over six skin color categories, found 415 nanometer excitation with a yellow filter improved bruise detection over white light with an odds ratio of 5.34, and 450 nanometers with yellow gave 4.08. The authors framed their own result carefully: alternate light enhances concurrent physical assessment in the presence of a known history of injury. The counterweight is Lombardi and colleagues in 2015, who measured specificity of roughly 51.6 percent on day one against 71.0 percent for white light and concluded that fluorescence under an alternate light source is not sufficient to accurately or responsibly diagnose subclinical bruising. Roughly half of uninjured control forearms fluoresced in a way readers called a bruise.

10.4Treat fluorescence as presumptive, never confirmatory, and shoot a white-light control frame of the identical field before you shoot the fluorescence frame.

Chapter 10 Checklist: Beyond Visible Light

Chapter 11Color You Can Defend

Color is the easiest thing in a photograph to correct and the hardest thing to defend. Those two facts are related, and the relationship is what this chapter is about.

What your camera does to color before you see it

Every photosite under the color filter array records one of three colors, so roughly two-thirds of the color in any digital photograph is interpolated rather than measured. On top of that interpolation the camera applies a white balance, which is nothing more than three gain numbers applied to the red, green, and blue channels to make the scene look neutral. Three numbers cannot invert a spectrum. A light source has hundreds of degrees of freedom in its spectral power distribution; if the illuminant emits nothing at 490 nanometers, no gain recovers what the subject would have reflected there, because that information was never collected. White balance neutralizes a cast. It does not restore missing reflectance. A maroon bruise, a paint chip, or a fabric can render differently under a light-emitting diode work light than under daylight, and no post-processing recovers the truth.

Color fidelity is therefore a capture problem, not an editing problem, and the fix is a physical object in the frame.

A gray card does two jobs, not one

A gray card serves two independent functions that practitioners conflate constantly. For white balance, all that matters is spectral neutrality: flat reflectance, so red, green, and blue read equal under any illuminant. The 18 percent figure is irrelevant to white balance. For exposure, all that matters is the 18 percent value, and neutrality is irrelevant. A card can be excellent for one purpose and useless for the other, and a cheap card that merely looks neutral will corrupt a custom white balance. Note also that 18 percent reflectance encodes to about 118 of 255 in sRGB, the standard red-green-blue color space that browsers and courtroom projectors assume, not to the 128 most people expect. Eighteen percent is the middle of perceptual lightness, not the middle of the number line.

What an opposing expert will ask about your chart

For serious work, use a color reference chart. The ColorChecker Classic, described in a 1976 paper by McCamy, Marcus, and Davidson, carries 24 patches in a four by six grid with six neutrals across the bottom row, and it lets an opposing expert re-derive your color correction from your own image rather than take your word for it. Two facts about that chart will be used against you if you do not know them. First, the colorants were reformulated in November 2014, which moved patch values by an average color difference of 0.85 and as much as 1.95, and the manufacturer publishes separate reference data for pre-2014 and post-2014 charts. Your report must state which set you applied. Second, no colorant is absolutely resistant to fading, and the manufacturer's maintenance guidance says to store the chart in its black cover and replace it every two years.

11.1Photograph the chart in the same light, at the same distance, and in the same plane as the subject, with exposure within a quarter stop of correct, a stop meaning a doubling or halving of light, and the light striking it at 20 to 45 degrees.

11.2Shoot every color-critical subject twice, once with the reference in frame and once without, and keep both.

That two-frame habit has real precedent. The American Board of Forensic Odontology's reference manual requires photographs both without and with a properly placed and labeled reference scale, on the same plane as the injury. The logic transfers directly: the frame with the reference makes your correction auditable, and the frame without answers the objection that the reference obscured the scene. Be candid, though, about where the authority runs out. No forensic standard requires a color reference inside the evidentiary frame. Several list one as optional equipment; one describes it purely as a white balance tool. The practice I am recommending is ahead of the published guidance, not derived from it, and you defend it by analogy to the odontology board's scale rule and to the reproducibility principle that governs image processing generally.

The PRESSURE 2 photographic validation sub-study protocol, published in Trials in 2017 for a multi-center randomized trial whose endpoint depended on judging tissue color, specified that a gray scale would be included in the photograph to correct white balance and that the computer monitor would be calibrated before review. A serious trial treated both as mandatory. So should you.

Color you captured correctly can still be shown wrong

Which brings in the display, because color you captured correctly can still be shown wrong. The International Organization for Standardization's proofing display standard, ISO 12646:2015, requires room illumination below 32 lux anywhere between monitor and observer, at roughly 5000 Kelvin, with a color rendering index above 90. The Federal Agencies Digital Guidelines Initiative sets its review stations to a gamma of 2.2 in a neutral matte gray room, with calibration verified weekly. None of that describes a courtroom. Say so in your report before opposing counsel says it for you.

A light that is not a blackbody radiator has no true color temperature, only a correlated one, and as Ohno's paper for the National Institute of Standards and Technology puts it, chromaticity is two-dimensional while correlated color temperature reports only one dimension. The missing dimension is Duv, the perpendicular distance from the light's chromaticity point to the Planckian locus. Two lights sharing a stated color temperature can sit on opposite sides of that curve and look plainly different.

11.3Capture raw so the white balance stays a reversible metadata parameter, choose the working space in processing, and deliver sRGB with the profile embedded for court display.

Dating a bruise from color has no scientific basis

Dating a bruise from its color in a photograph has no scientific basis. That is not my characterization. It is the conclusion of the systematic review by Maguire, Mann, Sibert, and Kemp in Archives of Disease in Childhood in 2005, which screened 167 studies, found only three that met inclusion criteria, and concluded that a bruise cannot accurately be aged from clinical assessment in vivo or on a photograph, that the practice of estimating the age of a bruise from its color has no scientific basis, and that it should be avoided. The 2013 update did not rehabilitate it.

The underlying data is worse than the summary suggests, and it fails at two separate layers.

The first layer is the photograph itself. Munang, Leonard, and Mok had three observers describe the color of 58 bruises on 44 children, both in person and from photographs. Complete agreement between two observers on the color description ran twenty-seven percent in person and twenty-four percent from photographs. Clinicians could not agree on what color the bruise was, and agreement got worse on photographs than in the room. Every dating opinion built on a photograph is stacked on top of that.

The second layer is the inference from color to age. Bariciak and colleagues, in Pediatrics in 2003, showed history-blinded examiners 50 children with accidental bruises of known age. Accuracy within 24 hours was 47.6 percent for emergency pediatricians, 29.4 percent for other physicians, and 36.8 percent for trainees, with individual accuracy ranging from zero to 100 percent. Interobserver reliability came out at a kappa of negative 0.03. Zero is chance. Negative means the observers agreed less often than random guessing predicts. The examiners relied primarily on color, and color showed no significant correlation with accuracy. Pilling and colleagues in 2010 ran the same test on 15 forensic experts using 132 images of 25 bruises of exactly known age from zero to 209 hours, and the median difference between estimated and actual age was 26.0 hours.

There is exactly one narrow inference the literature supports, and it runs in one direction only. Langlois and Gresham, studying 369 bruises across 89 subjects in 1991, concluded that it was only possible to conclude that a bruise with a yellow color was more than 18 hours old. That is the whole finding, and it does not run backwards. It does not make a red bruise fresh, a purple bruise two days old, or a non-yellow bruise less than 18 hours old, because any color can be present in fresh, intermediate, and old bruises. The finding was drawn from subjects under 65. And Thavarajah, Vanezis, and Perrett showed in 2012, using tristimulus colorimetry on 18 dark-skinned individuals, that yellowing cannot be reliably detected on darker skin because of pigmentation. The single surviving color-dating rule therefore has a built-in disparity in who it can be applied to at all.

11.4If you must say anything about color and time, say only that yellow suggests the bruise is not brand new, name the one-directional limit out loud, and stop.

Attack the measurement and the inference separately

So how do you dismantle an opponent who dates a bruise from a photograph anyway? Attack the layers separately, because each one falls on its own.

Ask what the illuminant was and whether its spectrum was recorded. Ask whether a neutral reference was in frame. Ask whether the file was raw or compressed, because chroma subsampling discards three-quarters of the color measurements and the yellow-green gradient at issue is exactly the low-amplitude color information that goes first. Ask what monitor the opinion was formed on and whether it was calibrated.

Then move to the second layer and ask for the study that supports color-to-age mapping, and when the answer is Langlois and Gresham, read them the sentence those authors wrote. Then hand over Bariciak's kappa. One line of questions defeats the measurement; the other defeats the inference. You do not need both, and you should use both.

Chapter 11 Checklist: Color You Can Defend

Chapter 12Scale, Measurement, and the Parallel Plane

A scale is not a ruler in a picture. It is a claim that the picture can be measured, and every claim invites a question.

The question is always the same one: what converts pixels into millimeters, and how do you know it is right? Three things answer it. A reference object of known dimension, placed in the plane of the evidence. A sensor plane held parallel to that plane. And a documented method for both. Miss any one and you have a photograph that looks measurable and is not.

The L-shaped forensic scale does three separate jobs

The standard reference object in forensic work is the American Board of Forensic Odontology No. 2 photomacrographic scale, described by Hyzer and Krauss in the Journal of Forensic Sciences in 1988. It is a rigid L-shaped plate, 105 millimeters on each leg and 1 millimeter thick, with graduations along both perpendicular legs, three circles, an 18 percent gray patch, and alternating black and white bars 1 centimeter wide. Those elements do three different jobs and should be described separately. The graduated legs give linear measurement on two orthogonal axes. The gray patch and the alternating bars are a tonal reference, letting you recover a measurement from a badly exposed frame because at least one bar edge survives clipping. And the three circles exist for geometry: they are the feature used to detect and correct the distortion produced when the camera is not square to the subject.

The board's published specification carries four numbers. Length accuracy of plus or minus 0.1 millimeter or 1 percent for the major centimeter graduations. Circle diameters of 19.75 millimeters internal and 23.00 millimeters external. Circle center spacing within 0.25 percent of a nominal 80 millimeters. And legs mutually perpendicular within 2 minutes of arc.

What happened when someone measured the scales themselves

The National Institute of Standards and Technology published a Dimensional Review of Scales for Forensic Photography in which it measured commercially sold scales against a system with roughly 0.001 millimeter uncertainty, nearly two orders of magnitude tighter than the tolerances being checked.

The linear graduations passed. Every vendor stayed well within the allowed error across the range tested, most within 0.1 millimeter at the 5 centimeter graduation.

The circles did not. None of the vendors achieved the specified 23.00 millimeter external diameter, with actual measurements running from approximately 19.176 to 20.396 millimeters, a shortfall of 2.6 to 3.8 millimeters on a 23 millimeter specification. Only one vendor came close on internal diameter. Only one met the spacing tolerance. More than half the rulers tested failed the perpendicularity specification. The institute's site visits found that manufacturers typically checked only the length graduations, often by visually comparing plastic scales against certified steel rulers, and applied their own tolerances rather than the published specification.

Understand what that means before you shrug at it. The circles are the feature relied on for perspective rectification. A circle in the subject plane images as an ellipse when the sensor is tilted, and the ratio of the ellipse's minor axis to its major axis is what tells you the tilt angle. That relation is geometry you can derive at the kitchen table, not a published forensic standard, so present it as derivation. Three circles at known separation, rather than one, are what make a full projective correction of the whole frame tractable instead of a single-axis fudge. If the circles are not the diameter they are supposed to be, and not the distance apart they are supposed to be, then the correction derived from them carries an error nobody has quantified, and the rectified image is offered to a jury as though it were geometry.

12.1Verify your own scale against a traceable standard, record the measured values, and be prepared to say on the stand whether the circles on that particular scale were checked.

Hold the sensor plane parallel to the subject

The better answer is to avoid needing the correction. The parallel plane rule is stated in every relevant guidance document. The Organization of Scientific Area Committees for Forensic Science, in its guide for crime scene photography, directs you to place the camera lens perpendicular to the subject. The legacy Scientific Working Group on Imaging Technology told you to mount the camera on a tripod with the focal plane parallel to the impression. The tread evidence working group said film plane parallel to impression. Different bodies, different decades, one instruction.

The arithmetic behind it is simple enough to put in a report. A subject plane tilted by an angle from the sensor plane is foreshortened along the tilt direction by the cosine of that angle. At 5 degrees you under-read by 0.38 percent. At 10 degrees, about 1.5 percent. At 20 degrees, 6.0 percent. At 30 degrees, 13.4 percent. At 45 degrees, 29.3 percent. Show the cosine when you present those numbers, because they are derived geometry rather than a published table, and derived geometry survives cross-examination when you show the derivation.

Two things make the real problem worse than the table. First, the error is directional. Only dimensions along the tilt axis shrink, so a tilted photograph does not merely scale wrong, it changes shape. A circular wound photographed 20 degrees off axis images as an ellipse flattened by six percent, and every aspect-ratio-dependent opinion is corrupted even if you never state an absolute dimension. Second, under true perspective the error is not uniform across the frame: near-side features are magnified and far-side features are minified, so a single scale placed at one point cannot correct the rest. That is precisely why the odontology board's scale carries three circles distributed across the field instead of one.

12.2Use a tripod and the camera's dual-axis level for every measurement image, and photograph the level indication if the geometry will be contested.

The scale must sit at the depth you measure

Placement is the other half of the geometry. The guidance is emphatic and consistent: the entire scale shall be on the same plane as the subject. For macro work, the scale goes at the edge of the frame and fills no more of it than necessary. For a three-dimensional impression, the scale sits at the depth of the feature you intend to measure, and the tread evidence guidance goes further and tells you to focus on the bottom of the impression rather than on the scale.

The reason is worth stating plainly to any jury, because it is intuitive once said. A scale calibrates the image only at its own distance from the lens. In a perspective projection, magnification varies as one over distance. A scale sitting 10 percent closer to the lens than the evidence is imaged 10 percent larger than the evidence, and every measurement you derive inherits that ratio error directly, with no warning in the picture.

When the scale belongs in the frame at all

The crime scene photography guide says photographs should be captured with a scale any time relative size is in question, and that if a scale is used, the photographer shall be certain a photograph is first captured without the scale. Note the mandatory verb and note the ordering. The scale is an object you introduced into the scene. The unscaled frame proves it is not concealing anything and preserves the scene in its found condition.

12.3Shoot the no-scale frame first, then the scale frame, from the same position without moving the camera.

There is a widespread belief that a 6 inch or 15 centimeter rule is the standard forensic scale length. I could not find any standard that designates one. Vendors sell many lengths. Choose a scale roughly the size of the item, say why you chose it, and do not claim a standard that does not exist.

Why your close-up work comes out soft

Close-up work adds its own trap. A lens is called a macro lens only if it reaches 1:1 magnification, meaning the image on the sensor is as large as the object itself. At high magnification the marked aperture stops being the working aperture. Effective f-stop equals marked f-stop multiplied by one plus the magnification, so at 1:1 the effective aperture is two stops smaller than the number on the barrel, a stop being a doubling or halving of light. A marked f/16 at 1:1 is f/32. That single fact explains why so many forensic macro images look soft: the photographer stopped down to get the tool mark or the wound into focus, and diffracted the detail away doing it. Depth of field at these magnifications is often measured in millimeters, and the honest resolution is focus stacking, many frames at a sharp effective aperture with focus stepped through the subject and merged.

12.4If you focus stack, preserve every source frame unaltered, document the steps to a reproducibility standard, and disclose in the report that the exhibit is a composite.

That last one is not optional politeness. A focus-stacked image is a computational composite, and the governing image processing guidance requires that the original be preserved, that processing be performed on a working copy, and that the steps be documented well enough for a comparably trained person to reproduce them. "Combined in Photoshop" fails that test. Presenting a composite as though it were a single photograph is the failure that gets exhibits excluded and experts remembered for the wrong reason.

Chapter 12 Checklist: Scale and Geometry

Chapter 13Raw, Compressed, and the Artifact That Looks Like Evidence

The most dangerous line in an evidence photograph is not the one you failed to record. It is the one the compressor drew for you.

What a raw file keeps and a JPEG discards

Bryce Bayer's color filter array, patented for Eastman Kodak in 1976 and long since expired, puts a single colored filter over each photosite: green over half, red and blue over a quarter each. Every photosite therefore measures one color and infers the other two from its neighbors, a reconstruction called demosaicing. Roughly two-thirds of the color in any digital photograph is interpolated rather than measured. A raw file preserves the measured mosaic and stores the camera's interpretation choices as metadata. A JPEG, the compressed format most cameras produce by default, contains only the camera's one-time guess, and the guess cannot be re-made.

That distinction is the strongest evidence-integrity argument available to you. In a raw file the white balance lives in a metadata tag, so a color correction changes an interpretation parameter, leaves the recorded measurement untouched, and can be reversed exactly. In a JPEG the white balance is baked into the pixel values and is gone. A raw file supports the testimony that you changed nothing. A JPEG cannot.

Bit depth follows the same pattern. Baseline JPEG is always 8 bits per channel, 256 levels. A 12-bit raw carries 4,096 levels and a 14-bit raw carries 16,384. That headroom is why raw survives the exposure error you made in a dark stairwell.

13.1Shoot raw for anything that will be examined, compared, or measured. Shoot JPEG only for orientation frames where the pixel values are not the evidence.

Two definitions of dynamic range, three stops apart

Dynamic range is where cross-examination gets sloppy on both sides. There are two published definitions and roughly three stops between them, a stop being a doubling or halving of light. Engineering dynamic range takes the low endpoint where the signal-to-noise ratio crosses 1; photographic dynamic range takes it at a signal-to-noise ratio of 20, adjusted for the sensor's circle of confusion. On one much-measured sensor the two figures are 12.37 stops and 9.00 stops. Same sensor, same settings, 3.37 stops of difference created purely by the choice of noise criterion. An opposing expert can say fifteen stops while you say eleven and you can both be telling the truth. A dynamic range figure without a stated noise criterion means nothing.

Compressed raw is not lossless, whatever the menu says

"Lossless compressed raw" is lossless in fact. "Compressed raw" is not, and neither are several format names a non-specialist reads as synonyms. Nikon's ordinary compressed format applies a lossy encoding before its entropy coding, reducing the distinct values from the analog to digital converter to a table of a few hundred to a few thousand entries; on a 12-bit sensor a 683-entry table retains about 16.7 percent of the discrete tonal levels. Sony's compressed format converts to an 11-bit tone curve and delta-encodes in 16-pixel blocks with 7-bit differences, which posterizes exactly where you least want it. Canon's compressed option quantizes a wavelet transform and lands 30 to 40 percent smaller than lossless. And Digital Negative, the archival raw container, is not automatically lossless either; its compression tag includes a lossy setting. "I converted it to Digital Negative" establishes nothing about data preservation.

What the compressor throws away, and in what order

Compression is worth understanding mechanically, because the artifacts are named and predictable. JPEG converts to a luminance-plus-color representation, subsamples the color, partitions the image into 8 by 8 blocks, applies a discrete cosine transform to each block, divides the resulting 64 coefficients by a quantization table, and entropy codes the result. Read the standard's sample luminance table and the policy is visible: the entry scaling overall brightness is 16, while the entries carrying the finest detail run past 100. The quantization table is a written record of what the camera decided to throw away. Real cameras use their own tables, so never speak of "the" JPEG quantization table as though only one exists.

The color side is more aggressive still. In the 4:2:0 subsampling used by most cameras and virtually every phone, one color value is shared by a two-by-two block of four pixels, so three-quarters of the color measurements are discarded before the transform even runs. Luminance is untouched. That asymmetry is why compression damage shows up as color error at edges, and why an opinion resting on subtle hue gradients rests on the first thing thrown away.

The Scientific Working Group on Digital Evidence publishes an artifact list in its compression guidelines: blocking, meaning pixel blocks coinciding with the macroblock size; banding, meaning visible steps in what should be a smooth gradient; local color distortion; and high frequency losses, in which edges go fuzzy and fine detail blurs. Notice what the list does not include. It does not name ringing or mosquito noise, and it contains no warning that an artifact may be mistaken for image content. The field's own governing guidance is thinner on this hazard than the hazard deserves, and saying so is a legitimate observation for an expert to make.

An artifact that looks exactly like a crack

Ringing is the Gibbs phenomenon in a picture. A sharp edge requires high-frequency transform coefficients; when quantization drives them to zero, the truncated cosine series overshoots and undershoots at the discontinuity, leaving faint light and dark bands parallel to every high-contrast edge. At high magnification a dark ringing band beside a bright edge is visually indistinguishable from a hairline crack, a fracture initiation line, a scratch, a fatigue striation, or a tool-mark line. Blocking has its own signature failure: a hard 8-pixel step across a painted surface, a weld, or a fracture face reads as a seam or a machining mark, and the tell is regularity, because real physical features do not align to a perfect 8-pixel lattice across a frame. Banding manufactures a contour line that reads as a shallow dent boundary. And the quietest failure is the opposite of a false positive: fine texture such as ridge detail, striations, or thread pitch simply disappears, and its absence is not visually obvious to anyone.

13.2When you see a suspicious line in a compressed image, check whether it lies on the 8-pixel grid before you write a word about it.

I want to be exact about the evidentiary status of that entire discussion, because the honesty is the point. I searched for a published case report or peer-reviewed forensic paper documenting a specific instance in which a compression artifact was misread as physical evidence in casework, and I did not find one. The forensic literature on compression is overwhelmingly about detecting compression to authenticate an image, not about compression creating false features. The closest published analogs sit in medical imaging. So the argument in the preceding paragraph is reasoned engineering inference, built on documented artifact mechanisms and standards guidance, not on documented casework. Say exactly that in your report if you rely on it. An expert who labels his own inference as inference is far harder to impeach than one who lets a jury assume there is a case behind it.

What the literature does supply is a quantified boundary, from fingerprints. The National Institute of Standards and Technology measured one-to-one fingerprint matching against compression ratio and reported true accept rates of 0.993 at 15:1, 0.988 at 20:1, and 0.982 at 30:1, concluding that compression up to 20:1 produces minimal negative effects and that beyond 20:1 the decline exceeds the specified limit. The Federal Bureau of Investigation did not use JPEG for fingerprint archiving at all; it commissioned a purpose-built wavelet codec and capped compression around 15 to 20 to 1. When a national laboratory needs that number, it runs the experiment.

The guidance is consistent with all of this. Lossy compression may be acceptable for documentation images such as scene and investigative photographs, and it is not recommended for images that will be subjected to forensic analysis or comparison. The working group's image categories document draws the same line by name: examination quality images should be captured in a lossless or uncompressed format at the highest resolution the device offers.

13.3Sort every planned frame into documentation or examination quality before you leave for the scene, and set the format accordingly.

The newer formats hide their damage better

Two newer developments close the chapter. The High Efficiency Image File Format, standardized internationally and now in a third edition published in 2025, is roughly two and a half times as efficient as JPEG at equal objective quality, since JPEG needs on average 139 percent more data for the same measured quality, which sounds like good news and is worse news for evidence. It uses variable-size transform units and directional intra prediction, so it does not produce the tell-tale 8 by 8 grid, and its dominant failure modes are smoothing and detail substitution: a block extrapolated from its neighbors can be filled with a plausible pattern rather than the pattern that was there. JPEG damage announces itself. This damage looks clean. I have found no published forensic characterization of these artifacts, so treat that as reasoned analysis from the coding structure, not settled literature.

The second development is generated detail. Adobe's denoising tool is a deep convolutional neural network that performs demosaicing and denoising in a single step and writes out a Digital Negative file, which everyone downstream reads as raw. Consumer tools advertise the recovery of skin pores and hair wisps, which cannot be recovered because they were never sampled. They are synthesized from a statistical prior. The moon controversy proved the point only because the moon is a known, fixed, publicly documented object. On a crash scene, a wound, a weld, or a fracture surface, there is no ground truth to catch it with.

13.4Record in the chain of custody whether any neural processing touched the file, and treat a machine-generated Digital Negative as a processed image, not an original.

Chapter 13 Checklist: Format and Artifacts

Part Three. The Scene

Chapter 14Before You Leave

The photograph that loses a case is rarely the one taken badly. It is the one nobody thought to take, on the only day anyone was going to let you on the property.

Start from the question the imagery must answer

Every documentation plan starts at the end. Name the question your imagery has to answer, write it in one sentence at the top of the page, and build the shot list backward from that sentence. "Was the guard in place on the day of the incident" is a question. "Photograph the machine" is not. The first sentence generates a list: the mounting bosses, the fastener holes, the paint witness at the boss faces, the operator's line of sight to the pinch point, the warning placard, the serial plate, and the same features on the identical machine two bays over. The second sentence generates forty pictures of a machine and a deposition you will not enjoy.

14.1Write the question the imagery must answer at the top of the plan before you write a single line of the shot list.

A plan is one page with four things on it: the question, the shot list built backward from the question, the equipment that list requires, and the constraints you already know about. Constraints are what turn a good plan into a possible one. Sunrise and sunset, if the case turns on what a driver could see. Tide and current, if the structure sits in water. The plant's shutdown window, because the line will not stop for you during production. The three hours the property owner agreed to, and not a minute more.

Who decides what you may photograph on site

Access is a legal question and it is not yours to answer. Someone negotiated your entry, and the terms of that negotiation govern your day. Know them before you leave the office.

A written site inspection protocol should tell you the date and the time window, the areas and systems in scope, who may attend, who may operate equipment, whether you may measure, whether you may test, whether you may remove anything, and what you may not photograph. Faces of employees, security systems, other tenants' property, and proprietary processes are commonly carved out. If nobody has written it down, ask retaining counsel to write it down.

Where entry comes by court order, the order controls, and the order's silence is not permission. Federal Rule of Civil Procedure 34(a)(2) governs entry onto land for inspection in federal civil cases, and Federal Rule of Civil Procedure 26(c)(1)(E) lets a court designate the persons who may be present while the discovery is conducted. Reading those provisions is counsel's work. Complying with them is yours.

There will be no second visit

Plan for the visit you get, not the visit you want, because there is almost never a second one. Scenes are repaired. Machines are rebuilt or scrapped. Vehicles are released to salvage. Guardrails are replaced, pavement is resurfaced, the tree is cut down, the building is sold. Even when the site survives, the second inspection costs another motion, another travel day, another round of objections, and it arrives after the other side has read your report and knows exactly what you missed. Shoot the site as though the site is about to disappear, because in litigation terms it usually is.

Carry two of everything that can fail

Redundancy is not gear enthusiasm, it is arithmetic. A camera body is a single point of failure standing between you and a scene that will not exist next month. Carry two bodies, and where you can, carry two bodies that take the same lenses and the same batteries. Carry more memory cards than you can imagine filling, and format them in the camera that will write them, not on a computer. Carry spare batteries for the bodies, the flashes, the triggers, and the light meter, and charge every one of them the night before. Do not delete a frame in the field. Do not reformat a card in the field. A card that is full is a card that goes in the labeled sleeve and stays there.

14.2Format every card in the body that will write it, fill it, then retire it to a labeled sleeve for the rest of the trip.

Your camera clock is an uncalibrated instrument

Now the clock, because the clock is the first thing a competent opponent attacks. A camera's date and time come from a free-running quartz oscillator with a small backup cell. There is no network time service, no satellite discipline, and no correction of any kind on a typical body. It is set once by a human being and then it drifts. To calibrate expectations: a temperature-compensated precision real-time clock component, the kind sold on the strength of its accuracy, is specified to roughly 2 parts per million, which is about a minute a year. An ordinary camera clock is far looser than that, and it also carries whole-hour errors from travel and daylight saving changes. The clock is an uncalibrated instrument until you calibrate it, and you calibrate it in the field, not in the deposition.

There is a second problem underneath the first. Image metadata timestamps are plain text in the camera's local time with no time zone attached. The metadata standard did not add fields for the offset from Coordinated Universal Time until version 2.31, published in July 2016, and plenty of bodies built after that date still write nothing into them. A timestamp with no offset is ambiguous by as much as fourteen hours. If your timeline turns on whether a frame was captured before or after an event, the timestamp alone will not resolve it.

The fix takes ninety seconds and it belongs at the start of every session. Photograph a time display you can defend, a satellite receiver's readout or a network-synchronized phone or the United States government time page, in the same frame as, or immediately adjacent to, your first evidence frame. Do it again at the end of the session. Then compute the signed offset: camera clock minus true time. Two measurements, one at each end, give you drift across the session and let you state a bound on the error rather than a point estimate.

Location tagging records where you stood, not the subject

Decide your location tagging policy before you leave, and write the decision into the plan. Location tagging embeds the coordinates of the photographer, not the subject, and it can also record the direction the lens was pointing. That cuts both ways. It corroborates where you stood and when, which is useful. It also travels with every frame you produce, including frames shot on the same body in an unrelated matter on the same day. Whatever you decide, decide it in advance and never strip coordinates from an original afterward.

14.3Set the policy on location tagging before the inspection, record it in writing, and never remove metadata from an original file.

Two last items belong on the page before you close it. First, weather and light: check the forecast, and if the case involves visibility, check the sun's position at the hour that matters, because you cannot rephotograph a shadow. Second, personal protective equipment, site orientation, and the escort you will be assigned. An expert who arrives without boots and a hard hat gets sent to the parking lot, and the three hours run anyway.

14.4Pack the plan itself, on paper, and shoot from it.

Chapter 14 Checklist: Before You Leave

Chapter 15Overall, Midrange, Close-Up

The close-up is the photograph everyone takes. It is also the photograph that proves the least.

A tight frame of a fractured weld proves that somewhere in the world there is a fractured weld. It does not prove the weld is the one in the complaint, that it was on the third-floor mezzanine, that it faced the walkway, or that it looked that way before anyone touched it. That is what the other two tiers are for. The close-up proves the detail. The midrange proves the location. The overall proves the context. Remove any one of the three and the remaining images float free of the case.

This is not a stylistic preference. It is the oldest working method in forensic photography, and it survives in the current guidance. The now-defunct Scientific Working Group on Imaging Technology built its impression photography guidance around documenting "the overall, midrange and close-up views of the area," and the sequence appears in the proposed Standard Guide for Crime Scene Photography published by the Organization of Scientific Area Committees for Forensic Science in June 2020. The reason the tiers exist is evidentiary, not aesthetic: each tier authenticates the tier below it.

Every close-up must trace back to an overall

Build the chain and never break it. Every close-up must appear inside at least one midrange frame. Every midrange must appear inside at least one overall. If you can put your finger on the detail in the close-up, then find that same detail in the midrange, then find that midrange region in the overall, the chain holds and you can lay foundation from the stand in ninety seconds. If any link is missing, you will spend ten minutes explaining, and the jury will spend those ten minutes deciding you are not sure.

15.1Make every close-up traceable to an overall through at least one midrange frame.

The four-corner walk, and the walls behind you

The overall tier is where most experts get lazy. Start with the four-corner walk. Stand in each corner of the space and shoot toward the diagonally opposite corner, at standing eye height, with a lens that does not exaggerate depth. Four frames and the room is covered from every direction, with overlapping content in each. Then, from the same four positions, turn and shoot outward at the near walls, because the wall behind you contains the panel, the placard, the exit, or the light switch that becomes an issue in month fourteen. Add a frame of the ceiling and a frame of the floor. Outdoors, the four-corner walk becomes a perimeter walk, and you add the approach: photograph the scene the way the plaintiff came to it, from the distance at which the plaintiff first could have seen it, standing at that person's eye height and not yours.

Overlap on purpose. Adjacent frames should share roughly a third of their content, so that a viewer can carry a feature from one image into the next without a jump. Overlap does two things: it proves nothing was skipped, and it makes the coverage reviewable by someone who was not there. One caution if those frames may ever feed a three-dimensional reconstruction. A camera that pivots on a tripod head about its own optical center produces no parallax, and without parallax there is no depth to recover. If the images are for a model, you must walk the camera between frames, not swing it.

Photograph the sign that names the room

Orientation frames are the cheapest documentation you will ever make. Before entering any area, photograph the thing that names it: the room number, the door placard, the street sign, the bay number, the serial plate, the equipment tag, the mile marker. Those frames act as dividers in a file of two thousand images, and they convert a folder into a narrative that a paralegal, a mediator, or a juror can follow without you standing next to them.

The frame without the scale comes first

Now the scale rule, and note the ordering, because the ordering is mandatory language. The Organization of Scientific Area Committees guide provides that photographs should be captured with a scale whenever the relative size of an item is in question, and that if a scale is used, the photographer shall be certain a photograph is first captured without the scale. Without-scale comes first. The reason is that the scale is an object you introduced into the scene. The unscaled frame proves you concealed nothing beneath it and preserves the item in its found condition. The same sequencing appears in the older impression guidance: overall, midrange and close-up views without scales and markers, then the same views with markers and scales added.

15.2Shoot the no-scale frame and the scale frame of every item, in that order, every time.

When the scale goes in, the geometry rules come with it. The entire scale must lie in the same plane as the subject, because a scale calibrates the image only at its own distance from the lens, and magnification in a perspective projection varies with one over distance. A scale sitting ten percent closer to the camera than the feature reads about ten percent large, and that error passes straight into every measurement you take from the frame. Put the camera lens perpendicular to the subject plane, use a tripod, and fill the frame with the evidence rather than the background.

What your photograph log has to record

The log itself takes discipline and pays it back at deposition. For each frame or short run of frames, record the frame number, the time, the subject, the camera position or direction, the lens and focal length, and one line of note. Write it as you go, before you move the tripod, not from memory in the hotel that night. Three columns and a pen beat a perfect memory, and the log is what lets you answer "which way were you facing" without guessing.

Photograph what does not matter yet

Photograph the thing you are confident is irrelevant. The adjacent machine that was not involved. The undamaged guardrail fifty feet upstream. The empty shelf. The weather. The condition of the floor twenty feet away. The absence of a warning sign, which can only be documented by photographing the place where a sign was not.

Cases change theory. The defect you were retained to examine gets conceded and the case becomes a notice case, and notice lives in the frames you took of things that did not matter yet.

I have never once regretted an extra hundred frames. I have regretted the missing one many times.

15.3Photograph the thing you are confident is irrelevant, and log it like everything else.

15.4Log the frame number, direction, and subject before you move the camera, not at the end of the day.

The three tiers are also how you will present. The overall orients the jury, the midrange places the detail, the close-up delivers it. Build the exhibit in the order you built the coverage, and the testimony writes itself.

Chapter 15 Checklist: Overall, Midrange, Close-Up

Chapter 16Photographing Injuries and People

A photograph of an injury is evidence. A photograph of a person is a relationship. You have to manage both at once, and the second one is harder.

Consent is a condition, not a form. It has to hold at the start of the session and still hold at the end. Explain what you are photographing, what will be visible, who will see the images, and that the person may stop at any time, decline any frame, and ask for a chaperone. Then do all of that. Photograph only what the injury requires and drape the rest. Keep the frame tight enough that the image proves the injury without exhibiting the human being, and shoot the version you would be comfortable projecting in a courtroom with the plaintiff's family in the gallery, because that is where it goes. Where the imaging happens at a defense medical examination, the scope is set by agreement or by order, and you photograph inside it and nowhere outside it.

Why a flash on the camera hides the injury

Now the light, and the light is where most injury photography fails. A flash mounted on the camera sits a few centimeters from the lens axis, so its light leaves along the optical axis and returns along it, and every shadow falls directly behind the thing that cast it, hidden from the lens. Skin has texture, and texture is what an abrasion, a laceration, a scar, a graft margin, or a patterned injury is made of. Axial flash flattens it. Worse, skin is oily and moist, and a near-axial source produces a mirror reflection of the flash tube that clips to pure white. Clipped highlights contain nothing recoverable. Any attempt to recover them afterward is a processing step you have to disclose.

Get the light off the axis. Current guidance from the Scientific Working Group on Digital Evidence describes direct lighting at roughly 45 degrees to the surface, with two to four lights on each side of the camera, chosen to avoid reflection from the surface, and endorses cross-polarization for glare: a polarizing filter between each light and the subject and another on the lens, rotated until the specular reflection extinguishes. Cross-polarization is the single most useful trick in injury work, because it removes the sheen from skin and leaves the color and texture underneath.

Meter for the skin that carries the evidence

Skin tone changes the exposure problem, and pretending otherwise produces bad images of some people and good images of others. A reflected-light meter reads a scene toward middle gray, so it will overexpose dark skin and underexpose light skin if you let it decide. Meter for the skin that carries the evidence, expose to protect the highlights, and check the histogram rather than the rear screen. On darker skin the useful information sits in a narrow tonal band, which means noise and compression damage do more harm there than they do on lighter skin, so shoot raw files, keep the sensitivity setting low, and add light rather than gain. Bounce only off neutral white or gray surfaces. A beige ceiling, a green corridor wall, or a red awning casts its color across the whole subject, and in injury work color is not decoration.

16.1Put a color reference target in at least one frame of every injury series, and record the light source you used.

Bodies are curved and scales are flat

Scale on the body follows the same geometry as scale anywhere else, with one complication: bodies are curved. The entire scale must sit in the plane of the feature you are measuring, adjacent to it, not covering it, and the sensor plane must be parallel to that local plane. On a forearm or a shin, the plane changes over a few centimeters, so photograph the feature in sections rather than fighting the curve in one frame. Adhesive-backed and flexible versions of the American Board of Forensic Odontology No. 2 photomacrographic scale exist for exactly this reason. Take the no-scale frame first, then place the scale. And know the limits of the tool you are using: a National Institute of Standards and Technology dimensional review of commercially sold scales found the linear graduations reliable across vendors but the printed circles, the feature relied on to correct oblique perspective, out of specification at every vendor tested. Use the legs for measurement. Do not build an opinion on the circles without verifying the particular scale.

16.2Take the no-scale frame first, then place the scale in the plane of the injury, never across it.

Do not date a bruise from its color

Do not date a bruise from its color, and be ready to take apart the expert who does. The systematic review literature on bruising concludes that a bruise cannot accurately be aged from clinical assessment in life or from a photograph, and that estimating the age of a bruise from its color has no scientific basis. It is worse than that on darker skin: Thavarajah, Vanezis and Perrett reported in 2012 in Medicine, Science and the Law that yellowing, the one color transition with any support behind it, cannot be detected on dark skin because of pigmentation. Stack that on top of the fact that the color in a photograph is a product of the illuminant's spectrum and the camera's white balance rather than of the tissue, and the inference collapses twice over. That argument is developed in full earlier in this book. Photograph the injury faithfully and let a treating clinician address timing.

Fix the variables now or the series proves nothing

Series over time is where photography earns its keep in a personal injury case, and repeatability is the whole point. Fix the variables at the first session and hold them for every session after: same camera body, same lens, same focal length, same camera-to-subject distance, same lighting setup and power, same background, same anatomical framing, same scale placement. Write those settings down and carry the sheet to the next session. A series shot that way shows change. A series shot casually shows the photographer's changing habits, and the other side will say so.

16.3Fix focal length, distance, lighting, and background at the first session and reproduce them exactly at every session after.

Record the day, do not direct it

Day-in-the-life documentation is closer to filmmaking than to evidence photography, and it draws its own objections about staging, cumulativeness, and prejudice. How far to push it is trial counsel's judgment, and the presentation questions come later in this book. Your job on the day is to record what happens without directing it, log what you did and did not stage, and resist the temptation to build the most affecting version of the truth.

The federal medical privacy law does not bind you

Now the medical privacy law, because almost everyone gets it backward. The Health Insurance Portability and Accountability Act, the federal medical privacy law, does not bind you. Its rules apply to covered entities, meaning health plans, health care clearinghouses, and health care providers who transmit health information electronically in connection with covered transactions, and, where the regulations so provide, to their business associates. That applicability provision sits at title 45 of the Code of Federal Regulations, section 164.104. The Department of Health and Human Services states the point plainly: the rules apply to covered entities and business associates, and an entity that does not meet either definition does not have to comply with them. A retained expert photographing a plaintiff at an examination or a site visit is ordinarily neither.

Correct the companion error while you are at it. The de-identification provision at section 164.514 lists eighteen identifiers that must be stripped, and item (Q) on that list is "full face photographic images and any comparable images." That list is not a definition of protected health information. It is a list of identifiers whose presence prevents information that is already protected health information from qualifying as de-identified. Saying "photographs are protected health information because they are on the eighteen-identifier list" inverts the regulation. The same photograph has two legal characters depending on who holds it: in the hospital's chart it is protected health information held by a covered entity, and in your file it is a photograph.

16.4Get the protective order's imaging and retention terms from retaining counsel in writing before the first session, and file that email with the case.

Chapter 16 Checklist: Injuries and People

Chapter 17Photographing Parts, Machines, and Failure Surfaces

Every failed component carries two records: the one the failure wrote, and the one the handling wrote afterward. Your job is to capture the first before anyone adds to the second.

Do not clean the part. Not with solvent, not with a brush, not with compressed air, not with a shop rag, not with your thumb. The debris on a fracture face may be the corrosion product that dates the crack. The film on a bearing may be the lubricant that was or was not there. The smear on a bracket may be paint transfer from the component that struck it. The dust pattern on a guard may show that the guard has not moved in a year. Cleaning is destructive testing performed without notice, without a protocol, and without the other side in the room, and it is one of the easiest spoliation arguments in this field to make.

17.1Do not clean the part, and do not mate the fracture halves.

Photograph the box before you open it

Document as-found before anything moves, and "anything" includes the box. Photograph the shipping container closed, with its labels and seals. Photograph the evidence tags, the chain of custody form, the packing material, and the way the part sits inside it. Then photograph the part in the packaging before lifting it out. If the part arrives assembled, photograph the assembly from six sides before a single fastener turns. Index marks, witness marks, alignment scribes, and the relative positions of mating components are all information that exists only until the first thing moves.

Relief is the evidence and only shadow records it

The features you are usually there to record fall into a few families, and each has its own lighting demand. Fracture surfaces carry beach marks, ratchet marks, shear lips, and origin regions, and every one of those is relief, not color. Wear patterns are polish, scoring, and directional texture. Witness marks are the impressions one part leaves in another, and their value lies in where they are and which way they run. Thread engagement is countable: the bright band where threads bore load and the dull band where they did not, and the number of threads standing proud of the nut. Fastener marks tell you about assembly and disassembly history, including socket and wrench marks on the flats, paint or lacquer witness across a joint, and torque stripe alignment.

Relief is recorded by shadow, and shadow is created only by light arriving at a low angle. For a feature of height h illuminated at angle θ measured from the surface plane, the shadow it casts has length h divided by the tangent of θ. At 45 degrees a 0.1 mm ridge casts a 0.1 mm shadow, which the camera may not resolve. At 20 degrees it casts 0.27 mm. At 5 degrees it casts 1.14 mm, an eleven-fold geometric magnification of relief into something a sensor records easily. Oblique light is a device that converts depth into contrast.

Two consequences follow and both get asked about. First, shadow length is not a measurement of depth unless you know and recorded the angle, so record the angle. Second, the lower the angle, the greater the exaggeration, which is why you shoot the same feature at more than one angle and say so.

17.2Record flash distance, angle from the surface plane, clock position, and manual power for every oblique frame.

Resist the urge to name a standard angle, because there is not one. The Scientific Working Group on Digital Evidence specifies a criterion rather than a number: place the light at a low enough height and angle to cast a shadow into the subject, and adjust the angle to produce the appropriate contrast. The correct angle depends on the depth of the relief, and an expert who testifies to a fixed universal number invites impeachment. Documented working values do exist in the practitioner literature for impression photography, at 25 degrees and 45 degrees, and near-grazing angles are used for the shallowest relief. Use them as starting points, not as a rule.

What the working group does prescribe is coverage. Its footwear and tire imaging guidance directs that for each impression the examiner take at least one photograph under ambient light, then photographs using oblique lighting from various angles, including the four cardinal points. That is the twelve, three, six and nine o'clock protocol under its current name, and the reason it is not optional is geometric: the shadow formula applies only to the component of relief perpendicular to the light. A striation running along the light direction casts almost no shadow across itself and is invisible at any angle. Lighting from one direction guarantees that some part of the evidence goes unrecorded.

Use manual flash power, not automatic metering

Use manual flash, not automatic through-the-lens metering, for anything that will be examined for detail or compared. Manual power is a number you can write down and reproduce: flash at one quarter power, 1.0 meter from the surface, 20 degrees from the surface plane, at f/11 and the base sensitivity setting. Automatic metering re-measures every frame, is deceived by exactly the shiny surfaces this work involves, and fights your bracket. It also describes no state that anyone can reproduce.

Why stopping down ruins a close-up frame

Close-up work has a trap in it that ruins more forensic images than any other single cause. As magnification rises, the working aperture is no longer the marked aperture. Effective f-number equals the marked f-number multiplied by one plus the magnification. At life size, that is a factor of two, so a marked f/8 behaves as f/16 and a marked f/16 behaves as f/32. Depth of field at these magnifications is already down to millimeters, so the instinct is to stop down, and stopping down drives you straight into diffraction two stops earlier than intuition says, a stop being a doubling or halving of light. The result is the soft, mushy macro frame that shows a fracture origin as a gray blur.

The answer is focus stacking: many frames at an aperture that is still sharp, with focus stepped through the subject, merged afterward. Do it on a rail or with the camera's own focus bracketing, on a tripod, with the part immobilized.

17.3Preserve every source frame of a focus stack and disclose the stack as a stack.

A tilted sensor plane changes shape, not just size

The measurement rules from earlier in this book apply with full force to parts. Put the sensor plane parallel to the surface being measured, because a plane tilted by θ foreshortens dimensions along the tilt by the cosine of θ: about 1.5 percent at 10 degrees, 6.0 percent at 20 degrees, and 13.4 percent at 30 degrees. Worse, the error is directional, so a tilted frame does not merely scale wrong, it changes shape, and any opinion that depends on an aspect ratio is corrupted even if you never state a dimension. Put the scale in the plane of the feature. Fill the frame. Shoot raw files or an uncompressed format at the highest resolution the camera body offers, because compression artifacts on a fracture surface can look exactly like the thing you are looking for.

Disassembly is destructive testing, so give notice

Then disassembly, which is where careers get complicated. Taking a machine apart is destructive: it destroys the as-found relationship between components, and some of what it destroys cannot be restored. Treat it as destructive testing, because that is what it is. The other side is entitled to notice, to a written protocol agreed in advance, and to attend with its own experts and its own cameras. Do not improvise this, and do not let a helpful plant maintenance supervisor start turning bolts because he is trying to be useful.

Photograph each step twice: once with the tool in position and the fastener untouched, and once immediately after the component comes free, showing both mating faces. Number the steps and number the parts, and photograph every removed item with its own label and its own bag before it goes into the bag. Run video of the whole disassembly on a locked-off camera as a continuous record while you shoot stills, because the video answers "what order did you do that in" without argument. Bag and tag as you go. Never combine two removed items in one container.

17.4Photograph every disassembly step before the tool touches the fastener and again the moment the part comes free.

Chapter 17 Checklist: Parts, Machines, and Failure Surfaces

Chapter 18Difficult Scenes: Fire, Water, Height, and Dark

The difficult scene is not the one that is hard to photograph. It is the one that is destroying the evidence and endangering you at the same time, on a schedule nobody consulted you about.

Four families of scene do this: fire and explosion, marine and underwater, elevated and rigged, and confined or dark. They share a structure. The evidence is perishable, the access is conditional, the second visit does not exist, and the thing that makes the scene interesting is the same thing that makes it dangerous. Everything in this chapter follows from that.

Fire writes in char, soot, and calcined surface

Start with fire, because fire scenes punish carelessness in the most expensive way. A burned structure is a record written in three materials: char, soot, and calcined surfaces. Char is directional and depth-varying. Soot is a deposition pattern, which means it records where hot gases traveled and where they did not. Calcination, the heat-driven change in gypsum and masonry, records duration as well as temperature. All three are surface phenomena, all three are fragile, and all three are erased by overhaul, by suppression water, by a boot, and by a gloved hand steadying a camera against a wall.

So the sequence is fixed. Photograph the exterior in a four-corner walk before anyone re-enters, then each elevation, then the ground and debris field outside the structure. Move inward compartment by compartment, working from the areas of least damage toward the areas of greatest damage, and photograph each compartment overall, midrange, and close-up before you photograph any detail within it. Photograph every protected surface, meaning every place where an object shielded the material beneath it, because a protected surface is a shadow of the fire's own geometry and it is the first thing lost when the object is moved. Then, and only then, begin layer removal, photographing each layer before you take it off. Debris excavation is destructive testing. The other side gets to see every step.

18.1Photograph the protected surface before you lift the object that created it, and photograph it again after.

Soot is the hardest thing you will ever meter

Exposure at a fire scene fights you in a specific way. Soot is one of the lowest-reflectance surfaces you will ever meter. A reflected-light meter is calibrated to render whatever it sees as a middle tone, so pointed at a charred wall it demands more exposure and gives you gray mush where the evidence is a difference between deep black and slightly less deep black. Meter manually, bracket deliberately, and put a gray card and a color reference target in at least one frame of every compartment. Color fidelity is not decoration here. The distinction between clean burn and soot deposit, between oxidized and unoxidized metal, between thermal and mechanical damage, often lives in hue and in a narrow band of dark tones.

Char relief responds to oblique light the same way a tool mark does. The current standards language does not give a fixed angle; it gives a criterion, which is to place the source low enough to cast a shadow into the feature. Set the flash off the camera, work the angle until the relief reads, and record the angle you used, because shadow length is not a measurement of depth unless the angle is known.

Painting a dark interior with the lens covered

Large burned interiors have no light in them. This is where painting with light earns its place, and the technique is documented rather than improvised. The Scientific Working Group on Digital Evidence, in its low light photography guidance, gives a starting point of bulb, f/8, and a sensitivity setting of 400, the value marked ISO on every camera. The camera goes on a stable tripod with a remote release, exposure mode manual, autofocus off. You keep an opaque cloth over the lens between flash discharges, which is what makes the whole method work: the sensor accumulates light only while you are firing, so your own movement between positions never records. Fire manual flash at full power, start at the point furthest from the camera and walk back toward it so you never re-walk a lit zone, aim the head away from the lens, and count your pops and distances so the exposure is describable in testimony rather than lucky.

Water bends the light before it reaches the lens

Water scenes change the optics, not just the logistics. Light entering a flat housing port is refracted, so everything appears larger and nearer than it is, by roughly a third. A scale placed anywhere but the plane of the subject is therefore worse than useless. Backscatter, the illuminated suspended particles that turn a strobe into a snowstorm, is defeated by moving the light source well off the lens axis, which is the same fix as on land for the same reason. Color goes with depth, red first, so a color reference target in frame is mandatory rather than optional. And a photogrammetric solution built from underwater imagery without in-water calibration is not a solution, because the projection model the software assumes is not the projection model the water produced.

Recovered hardware has a clock running on it

Recovered marine hardware has a clock on it. Salt water keeps working after recovery, so document the as-found condition wet, in place if you can reach it, and document again immediately on recovery, before anyone rinses, brushes, or preserves anything. Photograph the fastener before the wrench touches it.

Where was the load path

Elevated and rigged scenes fail on a single question: where was the load path. Anchors, connectors, terminations, and attachment points form a chain, and a chain is only as informative as its least documented link. Photograph the entire load path, anchor to attachment, in order, with an identifiable frame linking each element to the next. Photograph from the elevation of the event, not from the ground. A ground-level photograph of an anchor 40 feet up establishes that an anchor exists and nothing else; it cannot show gate position, wear at the bearing surface, or the direction of loading. Get up there on a lift, on rope, or not at all.

18.2Document rigging as found, and never adjust a component to improve a photograph.

Two rules protect you at height. Tether the camera and every accessory, because a dropped camera body is a projectile and it ends both the inspection and the relationship. And tie in independently of anything you are documenting, because using the evidence as your anchor is exactly the kind of fact that shows up in a deposition transcript.

A confined space is an atmosphere problem first

Confined and dark spaces add an atmosphere to the list of things that can hurt you. A permit-required confined space, meaning one with a hazardous atmosphere, engulfment risk, or a converging configuration, requires atmospheric monitoring, an attendant, and a retrieval plan before a camera is relevant. In a flammable atmosphere your flash and your focus-assist illuminator are ignition sources and the equipment must be rated for the environment or it does not go in. Nobody has ever been impeached for declining to enter. Careers do not end in caution; they end in a photograph nobody could take safely.

Optically, small spaces tempt you toward the widest lens you own, and a wide lens misrepresents distance inside a tank, a duct, or a crawlspace by pushing everything away from the viewer. Use the longest focal length the space permits, take more frames rather than wider ones, and if you must go wide, say so in the caption and put a scale in the plane you care about.

18.3Write the hazard, the protective equipment, and the access method into the photograph log for every difficult scene, on the same line as the frame numbers.

The unifying principle is that protecting yourself and protecting the evidence are the same discipline practiced in two directions. Both require you to slow down at the exact moment the scene is pushing you to hurry. Both require a sequence decided before you arrive. And both are judged later by people who were not there, reading a log written by someone who was.

Chapter 18 Checklist: Difficult Scenes

Chapter 19Night, Visibility, and the Honest Photograph

The night photograph lies in both directions. It shows the jury far more than the witness could see, and at the same moment far less. An exhibit that corrects only one of those errors has not been corrected.

The camera pools light no eye can gather

A rod, the receptor that carries vision at low light levels, integrates arriving photons over an interval of up to about 100 milliseconds and then starts over. A camera on a tripod integrates for as long as you hold the shutter open. Leave it open for four seconds and you have accumulated roughly forty times the light any observer's retina could pool, and the frame now contains detail no human being at that location ever had access to. This is not a subtle effect. It is the single most common way a night exhibit becomes an argument for a conclusion the physics does not support. The literature's own answer is to constrain the observer rather than expand the camera: the timed-exposure methodology reported in the Society of Automotive Engineers technical literature, paper 2017-01-1366, builds on Blackwell's work by using automatic shutters that limit observation to approximately one quarter of a second, and its results correlated positively with naturalistic driver data.

Your display cannot carry the range the scene had

The reproduction chain cannot carry the scene. The central methodology paper in this field, Society of Automotive Engineers paper 2016-01-1463 by Suway and Welcher, calibrates photographs and video for display by quantitative measurement rather than subjective assessment, and its abstract concedes in its own words that digital sensors and displays cannot reproduce the full dynamic range visible to human eyes. The people who built the method say so. You should say so too, out loud, before opposing counsel says it for you.

Adaptation is slow to build and fast to lose

Between those two errors sits adaptation, which is where most night cases are won and lost. Dark adaptation is not instantaneous and it is not linear. The cone mechanism carries vision above roughly 0.03 candelas per square meter; below that the rod mechanism takes over. After 5 to 8 minutes in darkness a second mechanism of vision comes into play, the rod branch of the adaptation curve, and the threshold keeps falling for a long time after that, reaching its minimum of about 0.00001 candelas per square meter only after approximately 40 minutes. Running the same relationship the other way gives the number that ends most arguments: bleaching fifty percent of the rhodopsin in the rods raises the visual threshold by ten log units, a factor of ten billion. A driver who glanced at a lit dashboard, passed a lit sign, or met an oncoming high beam was not the detector your tripod was.

Acuity and contrast collapse with it. Photopic contrast sensitivity peaks near half a percent; scotopic maximum contrast sensitivity is about eight percent. Photopic resolution runs to 50 to 60 cycles per degree; scotopic resolution tops out near 6. And the geometry inverts: maximum scotopic acuity sits at about 5 to 15 degrees off the line of sight, so at night the best part of the visual field is not the part the witness was looking with. Rod vision peaks near 507 nanometers and is achromatic, which means a color photograph of a night scene hands the jury chromatic information the driver's visual system could not have delivered.

19.1State in the exhibit caption that the image was made by a camera, not by an eye, and that the two are different detectors.

Put those facts together and the deepest problem in the field appears. A juror looking at a lit monitor in a lit courtroom is operating on the photopic branch of the contrast sensitivity function. The driver was operating on the scotopic branch. The juror is not a poorly informed observer of the driver's experience. The juror is a physically different instrument.

Match the conditions that change the number

Matching conditions is therefore the whole job, and it decomposes into a short list of controllable variables, each with published support. Moon phase is the one people assume matters most and it matters least. The moonlight study reported as Society of Automotive Engineers paper 2019-01-1005 by Neale and colleagues ran no-moon against full-moon conditions from moonrise to peak azimuth at thirty-minute intervals, with luminance stations at 75 feet front-left and 150 feet forward, low-beam headlamps, and calibrated photographs taken from the driver's perspective inside the vehicle. Its conclusion was that the full moon's contribution to visibility distance is negligible. Cloud cover, ambient roadway lighting, and pavement wetness move the number far more, and all three need to be recorded from the historical record rather than from memory.

Headlamps are where the real variance lives. Beam pattern, aim, lamp type, lens clouding, and bulb age change the illumination on the target by more than anything else under your control. The vehicle-specific headlamp mapping work in Society of Automotive Engineers paper 2021-01-0880 and the high dynamic range source-mapping method in paper 2020-01-0566 exist because a nominal beam pattern is not the beam pattern of the subject vehicle. Photograph the actual lamps, measure the actual aim, and if the subject vehicle is unavailable, say what you substituted and why.

Retroreflective material ages, and the aging is measurable rather than rhetorical. There is a published body of work on quantifying retroreflective materials from digital imagery, on the performance of conspicuity tape as a function of observation and entrance angle, and on the systematic degradation of retroreflective materials for testing purposes. A defense exhibit shot against fresh tape is not an exhibit of the trailer as it existed. Measure the actual material or match its condition, and disclose which you did.

19.2Record moon phase, cloud cover, precipitation, roadway lighting status, and pavement condition for the incident night and for the night you shoot, side by side, in the report.

The night exposure is chosen, not discovered

The calibration methods are old and they are honest about their own limits. The foundational paper, Society of Automotive Engineers 890730 by Holohan, Billing, and Murray, produced what it called a photograph that accurately illustrates the limits of perception, by having an observer view signs of varying shades of grey at the site under appropriate conditions and then selecting the correct density from a range of prints. Read that method again. The exposure is chosen. It is chosen against an observer standard, which is the honest part, but it is chosen, and absent that observer step the density of your night print is arbitrary. The video analogue, paper 960895, adds the point most experts miss: it calibrates recording and playback, using a hand-held grey scale chart on location. The courtroom monitor is inside the validated method. An uncalibrated display in the courtroom is outside it.

A parallel line of work treats the camera as a photometer, with published calibration methods for luminance estimation from still cameras and from video. That work is what allows an expert to say a measured quantity rather than an impression. It also imposes an obligation: if you use the camera as an instrument, you must state its linear response range and the settings at which you calibrated it, the same way you would for any other instrument.

Two warnings from inside the field are worth more than any warning from outside it. The abstract of paper 870600, by Olson, says that while such reconstructions can be helpful, they can also be very misleading. And paper 900369 frames the evidentiary difficulty precisely: the determination of what would have been visible to a reasonably alert person must be made with proper consideration of both inter-person and intra-person variability. Neither of those sentences was written by a defense lawyer. Both were written by the people who built the discipline.

19.3Shoot from the driver's eye position inside the subject vehicle, and record the seat and mirror positions you used.

19.4If you cannot match a condition, substitute deliberately, measure the substitution, and disclose it in the caption rather than in a footnote.

Print the conditions on the face of the exhibit

A night exhibit is honest only when the method for matching conditions is documented and disclosed on the face of the exhibit itself. Not in the appendix. Not in the deposition answer you hope to be asked. On the exhibit: date and time, moon and cloud conditions, lamp configuration and aim, camera position and height, exposure settings, the calibration reference used, and the observer procedure if there was one. An exhibit carrying that block is a piece of evidence. An exhibit without it is a picture of a dark road, and a picture of a dark road proves nothing except that it was dark.

Chapter 19 Checklist: Night and Visibility

Chapter 20The View From Above

A drone is a tripod that flies, and every legal problem it creates comes from the flying rather than from the camera.

The certificate you need before you touch the controls

Start with certification, because it is the first question and the easiest one to fail. Civil small unmanned aircraft operations in the United States are governed by Title 14 of the Code of Federal Regulations, Part 107, which reaches aircraft weighing less than 55 pounds on takeoff including everything on board or attached. Section 107.12 prohibits manipulating the controls without a remote pilot certificate carrying a small unmanned aircraft systems rating, or direct supervision by a holder. Section 107.61 sets eligibility: at least 16 years old, able to read, speak, write, and understand English, with aeronautical knowledge established either by the initial knowledge test or, for a current Part 61 certificate holder, by the training specified for that path. Verify every section number in this chapter against the electronic Code of Federal Regulations before you rely on it, because this part moves.

Currency is the part practitioners get wrong. Section 107.65 is not a recurrent-test rule. Within the previous 24 calendar months the remote pilot must have accomplished one of several alternatives: passed an initial knowledge test, completed recurrent training, or, for a current Part 61 certificate holder, completed the training specified for that path. The recurrent knowledge test was replaced by recurrent training. The 24-month interval was not changed. Know which alternative you are relying on, because you will be asked.

Registration carries the same trap. Under Part 48, the exemption for aircraft weighing 0.55 pounds or less applies only to aircraft operated exclusively under Title 49 of the United States Code, section 44809, the recreational provision. An expert flying a sub-250-gram aircraft to document a scene is operating under Part 107 and must register it regardless of weight, with the unique identifier legibly displayed on an external surface. Remote identification under Part 89 has applied since September 16, 2023, satisfied by built-in remote identification, by a broadcast module flown within visual line of sight, or inside a Federal Aviation Administration Recognized Identification Area. The date many operators quote, in March of 2024, was the end of a published enforcement discretion policy, not a change to the regulation.

20.1Register every aircraft you fly for a case, whatever it weighs, and photograph the identifier on the airframe as your first frame of the day.

How high, how fast, and whose airspace

The operating limits sit in section 107.51. Groundspeed may not exceed 87 knots, which the rule itself expresses as 100 miles per hour. Altitude may not exceed 400 feet above ground level unless the aircraft is flown within a 400-foot radius of a structure and does not go higher than 400 feet above that structure's immediate uppermost limit. Minimum flight visibility is 3 statute miles, and the aircraft must remain 500 feet below and 2,000 feet horizontally from clouds. Section 107.31 requires that it stay visible throughout the entire flight. Section 107.41 requires air traffic control authorization in Class B, C, and D airspace and in the surface area of Class E designated for an airport; Class G requires none, and the Low Altitude Authorization and Notification Capability system supplies authorization at or below 400 feet at hundreds of airports.

Two limits matter disproportionately to scene work. Night operation under section 107.29 no longer requires a waiver, provided the remote pilot completed the knowledge test or training after April 6, 2021 and the aircraft carries anti-collision lighting visible for at least 3 statute miles. Only the lighting requirements are waivable; the training requirement is not. And section 107.145 governs flight over people inside moving vehicles: absent a closed or restricted access site with notice to those inside, the aircraft must not maintain sustained flight over them. Documenting an open roadway scene is squarely a section 107.145 problem, and transient transit is the only unwaived option. Note also that section 107.23, careless or reckless operation, does not appear on the waiver list. There is no waiver to operate carelessly.

Beyond visual line of sight operations still require a section 107.31 waiver, because Parts 108 and 109 are marked reserved and the proposed rule published in August 2025 remained proposed as this was written. And the falsification provision that used to sit at section 107.5 is now reserved, consolidated into sections 3.401 through 3.405. Any checklist you inherited citing section 107.5 is stale.

The Fourth Amendment cases do nothing for you

The Fourth Amendment aerial surveillance cases are not your rule of decision. California v. Ciraolo, 476 U.S. 207 (1986), Dow Chemical Co. v. United States, 476 U.S. 227 (1986), and the plurality in Florida v. Riley, 488 U.S. 445 (1989), all address government observation from navigable airspace.

A privately retained expert is not a state actor, so those cases neither govern you nor protect you, and the argument that the Supreme Court blessed aerial observation so your flight is fine is a category error.

What reaches you is state privacy tort, principally intrusion upon seclusion; common-law trespass, informed by the reasoning in United States v. Causby, 328 U.S. 256 (1946), that a landowner must have exclusive control of the immediate reaches of the enveloping atmosphere; and state statute. Altitude authorization is not a trespass defense.

Long Lake Township v. Maxon, decided by the Michigan Supreme Court on May 3, 2024, is the case most often described incorrectly. It is an exclusionary rule case, holding that the exclusionary rule may not be applied to civil enforcement proceedings that effectuate local zoning and nuisance ordinances and seek only prospective, injunctive relief. The court expressly declined to decide whether the warrantless drone surveillance was a Fourth Amendment search. Anyone who tells you Long Lake Township held that warrantless drone surveillance is a search has not read it.

20.2Assume the Fourth Amendment cases do nothing for you, and locate your authority to fly in consent, in a court order, or in a statutory exception.

The law that reaches you is the state statute

State anti-surveillance statutes are the live risk, and they are in force. In National Press Photographers Association v. McCraw, 90 F.4th 770 (5th Cir. 2024), the Fifth Circuit reversed the district court and upheld the Texas statute, reasoning that the law regulates not what image is captured but where it is taken from and how, and that drone operation is not inherently expressive conduct. Sources written before 2024 reporting the statute as struck down are wrong. Florida's statute, section 934.50, reaches private persons directly and shifts attorney fees. Oregon's private-actor provision turns on repetition and notice: an owner may sue where the operator flew over the property at least once before and the owner then objected, with treble damages and no altitude threshold. Nevada sets a 250-foot threshold with prior notice and carves out licensed land surveyors. California Civil Code section 1708.8 reaches entry into the airspace above the land, and its constructive invasion provision reaches capture by any device regardless of physical trespass.

20.3Before the flight, read the statute of the state you are standing in, and write the exception you are relying on into the engagement file.

What the overhead view shows that the ground cannot

Overhead geometry is the exhibit a ground-level photograph cannot produce. A plan view shows the spatial relationships among vehicles, marks, debris, obstructions, and roadway features without the perspective compression that makes a ground photograph arguable. An oblique aerial at a controlled height demonstrates sight lines, showing what a berm, a hedge, a parked trailer, or a crest concealed, in one image a jury reads without instruction. With control points on the ground, that coverage becomes measurable rather than merely illustrative. And when an opponent calls aerial mapping novel, recall that the Supreme Court in Dow Chemical described the camera before it as conventional, commonly used in mapmaking rather than an exotic sensory device.

Saying the drone took the picture is not foundation

Foundation is where careless aerial work dies. Federal Rule of Evidence 901(b)(9), authenticating a result produced by a process or system, is the workhorse for drone imagery. No decision was located applying it to drone footage, so treat what follows as analysis and not as authority: a conclusory foundation invites the objection that nothing was established about camera reliability or system specification. It is not enough to say the drone took the picture. Establish the aircraft and sensor make and model, firmware version, calibration status, satellite lock quality, how the imagery was written and transferred, and hash verification at acquisition. Rules 902(13) and 902(14) let you avoid a live foundation witness, but only with advance notice.

20.4Preserve flight logs and telemetry as evidence, not as an application cache.

Flight logs, telemetry, and raw files are electronically stored information. Reformatting the card after export, letting a cloud logging account lapse, or producing only the finished orthomosaic while discarding source frames are realistic exposures under Federal Rule of Civil Procedure 37(e). Preserve both altitude records and know the difference: barometric altitude is referenced to a takeoff-point zero, satellite altitude to mean sea level, and section 107.51 is written in feet above ground level. An opponent who catches you conflating them has a cheap and effective cross.

Chapter 20 Checklist: Aerial Documentation

Chapter 21Three Dimensions

There is no gate here. No published decision was located that squarely addresses the admissibility of laser scanning, structure from motion, or three-dimensional reconstruction, and the photogrammetry decisions that do exist are exercises in deference rather than scrutiny. The rest of this chapter is about what you do with that.

What structure from motion does with your photographs

Technique first, because your reliability record is built out of technique. Structure from motion recovers camera positions and a three-dimensional point set from ordinary overlapping photographs. The pipeline is published and inspectable: detect features, match them across images, solve for camera poses and point coordinates in a bundle adjustment, then fuse depth into a dense cloud. Nothing about it is exotic. Everything about it depends on how you shot.

Overlap is the first published number. The Metashape manual specifies eighty percent forward and sixty percent side overlap for aerial capture, and sixty to seventy percent for close-range work. Pix4D's planning guidance sets a general case of at least seventy-five percent frontal and sixty percent side, rising to ninety percent for large vertical objects. For an object you orbit, the figure is one image every 5 to 10 degrees, roughly 36 to 72 images per circuit. That converts directly into a shooting plan for a damaged vehicle.

The capture rules are equally concrete. Prefer a fixed focal length; if you use a zoom, set it to an end stop so the setting is repeatable. Keep the sensitivity setting, marked ISO on every camera, at its lowest value, choose an aperture deep enough to hold the object sharp, and keep the shutter fast. Shoot raw files, convert losslessly, and never crop or geometrically transform the images: the software solves for one consistent lens model per camera, and any per-image remapping breaks that assumption.

Some surfaces do not reconstruct at all. Untextured, shiny, reflective, and transparent objects are named failure cases, along with flat scenes, moving objects, and direct sunlight. The reason matters more than the list. A specular highlight is not attached to the surface, so the matcher tracks a phantom and the polished bumper you came to measure comes back as noise.

21.1Photograph a damaged vehicle in full orbits at 5 to 10 degree intervals, at three heights, before you photograph anything in detail.

Nothing in the photographs fixes the size of anything

A purely image-based reconstruction is determined only up to a similarity transform: position, orientation, and scale are arbitrary, and nothing in the images fixes the size of anything. Scale has to be imported, through certified scale bars oriented in several directions or through surveyed control points. And a subset of those points must be withheld from the solution and used only to measure residual error. A model validated only against the points used to constrain it is not validated at all.

How accurate photogrammetry is, in published numbers

Accuracy expectations are published and they are good. In Society of Automotive Engineers technical paper 2010-01-0065, Randles and colleagues measured vehicle crush by hand and by photogrammetry against a total station baseline: hands-on differed by 0.6 centimeters with a standard deviation of 1.4, photogrammetry by 0.1 centimeters with a standard deviation of 1.0. Close-range photogrammetry of crush is accurate to roughly a centimeter, at least as good as a tape measure in a trained hand. At scene scale, paper 2019-01-0423 placed evidence locations within 3.0 inches, about 8 centimeters, and had five independent analysts reach consistent results. Inter-operator reproducibility data is rare, and that is the citation for it.

Read the scanner specification with its conditions attached

Terrestrial laser scanning is the other route, sweeping a beam over the scene and recording range and angle for millions of points. Read the specifications with care. One instrument, as tabulated in a peer-reviewed laboratory study, publishes distance accuracy of 1 millimeter plus 10 parts per million at sixty-eight percent confidence, for a single-shot measurement to a surface of eighty-nine percent reflectance. Sixty-eight percent confidence is one standard deviation; a ninety-five percent figure would be roughly double. Eighty-nine percent reflectance is not a black tire, wet asphalt, or dark clothing.

That study also found the instrument stayed inside its manufacturer specification while showing clear systematic error components, including a striped pattern with discontinuities of about 2 millimeters at 20 meters. Systematic error does not average out with more points. Random noise does. An expert who reports a measurement as though millions of points had washed the error away has misunderstood the instrument.

Geometry compounds it. Beam divergence of half a milliradian grows the footprint by roughly half a millimeter per meter, so every returned point is the centroid of an illuminated patch and edges become mixed returns straddling foreground and background. Registration is a separate error budget again, compounding across setups, whether you register on targets of known geometry, by cloud-to-cloud matching, which fails predictably in a featureless corridor, or by the instrument's own tracking. Instrument performance is characterized by ASTM International test methods E2938 and E3125.

21.2Archive point data in the open interchange format specified by ASTM E2807, so another expert can independently examine what you produced. Note the naming confusion rather than repeating it: E57 is the committee, and also the common name of the point cloud format, but the format itself is specified by E2807.

What a tablet scanner can and cannot carry

The sensor in current Apple devices emits an 8 by 8 array diffracted into 3 by 3 grids, 576 points in total, with a stated maximum range of 5 meters. Against 2 million points per second, the envelope is obvious. Published evaluations give roughly plus or minus 1 centimeter on small objects and plus or minus 10 centimeters across a 130 meter cliff. Society of Automotive Engineers paper 2022-01-5083 compared a tablet scanner against a terrestrial scanner across ten vehicles and found an average difference of plus or minus 0.4 inch, with less than one percent error in the resulting change in velocity, called delta-V. Against that, a laboratory comparison could not produce an evaluable mesh at all on small objects, and noted that black is filtered out. Most consumer applications deliver a mesh rather than exportable raw points, which is fatal wherever the raw data is the deliverable.

The same caution applies to the view-synthesis methods now being marketed for reconstruction: they optimize photorealism, their headline metrics measure image similarity rather than geometry, and one published comparison found average distance differences around 1.3 meters on vertical surfaces. A photorealistic render is not a measurement.

The photogrammetry decisions are deference, not scrutiny

In United States v. Quinn, 18 F.3d 1461 (9th Cir. 1994), the court approved photogrammetric height testimony because the district court found the process nothing more than computer-assisted calculation involving no novel or questionable technique, and because the defendant pointed to nothing calling its reliability into question.

That is a burden-of-production ruling, not a reliability adjudication. United States v. Johnson, 114 F.3d 808 (8th Cir. 1997), contains no Rule 702 analysis at all, because admissibility was never challenged on appeal, and United States v. Kyler, 429 F. App'x 828 (11th Cir. 2011), is unpublished and was reviewed for plain error.

On the civil side, Chapman ex rel. Estate of Chapman v. Bernard's, Inc., 167 F. Supp. 2d 406 (D. Mass. 2001), is the leading federal decision, and its reliability holding runs one sentence: applying mathematics to photographs to derive measurements appears reasonable and does not strike the court as junk science. No error rate. No validation study. No factor analysis.

The most substantive treatment is State v. Matthews, 479 Md. 278, 277 A.3d 991 (2022), where an examiner performed reverse projection photogrammetry on scene with a calibrated height chart. The court held that the unknown degree of uncertainty in the estimate went to weight rather than admissibility, over a dissent. Make one terminology correction while you are there: neither Quinn nor Johnson uses the phrase reverse projection, and both describe scaling within a single image from objects of known size. Classical reverse projection requires re-photographing the scene with a reference of known height. That is what Matthews did.

21.3Never represent that photogrammetry or laser scanning has been searchingly validated by the courts. Say what the decisions hold, and say it first.

As for scanning itself, the negative finding rests on more than a failed search. The forensic community's own 2025 compilation of legal support for image examinations cites no case involving three-dimensional scanning, laser scanning, or point clouds. Those terms do not appear in it. State it the way it should be stated: no reported decision was located, and the leading compilation cites none. Not that none exists.

21.4Build the reliability record the courts have not built for you, and put it in the report before anyone asks.

That record is not mysterious, and Chapter 4 already told you what it is made of. Where no standard governs, conformance is demonstrated by documented, reproducible method. Here that means naming the instrument and quoting its specification with the conditions attached; stating capture geometry, overlap, and settings; publishing residuals at the withheld checkpoints; reporting registration statistics; reporting every measurement with its uncertainty, which Federal Rule of Evidence 702(d) as amended makes an admissibility matter; archiving in an open format; and preserving source frames and raw scans. Do that, and the absence of a case on point stops being your problem and becomes the other side's.

Chapter 21 Checklist: Three-Dimensional Capture

Part Four. The Record

Chapter 22Video Is Not Photography

A photograph measures a moment. A video measures time, and time is the thing people get wrong.

Everything in the first three Parts still applies. Exposure, focus, focal length, light, color, scale: a video frame is a photograph and obeys every rule a photograph obeys. Video adds one variable, and that variable is a clock. Almost every video case that falls apart falls apart at the clock, because a recording can be sharp, well lit, correctly exposed, and still produce a speed that is wrong by a factor you cannot defend.

Frame rate is a fraction, not a round number

Start with frame rate, because nothing downstream works until you know it. The rates people say out loud are 24, 30, and 60 frames per second. The rates cameras run are often 24000/1001, 30000/1001, and 60000/1001, which come out to 23.976, 29.970, and 59.940 frames per second. This is not a manufacturing tolerance. It is a deliberate 1000/1001 slowdown adopted when color was added to American analog television, so that the color subcarrier would sit in the right relationship to the sound carrier already in the air. Title 47 of the Code of Federal Regulations, Section 73.682, in the 2010 edition, states it in plain regulatory language: the vertical scanning frequency "corresponds nominally to 60 Hz (the actual value is 59.94 Hz)." Cite the 2010 edition, because the current text has removed the analog values and incorporates other standards by reference.

A file of 108,000 frames at 30000/1001 frames per second is 3,603.6 seconds long, not 3,600. Call it 30 frames per second and you have under-reported elapsed time by 3.6 seconds in every hour, which is one part in 1,001. Speed equals distance divided by elapsed time, so a shortened denominator inflates the answer by exactly one tenth of one percent. A true 60 miles per hour reads as 60.060.

22.1Write the frame rate into your notes as a fraction, never as a decimal, before you compute anything.

Now concede something, because conceding is how you keep the jury. One tenth of one percent is not the error that decides cases. It is the error you fix first so that the true one becomes visible, and the true one is in Chapter 25. An expert who presents the 29.97 problem as the great timing trap of video analysis has told the jury something true and unimportant, and has left the door open for the error that matters.

The shutter angle that looks right measures wrong

Motion blur is the next variable, and it is the one cinematographers have trained the world to get wrong. On a rotating disc shutter, exposure time equals the shutter angle divided by 360, times one over the frame rate. The 180 degree convention therefore gives an exposure of one over twice the frame rate: 1/48 second at 24 frames per second, 1/60 at 30, 1/120 at 60. That convention exists because it looks right. It is an aesthetic standard, and it is the wrong choice for measurement.

Work the numbers. A 50 mm lens on a sensor with 5.0 micrometer pixels, aimed at a car crossing the field of view at 25 meters per second, which is 55.9 miles per hour, from 30 meters away. One pixel covers 3.000 millimeters of that car. At 1/60 second the car travels 416.7 millimeters during the exposure, which is 138.9 pixels of smear. At 1/1000 second it is 8.3 pixels. At 1/4000 second it is 2.1 pixels. To get the smear under a single pixel at that geometry you need 1/8333 second.

Read that again with a measurement cursor in your hand. At the film-look setting, every edge on that vehicle is a gradient 139 pixels wide, roughly the width of its own headlight in the image. There is no edge to click on. The cost of fixing it is real: you lose about four stops of light, a stop meaning a doubling or halving of light, compared with a 180 degree shutter at 60 frames per second, and fast shutters produce strobing rather than smooth motion. Pay that cost on purpose and say in your report that you paid it.

22.2For any video you shoot to measure motion, set the shortest shutter the light allows, and record the shutter setting in the log at the moment you set it.

What a group of pictures actually stores

Modern video does not store pictures. It stores a group of pictures, and the Scientific Working Group on Digital Evidence defines the parts of it precisely in its Technical Overview of Digital Video Files, document 17-V-001. An intra frame, called an I frame, contains "all newly encoded information." A predicted frame, or P frame, contains "information relating to changes from the previous I-frame." A bidirectional frame, or B frame, is "encoded based on interpolation from the nearest I- and/or P-frames." Longer groups compress better and, in the working group's words, "do not capture quick transitions as effectively."

What that means in bytes is more persuasive than any definition. In a short test clip encoded with a group length of 15, the leading I frame occupies 3,683 bytes and the three frames that follow it occupy 78, 41, and 27 bytes. Nobody looking at 27 bytes believes it is an independent picture of anything. It is a set of instructions for modifying other pictures. Five things follow, and you should be able to say all five from the stand: you cannot verify a single frame in isolation, different decoders produce different pixels, damage to one predicted frame propagates until the next self-contained frame, open groups make frame extraction depend on data outside the group, and re-encoding will never reproduce the file bit for bit. The stored order is not the display order either, which the working group states directly in its Core Technical Concepts for Time-Based Analysis of Digital Video Files: "the order and timing in which frames are presented is not necessarily the same as the way the frames are stored within the multimedia container."

For video you create, use a codec that encodes every frame independently. Apple's published data rates for 1920 by 1080 at 29.97 frames per second run from 45 megabits per second for the proxy tier to 220 for ProRes 422 HQ and 495 for ProRes 4444 XQ. Storage is cheap. Explaining a predicted frame to a jury is not.

Two kinds of timecode, and how each is carried

The Society of Motion Picture and Television Engineers Standard ST 12-1:2014, Time and Control Code, defines it. Linear timecode is 80 bits per frame carried as an audio-band signal running from 960 Hz to 2400 Hz, which is why a spare audio track can hold it, and why it becomes unreadable when the transport slows or stops. Vertical interval timecode carries 90 bits, of which 32 are the time code, 32 are user data, 18 are synchronization, and 8 are a checksum, and it can be read on a paused frame.

Drop-frame timecode skips labels, not recorded frames

Drop-frame timecode is the single most misunderstood item in this field, so learn the sentence and say it three times. Drop-frame drops numbers, not frames. Every recorded frame is present in the file. What skips is the label: frame numbers 0 and 1 of the first second of every minute, except when the minute is divisible by ten, which removes 108 numbers per hour. Non-drop-frame timecode does not do this, so it reads 01:00:00:00 only after one hour and 3.6 seconds of real time. Drop-frame leaves a residual of about 0.108 frame per hour, roughly 3.6 milliseconds, and does not accumulate a full frame of error for about nine hours.

22.3When two cameras must agree, remember that genlock locks the rate and jam sync only sets the value; a jam-synced pair holds the same number and drifts apart all afternoon.

Chapter 22 Checklist: Video Fundamentals

Chapter 23The Measurements That Lie

Most bad video measurements come from bad video. This one comes from good video, correctly exposed, in focus, and wrong by a fifth.

Your sensor reads one row at a time

Nearly every camera sensor in service today reads out one row of pixels at a time, top to bottom, instead of capturing the whole frame at one instant. That is a rolling shutter. Every row in a single frame therefore has a different exposure start time, and the delay between adjacent rows equals the full sensor readout time divided by the number of rows. A propeller photographed this way bends. A car photographed this way leans. And if you measure that car, you will be wrong.

Convert the subject's speed into pixels per second on the image plane, which is focal length times world speed divided by the product of pixel pitch and distance. Multiply that by the per-row delay. The result is the tangent of the angle by which every vertical edge in the frame leans away from true vertical. The object's own size cancels out of that expression, which is the elegant part and the testable part: a lamp post, a door frame, and a building corner all lean by the same angle, so any known-vertical feature in the scene lets you solve for either the readout time or the subject's speed.

Put numbers on it with the same camera geometry used in the last chapter: a 50 mm lens, 5.0 micrometer pixels, a subject at 30 meters moving at 25 meters per second, which works out to 8,333 pixels per second of image-plane motion, on a sensor 3,000 rows tall. At a readout time of 66.7 milliseconds, one fifteenth of a second, every vertical edge leans 10.49 degrees and the top of the frame is displaced 555.6 pixels from the bottom. At 33 milliseconds it is 5.29 degrees. At 6 milliseconds it is 0.96 degrees. At 3.7 milliseconds it is 0.59 degrees. On a true global shutter it is zero.

Those readout times are not guesses. Jim Kasson has published oscilloscope measurements of shutter transit made by photographing a free-running function generator on a scope with a known time base and counting the displacement between the top and bottom of the frame. His figures: 3.7 milliseconds for a stacked Nikon Z9 in electronic shutter, about 6 milliseconds for the first-generation Sony a9, about 64 milliseconds for a Nikon Z7 in silent mode, and about 71 milliseconds for a Sony a7R II in silent full-frame mode. His class statement for pre-stacked Sony camera bodies is that electronic scan times "vary from about 1/14 second to 1/30 second for a full frame image," which is 33 to 71 milliseconds. Global shutter exists but is rare and recent: Sony's Alpha 9 III product page describes a full-frame stacked 24.6 megapixel sensor with a global shutter system and a "maximum 1/80000 second shutter speed with flash sync," and that flash sync claim is the diagnostic signature.

23.1Before you measure anything off a video frame, find out whether the sensor rolled, and how fast.

The length error is bigger than the skew

Skew is the artifact people know about. Length error is the one that decides cases. When the subject moves along the scan direction rather than across it, the sensor is chasing or retreating from the moving edge, and the object is stretched or compressed. Apparent height equals true height divided by one minus the image-plane speed times the per-row delay when the motion runs with the scan, and divided by one plus that product when it runs against. At the geometry above, a 66.7 millisecond readout elongates the object by just under twenty-three percent in one direction and compresses it by just under sixteen percent in the other. A 33 millisecond readout gives about ten percent and about eight percent. A 3.7 millisecond readout gives about one percent either way.

Twenty-three percent is not a rounding issue. It is the difference between a defensible measurement and an inadmissible one, and it is invisible in the image. The frame looks fine. The edges look like edges. Nothing in the file announces the error. Measuring position or length from a single rolling-shutter frame of a fast-moving object is invalid unless the readout time has been characterized and the geometry corrected, and that sentence belongs in your report in roughly those words.

How to spot a rolling shutter in the picture

The good news is that rolling shutter is diagnosable from the picture itself. A flash fires for tens to hundreds of microseconds, far shorter than a 33 to 67 millisecond readout, so only the rows integrating at that instant are lit, producing a bright band across part of the frame. That band proves rolling shutter and lets you estimate the readout timing. Pulsed light emitting diodes, which is to say most vehicle taillights, traffic signals, and dimmed interior lighting, produce horizontal banding whose spatial period encodes the switching frequency, and that banding can independently establish whether a lamp was energized. Curved propeller blades and elliptical wheels are the same evidence in cruder form.

Twenty years of research, and no published standard

There is real literature here, and citing it shows the field has thought about this for twenty years. Ait-Aider, Andreff, Lavest, and Martinet showed at the 2006 European Conference on Computer Vision that a single rolling-shutter frame encodes the subject's instantaneous velocity. Hedborg, Forssén, Felsberg, and Ringaby published rolling shutter bundle adjustment at the 2012 Conference on Computer Vision and Pattern Recognition, reformulating structure from motion so that each scan line carries its own camera pose. Dai, Li, and Kneip showed in 2016 that the ordinary epipolar constraint does not hold on rolling-shutter imagery. Albl, Kukelova, Larsson, and Pajdla published rolling shutter absolute pose in the 2020 volume of the Institute of Electrical and Electronics Engineers Transactions on Pattern Analysis and Machine Intelligence. Oth, Furgale, Kneip, and Siegwart published a calibration method in 2013, which matters because the line delay is a measurable camera parameter and not an excuse.

In the research behind this book, the complete published index of the Scientific Working Group on Digital Evidence was reviewed, including every video, imaging, photography, and multimedia document. No published guidance addressing rolling shutter was identified. That is not proof that none exists anywhere, and you should say it that carefully. But it means that if you wait for a standards body to hand you a method, you will testify without one.

23.2Close the standards gap in your own report, before your opponent opens it: name the effect, state the readout time you measured or the range you could not exclude, and state the resulting bound on your measurement.

Interlaced video is two pictures pretending to be one

Interlacing is the older version of the same problem, and it is still everywhere in legacy footage. A 1080i frame is not a frame. It is two fields, odd lines and even lines, captured at different instants and composited afterward. At 59.94 fields per second the field interval is 16.683 milliseconds and the frame interval is 33.367. When the subject moves between the two field captures you get combing, the horizontal teeth on vertical edges that people mistake for compression damage.

Every deinterlacer destroys information. Weaving recombines both fields and keeps the combing. Bobbing outputs each field as a full-height frame with the missing lines interpolated, which means half the pixels are invented. Motion-adaptive deinterlacing weaves the still areas and bobs the moving ones, which looks best and is forensically worst, because you cannot tell by inspection which pixels were recorded and which were manufactured.

Work from the fields, not from deinterlaced frames

For analysis you want the fields, not the frames. Separating the fields of a 720 by 480 interlaced clip yields twice as many images at 720 by 240. You trade half the vertical resolution for double the temporal resolution: 59.94 time samples per second instead of 29.97, and timing quantization cut from 33.4 milliseconds to 16.7. For a speed calculation that doubles your data points. No working group guidance on field-based analysis was identified either, so present it on the engineering, and note that any reader with the free tools can reproduce the field count.

23.3Check the declared field order and the per-frame interlace flags before you assume a legacy file is progressive.

Chapter 23 Checklist: Rolling Shutter and Fields

Chapter 24High-Speed Video and Mechanism

Some events are faster than the case can see. A discharge, an impact, a fastener letting go, a rope releasing: the whole mechanism lives in a few milliseconds, and ordinary video records the before and the after with nothing in between.

High-speed video answers a different question from the video in the last two chapters. It is not used to measure a scene. It is used to establish a sequence, which part moved first, whether the failure preceded the impact or followed it, whether the guard was in position when the energy arrived. That is a causation question, and causation is where mechanism testimony earns its fee.

The light requirement rises with the frame rate

The first constraint is light, and it is arithmetic rather than opinion. Exposure time shrinks in proportion to frame rate, so the light requirement rises in proportion. Relative to 60 frames per second, 500 frames per second needs 8.3 times the light, which is 3.06 stops, a stop meaning a doubling or halving of light. One thousand frames per second needs 16.7 times, or 4.06 stops. Five thousand frames per second needs 83.3 times, which is 6.38 stops. Ten thousand frames per second needs 166.7 times, 7.38 stops. That is why high-speed impact work runs continuous lighting arrays measured in tens of kilowatts. The cameras are not insensitive. There is simply no time to collect photons.

24.1Budget the lighting before you budget the camera, because the lights are the part that will stop the test.

The sensors compensate with enormous pixels. Photron publishes a 20 by 20 micrometer pixel pitch for the FASTCAM NOVA S20, which is roughly ten to twenty times the linear pitch of a phone sensor and one hundred to four hundred times the collecting area per pixel. That camera is specified at 18,750 frames per second at 1,024 by 1,024 in 10-bit mode, at 20,000 frames per second at 1,024 by 992, and at up to 1,100,000 frames per second at a reduced 128 by 16. Its minimum exposure is 0.2 microseconds, and Photron states that minimum is independent of frame rate, which matters more than the headline number: the frame rate sets how often you sample, and the exposure sets how sharp each sample is. Across the same product line at full 1,024 by 1,024, the S6 runs 6,400 frames per second, the S9 runs 9,000, the S12 runs 12,800, and the S16 runs 16,000. The smaller Mini AX series runs 2,000, 4,000, and 6,400 frames per second at that resolution with a shutter adjustable from 1 millisecond to 1 microsecond. The Orion S40 is specified at 31,250 frames per second at 1,280 by 1,024 with a 1.0 microsecond minimum exposure.

At the accessible end, the Chronos 2.1-HD records 1,920 by 1,080 at 1,000 frames per second on a 10 micrometer pitch sensor with a minimum shutter of 10 microseconds, at a published base price of $6,800. Kron publishes its record times: 2.7 seconds on 8 gigabytes, 5.5 seconds on 16, and 11 seconds on 32.

You get seconds of recording, not minutes

Those record times are the second constraint, and they are the one that surprises people. Work the arithmetic yourself: bytes per frame equals pixels times bit depth divided by eight, and record time equals memory divided by bytes per frame times frame rate. For the Chronos, 1,920 by 1,080 is 2.0736 megapixels, 12 bits packed is about 3.11 megabytes per frame, and 1,000 frames per second is about 3.11 gigabytes per second, so 8 gigabytes buys about 2.6 seconds. The manufacturer says 2.7. The arithmetic is honest.

You cannot press the button on a collision

You are recording for a few seconds, and you cannot press the button on a collision. These cameras run a continuously overwriting memory buffer and keep a window of frames around a trigger. The Chronos documents an End Trigger mode that "records until defined delay after trigger," which is functionally that behavior. Photron's product pages document memory segmentation rather than trigger modes. Manufacturer-documented pre-trigger percentages were not obtained in the research for this book, so specify them from the manual for the camera you rent, and put that page in your file.

24.2The trigger design is the experiment. Write the trigger scheme, the delay, and the pre-event window into the test plan before anyone loads a specimen.

A reported frame rate is a setting, not a measurement

Timebase is the third constraint, and the one experts skip. A camera that reports 5,000 frames per second is reporting a setting, not a measurement. If you intend to derive a velocity or a duration from frame counts, put an independent time reference in the frame. A calibrated timing light, a strobed source of known frequency, or a free-running signal displayed on an oscilloscope in the corner of the shot all work, and all of them let you verify the interframe interval from the recording itself rather than from the menu. This is the same trick used to measure sensor readout times, and it is good engineering rather than a published standard, so present it as method and show the result.

Slow motion is more persuasive than it deserves

Then there is the presentation problem, which is where high-speed work gets excluded. Slow motion is persuasive out of proportion to its content. Fifty milliseconds of a machine guard failing, played back over eight seconds, will feel to a jury like a leisurely, avoidable, obvious sequence of decisions. It was not. It was fifty milliseconds.

No decision addressing slow motion as an independent ground for exclusion surfaced in the research behind this book, and I will not pretend otherwise. But the objection has two doctrinal homes and both are live. Under Federal Rule of Evidence 403 the concern is that the exhibit misleads. Under Fusco v. General Motors Corp., 11 F.3d 259 (1st Cir. 1993), the operative question is whether a demonstration is "sufficiently close in appearance to the original accident to create the risk of misunderstanding by the jury," and slowing an event by a factor of 150 changes appearance more than any lens choice ever will.

The fix is disclosure, stated on the exhibit and stated from the stand. Burn the capture rate and the playback rate into the corner of every clip you show. State the real elapsed time of the segment in words before you play it, and state it again after. If the whole event is 42 milliseconds, say so, twice.

24.3Label every high-speed exhibit with capture rate, playback rate, and true elapsed time, on the image, not in a footnote.

Keep every take, including the misfires

High-speed footage is enormous, and it will tempt you to keep only the good takes. Keep everything, including the misfires and the frames before the trigger, because a partial production of a test series is a spoliation argument waiting to be made. And remember that a high-speed clip is a test result. Chapter 27 governs it, and the substantial similarity of your test setup to the real event is a foundation you must lay, not an objection you get to answer later.

Chapter 24 Checklist: High-Speed Capture

Chapter 25Surveillance Video: Get the Native File

Every surveillance video case has the same fork in it. One road leads to the file the recorder wrote. The other leads to a copy of a copy, and every measurement you make on that copy inherits an error nobody can bound.

Native export first, and the player with it

Take the first road. The Scientific Working Group on Digital Evidence, in its Best Practices for Data Acquisition from Digital Video Recorders, document 17-V-002, prefers a native or proprietary export because it is "likely to provide best evidence for legal authenticity purposes as it is closest to the original manner of recording," and it requires that you also take the viewer application and codec from the device itself. The working group does not publish a formal ranked list, so do not tell a court it did. The preference order is implied by the document's structure, and it runs from native export, to an open-format export taken as a secondary backup because it "should also be acquired if time is available," to network export, to a memory stick export, to optical media on older units, to a hardware capture device, to imaging the drive directly, to seizing the whole unit.

A disk image with the native files and the manufacturer's player preserves per-frame timestamps, the original compressed bitstream, camera identities, event and alarm logs, burned-in overlay text, and deleted fragments in unallocated space. It does not preserve the recorder's configuration, because as the working group notes, those settings "are retained on the device, not the hard drive." A native export through the unit's own export function keeps the codec and the timing and usually the overlay, and loses the deleted data and often the logs. A transcoded export to a common container keeps an approximate picture and throws away the per-frame timing, replacing it with an assumed constant rate, plus one full generation of lossy re-encoding, plus the true pixel aspect ratio. A screen recording of the player adds monitor scaling, monitor gamma, refresh-rate beating with the capture, the player's own deinterlacing, and a second generation of loss.

A screen recording of a digital video recorder is not evidence of what the camera saw. It is evidence of what one particular computer displayed on one particular day.

25.1Ask for the native file in writing, in the first letter, and quote the working group's preference language in the request.

The nominal frame rate in the file is a fiction

Confusing 29.97 with 30 frames per second costs you one tenth of one percent.

Surveillance recorders do not write at a constant frame rate. A sixteen-channel unit shares one encoding and disk pipeline across sixteen cameras, so each camera gets whatever is left. Motion and alarm triggers change the rate in the middle of a recording: the working group's Considerations for the Use of Time-Based Analysis of Digital Video for Court, document 19-V-003, gives the concrete example of systems that shift from 5 frames per second during idle periods to 15 when motion is detected, which is a threefold rate change inside one file. Under processor or disk pressure the unit drops frames silently. And the file does not contain a frame rate at all. It contains a timestamp for each frame, from which an average can be computed after the fact. The working group's Core Technical Concepts document says it directly: "Timing information in a multimedia video file is not directly stored in a single location; instead it must be decoded and calculated based on a number of data elements stored throughout the file."

In a test file built to imitate exactly that motion-triggered pattern, the container declares a nominal rate of 10 frames per second while the true average is 6.45, and there is a single gap between consecutive frames of 1.2 seconds. An analyst who counts one frame across that gap and calls it one tenth of a second, because the file says 10 frames per second, is wrong by 1200 percent. Compare that with the one tenth of one percent from calling 29.97 by the name 30. The ratio between those two errors is roughly four orders of magnitude.

Sit with that. The error most video experts talk about is one part in a thousand. The error hiding in the actual evidence is twelve to one. A pedestrian who covered 6 feet in a "one tenth of a second" gap is moving at 41 miles per hour, which is absurd on its face, and I have seen absurd numbers survive all the way to a report because the analyst trusted the nominal rate. The working group's own document says the formula "speed = total distance/elapsed video time" "may not accurately reflect movement in video." That sentence is citable and it is the whole chapter.

25.2For every measurement interval, use the actual elapsed time between the two specific frames you used. Never use frame count divided by nominal rate.

That failure mode is worth naming for its own sake, because it is the shape of most confident wrong answers in this field: a respected source, followed faithfully, on a tool that changed underneath it, producing silence that looks like agreement.

How to test a file for a variable rate

Work from the native file in the manufacturer's player. Extract per-frame presentation timestamps. Compute the difference between consecutive timestamps rather than an average rate. Tabulate or plot those differences, because the irregularity becomes obvious to a juror the moment it is drawn. Two independent checks tell you whether the file is variable in the first place: the container's nominal rate compared against its average rate, where a mismatch is the primary indicator, and the dedicated variable-frame-rate detector filter in the FFmpeg tools, which reports the fraction of intervals departing from the mode along with the minimum, maximum, and average frame duration. Only a zero result with matching minimum and maximum is a clean answer of constant rate. Where the container has no reliable timing at all, which is common in proprietary formats, establish timing empirically by recording an event of known duration through the same unit at the same settings, or by reading the recorder's own burned-in timestamps, and document which you did.

Capture the clock offset before you leave

The recorder's clock is usually wrong, sometimes by hours, because of daylight saving transitions, a dead backup battery, or an installer who never set it. Capture the offset at the moment of collection, because it is not recoverable later. Photograph the unit's on-screen clock in the same frame as a phone displaying a known-good time source, and write the offset down as a signed quantity with the direction stated in words. And do not change the time and date on the recorder. That instruction appears in the working group's acquisition document as a flat prohibition.

Legacy surveillance pixels are not square

Legacy surveillance formats do not use square pixels. A 720 by 480 frame intended for a 4 by 3 display needs a sample aspect ratio of 8 to 9; displayed naively as square pixels it is 11.11 percent too wide. A 704 by 480 frame is 9.09 percent too wide. The distortion is horizontal only, so it does not cancel, and it silently corrupts every horizontal length, every reverse-projection solution, and every facial or vehicle proportion you might compare.

25.3Check the encoder tag in the file metadata. A "native export" whose tags name a general-purpose media library was produced by software, not by the recorder.

25.4Collect a margin of time on both sides of the incident, not just the incident window, because you will want the idle-period frame rate to compare against.

Chapter 25 Checklist: Surveillance Video Collection

Chapter 26Recording the Witness

A bad recording of a good interview is a transcript full of the word "inaudible." Every one of those is a fact you paid to collect and then threw away.

Choosing the microphone, and getting it close

Technique first, because most of what goes wrong is physics rather than law. A condenser, which includes every lavalier and every small recorder's built-in capsule, responds to the diaphragm's displacement, which gives it extended high-frequency response and high sensitivity. A dynamic moving-coil microphone responds to velocity instead, which rolls off the top end, needs no power, tolerates abuse, and is markedly less sensitive. That last property is a feature in a bad room, because an insensitive microphone forces you to work close, and working close is the whole game.

Pattern follows construction. A capsule open only at the front is omnidirectional. One open front and back responds to the pressure gradient and is directional, which brings proximity effect, meaning bass boost as you approach, and heightened sensitivity to wind and handling noise. An omnidirectional capsule brings neither.

What a directional pattern buys you is one number: distance factor. Published directivity data gives an omnidirectional capsule 1.00, a cardioid 1.73, a supercardioid 1.93, and a hypercardioid 2.00. That number is how much farther the microphone can sit from the mouth and still hold the same ratio of direct sound to room sound. A hypercardioid buys you twice the working distance. That is all it buys you.

26.1Get the microphone close. Everything else is a rounding error against that.

The reason is the inverse square law. Direct sound falls 6 decibels for every doubling of distance, while the reverberant field is roughly uniform throughout the room. Move from 60 centimeters to 15 and you have gained 12 decibels of direct signal against an unchanged room. No blanket, no pattern, no filter buys 12 decibels.

Which is why the shotgun microphone is usually the wrong tool indoors. An interference tube cancels off-axis sound by path-length difference, and cancellation requires that difference to be a meaningful fraction of a wavelength. At 200 Hz a wavelength is 1.7 meters, and a short tube cancels nothing. Published measurements of a modern shotgun show a directivity index flat at 6 decibels from 63 Hz through 1 kHz, which is what a plain hypercardioid delivers, and a small room's reverberant energy lives in exactly that region. Outdoors the shotgun is the right tool. In an office it is a boom pole holding a hypercardioid.

Placement has a documented cost too. The microphone manufacturer DPA publishes measurements showing that for chest and neck placement, "the range between 2-4 kHz is practically removed from the sound spectrum, which results in poor speech intelligibility." That band carries the consonants. When a chest-mounted lavalier transcript is peppered with "inaudible," whether that reflects the speaker or the microphone is a real technical question with a real answer.

Choose the cluttered office over the glass conference room

The room matters more than the gear. Reverberation time, the interval a sound takes to decay by 60 decibels after the source stops, rises above about half a second in a bare conference room, and speech becomes hard to understand and harder to enhance. The heating and air conditioning system is the noise nobody hears during the interview and everybody hears on the recording. Choose the small cluttered carpeted office over the impressive glass conference room every time.

26.2Aim the directional microphone's null at the noise source, not just its front at the witness.

Where to set levels, and what format to record

Aim dialogue peaks at 12 to 6 decibels below full scale and keep the long-term average near 20 below. Be precise about the broadcast standards, because conflating them is an easy impeachment: European Broadcasting Union Recommendation R 128 sets a programme loudness target of negative 23.0 units relative to full scale, and Advanced Television Systems Committee Recommended Practice A/85 sets negative 24. Those are delivery targets, not recording targets. Record at 48 kilohertz, which Audio Engineering Society standard AES5 recommends "for the origination, processing, and interchange of audio programs," and at 24 bits, not because you will hear 146 decibels of range but because 24 bits turns level-setting from a risk into a non-issue. Record uncompressed, to a Broadcast Wave file, at roughly 8 megabytes per minute per channel. Never record an interview to a compressed format; perceptual coders discard exactly the content that later authentication analysis depends on.

26.3Record on two independent devices, with separate power and separate media. Not two tracks on one recorder.

Federal law is a floor, not a ceiling

Now the law, and the answer is that this is counsel's question and not yours. Title 18 of the United States Code, Section 2511(2)(d), provides that it "shall not be unlawful under this chapter" for a person not acting under color of law to intercept a communication "where such person is a party to the communication." Federal law is a floor, not a ceiling. It disclaims illegality only "under this chapter," and it never displaces a state statute.

A minority of states require the consent of all parties, and that minority is more complicated than any list will tell you. California, Florida, Maryland, Pennsylvania, and Washington are solid all-party jurisdictions reaching both in-person and telephone conversations. Massachusetts, Illinois, and Montana prohibit secret, surreptitious, or hidden-device recording rather than requiring consent, so open announced recording is lawful in all three even over objection. Connecticut's all-party rule is civil only and telephone only. Nevada requires all-party consent by telephone and one-party consent in person. Michigan's statute is ambiguous as to participants and the offense is a felony. Delaware has two statutes in conflict, and Vermont has no wiretapping statute at all.

Do not rely on that paragraph. It is a summary, it will drift, and it is not legal advice. Confirm the rule with retaining counsel, in writing, for every jurisdiction the conversation touches, before the interview.

Oregon, and the vacated opinion still in circulation

Oregon deserves its own correction, because most sources still report a holding that no longer exists. Oregon Revised Statutes Section 165.540 splits: telephone communications require one participant's consent, while an in-person conversation may not be obtained if "not all participants in the conversation are specifically informed that their conversation is being obtained." That is notice, not consent. On July 3, 2023, a Ninth Circuit panel held the in-person provision facially unconstitutional. On March 19, 2024, the court granted rehearing en banc and stated that "the three-judge panel opinion is vacated." On January 7, 2025, the en banc court affirmed dismissal, holding the statute content-neutral and concluding that it survives intermediate scrutiny. Project Veritas v. Schmidt, No. 22-35271 (9th Cir. Jan. 7, 2025) (en banc). Any text telling you the Ninth Circuit struck that statute down is describing a vacated opinion.

When the conversation crosses a state line

Across state lines there is no uniform rule. California's Supreme Court has applied a comparative impairment analysis to hold that California law governed calls recorded from another state into California, in Kearney v. Salomon Smith Barney, Inc., 39 Cal. 4th 95 (2006). Confirm the current citation and holding with retaining counsel before you rely on it.

In Becker v. Computer Sciences Corp., 541 F. Supp. 694 (S.D. Tex. 1982), a court applying the most-significant-relationship test reached the opposite result on similar facts. Washington's statute reaches communications "between points within or without the state" by its own terms.

Apply the most restrictive rule among every jurisdiction the call touches, and ask the witness where he physically is. Area codes prove nothing.

Announce the recording on the recording

Announce the recording on the recording. Your name, the date, the statement that you are recording, the question whether you have permission, and the answer. That satisfies every all-party statute surveyed, satisfies the notice-only statutes, defeats the secret and surreptitious elements in Massachusetts, Illinois, and Montana, and creates its own proof.

Ask counsel one more question before you press record: whether to record at all. A recording is a document, and whether it is discoverable is a legal question with consequences. What you do with the resulting statement belongs to a different book in this series, Being a Great Expert Witness. This one stops at the file.

Chapter 26 Checklist: Recording the Witness

Chapter 27Testing, Reenactment, and Substantially Similar Conditions

Documenting a scene and staging an event are different jobs with different rules. Experts get in trouble when they do the second while believing they are doing the first.

Documentation records a condition that already exists. Testing creates a condition and records what happens. The moment you build, load, drop, or drive something for the camera, you have entered a doctrine with its own foundational burden, and that burden falls on you before the exhibit is admitted rather than on your opponent afterward.

Substantial similarity is your burden, not theirs

The best-written statement of it is Fusco v. General Motors Corp., 11 F.3d 259 (1st Cir. 1993). Judge Boudin conceded the power of the exhibit at the outset: the test track footage was "vivid and pertinent." Then he explained why courts distrust it. The concern "lies not with use of tape or film" but "with the deliberate recreation of an event under staged conditions," because "jurors may be misled because they do not fully appreciate how variations in the surrounding conditions, as between the original occurrence and the staged event, can alter the outcome." From that the court drew the rule this chapter is built on, describing it as a doctrine predating and now loosely appended to Federal Rule of Evidence 403, one "that requires a foundational showing of substantial similarity in circumstances."

First, similarity is your affirmative burden. Second, and this is the point most experts miss, the label you attach does not control. General Motors argued the footage illustrated general scientific principles rather than recreating the accident. The court's answer: "the critical point is not one of labels. The issue for us is whether the demonstration is sufficiently close in appearance to the original accident to create the risk of misunderstanding by the jury."

Sufficiently close in appearance. That is a photographic test written into evidence law. Focal length, camera height, framing, and lighting all change appearance without changing a single physical fact, which is why every craft chapter in this book has a home here.

27.1Decide before you shoot whether the exhibit is illustrative or experimental, write that decision down, and then build the shoot to match the answer.

Illustrative or experimental, and the foundation each owes

The decision matters because it changes the foundation you owe. Four Corners Helicopters, Inc. v. Turbomeca, S.A., 979 F.2d 1434 (10th Cir. 1992) states the whole structure in four consecutive paragraphs. Where imagery is illustrative, "admissibility of experimental evidence does not depend on identical actual and experimental conditions" and "typically, dissimilarities go to the weight of the evidence rather than its admissibility." Where it purports to simulate the event, experiments "may be admissible if made under conditions which are substantially similar," and while "the conditions need not be identical, they must be sufficiently similar to provide a fair comparison." Then the two sentences to tape above your desk: "if the evidence is offered to merely show physical principles, the experiment should be conducted without suggesting that it simulates actual events," and "it is important then that the jury be instructed that the evidence is admitted for a limited purpose only."

Four Corners also shows what happens when a party tries to hold both positions. Turbomeca argued substantial similarity and, in the alternative, physical principles. The exclusion was affirmed on both grounds. You get one label, and the exhibit has to look like it.

How close is close enough, case by case

Gladhill v. General Motors Corp., 743 F.2d 1049 (4th Cir. 1984) shows what mismatch costs. The accident happened at night, on a downhill grade, at a sharp curve. The braking demonstration was conducted "on a flat, straight, asphalt surface in daylight by an experienced test driver." The Fourth Circuit reversed and directed that "evidence of this test should be excluded entirely on retrial," observing that "it is elementary physics that automobiles traveling downhill with their front wheels turned do not behave the same way with their brakes locked as those traveling on a flat surface with their front wheels straight." Four variables were mismatched: illumination, grade, curvature, and driver expectancy. Two are photographic conditions and two are physical. Courts do not separate them.

The doctrinal language across circuits is consistent enough to quote. Jackson v. Fletcher, 647 F.2d 1020 (10th Cir. 1981), where a test truck weighed 37,000 pounds less than the real one, states the purpose plainly: "the object of the rule which requires substantial similarity of conditions is to prevent admission of evidence which tends to mislead and perhaps confuse the jury." Barnes v. General Motors Corp., 547 F.2d 275 (5th Cir. 1977) and Hall both require conditions "so nearly the same in substantial particulars as to afford a fair comparison." Sanchez v. Denver & Rio Grande Western Railroad Co., 538 F.2d 304 (10th Cir. 1976) adds that where motion pictures "purport to represent a reenactment of human conduct the court should scrutinize the foundation with great care as to detail."

The other side of the ledger is where your exhibit lives if you build it correctly. In Randall v. Warnaco, Inc., 677 F.2d 1226 (8th Cir. 1982), a videotaped experiment came in, the court noting that "admissibility does not depend on perfect identity between actual and experimental conditions." In Bannister v. Town of Noble, 812 F.2d 1265 (10th Cir. 1987), the tape came in for a limited purpose under an instruction worth copying into your own proposed language: "The film is not being introduced for the purpose of attempting to recreate the accident involved in this case. ... The film is introduced only to demonstrate certain physical principles."

27.2If your exhibit is illustrative, draft the limiting instruction yourself and hand it to retaining counsel with the exhibit.

Federal Rule of Evidence 901 is the stronger attack

A second, independent route of attack lives in Federal Rule of Evidence 901 rather than Rule 403. United States v. Stearns, 550 F.2d 1167 (9th Cir. 1977) holds that "a photograph may be distorted, and thus inadmissible as a technically inaccurate representation of the scene photographed." Under Rule 403 your opponent must show that probative value is substantially outweighed. Under Rule 901 you must produce evidence sufficient to support a finding that the image is what you claim it is. Claim an image shows what the driver saw, and if your focal length compressed or expanded apparent distance, it is not what you claimed. That is the stronger objection, and the one to make against the other side's exhibit.

Document the setup as carefully as the result

Document the setup as thoroughly as the result. The model comes from the losing side in Four Corners, where the opposing expert catalogued the dissimilarities one variable at a time: a high-torque lathe against a low-torque engine, roughly 1,000 revolutions per minute against 35,500, all of the torque applied to the screw against a small fraction, a test running 45 to 54 seconds against an event lasting under two seconds, and no support behind the screw head and therefore no wedging. Quantified, variable by variable. That is the affidavit that killed the exhibit, and it is the affidavit you should write about your own test first.

27.3Photograph the rig before the test, during the test, and after the test, and photograph the instrument settings and the specimen identification in the same frames.

27.4Keep every take, including the ones that did not work, and produce them.

Give notice before a test destroys the article

A test destroys your opponent's opportunity to examine the same article in the same condition. Give notice, publish the protocol in advance, and invite attendance. An opponent who declined to attend a properly noticed test is in a much worse position than one who was never told.

Chapter 27 Checklist: Testing and Reenactment

Part Five. The Chain

Chapter 28The Testimony Inside the File

A photograph is a picture with a statement attached to the back, written by the camera, in a language nobody in the courtroom reads.

That statement is the image metadata: a block of structured text the camera writes at capture, recording what made the file, when, with what lens, at what settings, and sometimes where. It has a name, the Exchangeable Image File Format, and one fact about it surprises people: it is a camera industry standard published by a Japanese trade association, not an international standard. The companion framework, the Extensible Metadata Platform, is the one with an international designation.

Which fields can carry an opinion

Most of the block is device housekeeping. A handful of fields are load-bearing, meaning a timeline or an opinion can rest on them. Read those first and quote them in the report.

Start with time, because there are three time fields and experts confuse them constantly. DateTimeOriginal records when the scene was captured, at shutter release. DateTimeDigitized records when the data was written as digital data, the same instant in a native digital camera and a different one for a scanned film frame. DateTime records when the file was last changed, and editors update it on save. A modification date later than the capture date proves nothing. A modification date three weeks later, sitting beside a Software field naming an image editor, is a reason to ask the next question.

Make and Model name the camera body. Software records the name and version of the program or firmware that generated the image, the fastest tell that a file passed through an editor. BodySerialNumber ties the file to a specific instrument rather than a model, and LensModel and LensSerialNumber do the same for the optic. Then the exposure record: exposure time, aperture, the sensitivity field the standard now calls PhotographicSensitivity and older cameras label ISOSpeedRatings, and two focal length fields, one actual and one giving the equivalent on a 35 millimeter film camera.

28.1Read the metadata of every file you receive before you form an opinion from it, and put the reading in your file.

Why no camera clock keeps honest time

Now the clock, the first thing a competent opponent attacks. A camera's time comes from a free-running quartz oscillator backed by a small internal cell. No network time service, no satellite discipline, no correction of any kind. A human sets it once and it drifts from there. No manufacturer publishes a clock accuracy figure for a camera. The best anchor available is component level: a temperature-compensated precision real-time clock part sold on its accuracy is specified at plus or minus 2 parts per million from 0 to 40 degrees Celsius, about a minute a year. An ordinary camera clock is not that part and is not that good.

The second half is worse. Image metadata date and time values carry no time zone information at all. Fields recording the offset from Coordinated Universal Time were added only in version 2.31 of the format, in July 2016, so any body designed before then, and plenty after, writes a timestamp ambiguous by up to fourteen hours. If your case turns on whether a photograph was made before or after an event and the file carries no offset field, the timestamp alone cannot resolve it.

At the start of every session, photograph a known-accurate time display, a satellite receiver readout or a network-synchronized clock, in or immediately beside your first evidence frame. Repeat at the end. The capture time of that frame, compared against the time pictured in it, gives a signed offset, and bracketing the session gives drift across it, so you can state a bound rather than a point estimate. Write down the body and serial number, the reference, the time it showed, the time the camera recorded, the offset, and the zone you were standing in.

28.2Establish the clock offset by photographing a known time reference at both ends of every session, and record it in writing the same day.

Then do not correct the originals. Read-only first, always: examine with a tool running in read mode, hash the file before and after, and show the values match. If a correction is warranted, apply it to working copies and report it as a documented correction with the original values preserved. Rewriting timestamps on an original destroys the record you were retained to keep.

What the location block proves about your clock

Location tagging carries both the strongest cross-check in the file and the largest unforced error in the field. The location block holds latitude, longitude, altitude, and a heading, and its own time stamp and date stamp derived from the satellite signal in Coordinated Universal Time. That is a second clock, independent of the camera's, inside the same file. Subtract one from the other and you have measured the camera's clock error directly, after the fact, from the file alone, including on files you did not shoot and cannot re-examine. A difference of seven hours and four seconds means seven zones west plus four seconds of drift, not a fabrication.

Before relying on a coordinate, read the processing method field, which records whether the fix came from satellites, a cell tower identifier, or a wireless network database. A wireless or cell-derived fix is a positioning inference that can be wrong by hundreds of meters, and its time is not satellite-derived. Read the horizontal positioning error field too, the closest thing to an uncertainty bar the format offers. The published figures for the global positioning system are commitments about the signal and the constellation, not about the receiver in your pocket: a worst case pseudorange accuracy of 7.8 meters at a 95 percent confidence level, and horizontal error of 8 meters or better 95 percent of the time as a global average, with the caveat that actual accuracy depends on atmospheric effects, sky blockage, and receiver quality. An expert who testifies that location tagging is accurate to five meters has quoted a system specification as a device result.

It cuts at you as well. Your geotags disclose where you stood, when, and which way you faced. A request for all photographs from the site inspection can sweep in your rental car, your hotel, and an unrelated matter shot with the same body that afternoon. Decide the policy before the inspection, and never strip coordinates from an original afterward, because stripping metadata from an original is a spoliation problem.

28.3Decide whether location tagging is on or off before the inspection, and never change an original to match a decision you made later.

How ordinary handling strips the metadata out

Now the uncomfortable part, because it is about your workflow rather than somebody else's. Metadata dies from ordinary handling, and every path below is one experts use weekly. Email an image and the client, the server, or a resizer in between may rebuild it from decoded pixels, starting from an empty metadata block. Text it and the platform re-encodes and downscales. Download it from a social platform and presume the embedded metadata is gone: periodic testing by the photo metadata working group at the Embedded Metadata Initiative has repeatedly found that leading platforms display no embedded metadata and strip most of it from retrievable files. Those rounds are roughly a decade old; say so when you cite them. There is no verified public data on the major messaging applications, so do not testify about them from memory.

Cloud photo libraries look like storage and behave like a publishing pipeline. They re-encode, they generate derivatives, and what you retrieve may carry a creation date matching the day you asked for it. A screenshot is not a copy at all. It is a new photograph of a screen, carrying the metadata of the screenshot, meaning the device and the moment you took it, and none of the metadata of the photograph it depicts. Pasting an image into a document or a slide is the same operation wearing a suit: the container keeps a re-encoded rendering and discards the block.

When a file arrives with no metadata, that absence is evidence of the path the file traveled, not evidence that the photographer tampered with anything.

The exchange to rehearse before you testify

Counsel asks how you know when the photograph was taken. The weak answer is that the file says so. The answer that holds is that you verified the clock at both ends of the session, recorded a measured offset, kept the originals read-only, and can produce the log entry that says so. Same photograph. Different verdict.

28.4Never offer a metadata timestamp as a fact without stating how the clock was verified and what the offset was.

Chapter 28 Checklist: The File's Own Testimony

Chapter 29Chain of Custody for Files

There are two chains of custody in every digital case, and most experts keep only one.

The formulation is not mine. The National Institute of Justice's courtroom guide to digital evidence, catalog number 211314, states it plainly: two chains of custody are involved, the physical item itself and its associated data, and the issues regarding the data are additional to those regarding the item. The memory card is one chain. The file is a second, parallel chain, because a file can be altered without anyone touching the card it came from.

What a hash does, and what it does not

The instrument that holds the second chain together is a hash. A hash function takes a file of any size, runs every bit of it through a fixed procedure, and produces a number of fixed length written as a string of characters. Run it on a 40 megabyte raw file and you get the same length of output as from a one-line text file. Change one bit anywhere, one pixel or one character in one metadata field, and the output changes completely and unpredictably.

It is not encryption; you cannot recover the file from it. It is a fingerprint.

The tools are free and already installed on every operating system you own. For a folder of evidence, hashdeep hashes a whole directory tree, writes the values to a known-value file, and later audits the tree against it and reports every discrepancy. That audit is what you run four years from now, and it is why you build the list today.

Collision resistance is not preimage resistance

Which algorithm to use has become a deposition fight, and the fight rests on a misunderstanding worth getting exactly right. Hash functions resist two different attacks. Collision resistance asks whether an attacker can find two files, both of his own choosing, producing the same hash. Preimage resistance asks a different question: given a hash you already recorded, can an attacker construct a different file producing that same value. The National Institute of Standards and Technology quantified both in Special Publication 800-107, since withdrawn, and the asymmetry is the point and survives the withdrawal. Collision resistance strength in bits is half the output length; preimage resistance strength equals the full output length. SHA-1 produces 160 bits, so its collision strength was never above 80 bits while its preimage strength is 160. MD5, the older Message Digest algorithm, has never been an approved algorithm at all.

The published attacks are real, and know them by name because your opponent may. Collisions in MD5 were demonstrated in 2004. Collisions in SHA-1 were demonstrated on February 23, 2017 by a Dutch national research institute working with Google, at a cost of about 6,500 processor-years and 110 graphics-processor-years. In January 2020 a French and Singaporean team demonstrated a chosen-prefix collision for about $75,000, using roughly 900 consumer graphics cards for two months.

Every one of those is a collision attack. In every one, the attacker chose both files. That is not your situation.

Evidence integrity rests on preimage resistance and nothing else. You hashed a file at acquisition and wrote the value into your log. The question a court cares about is whether somebody could later build a different file matching that value. No published attack does this against any of these algorithms. The Scientific Working Group on Digital Evidence, the live standards body in this field, published a position on exactly this question on September 29, 2019, and two of its sentences belong in your file. It is appropriate to use both MD5 and SHA-1 for integrity verification, the working group wrote, provided the hash is securely stored or recorded in the examination documentation. And: since there are no preimage attacks on any of the four hashing algorithms discussed, the only way to manipulate the evidence without detection is to do it before it is hashed.

Read the conditional twice, because it carries the argument. The protection comes from the hash having been recorded contemporaneously, in a document existing independently of the file. A hash computed today from a file of unknown provenance proves the file has not changed since today. Nothing about the four years before that.

29.1Record the acquisition hash the day you acquire the file, in a document stored separately from the file.

None of that is a reason to keep using the old algorithms. The institute has announced a transition away from the Secure Hash Algorithm 1, written SHA-1, for all applications by December 31, 2030. Use the 256-bit member of that family, SHA-256, as your primary. The working group's best practice for digital evidence acquisition, revised March 3, 2025, directs hashing before and after a copy and again before storage. Two algorithms cost one extra command.

29.2Hash with SHA-256 and add a second algorithm, because the cost is one command and the benefit is a line of cross-examination you never answer.

The original, the working copy, and the line between them

The primary image is the first instance in which an image is recorded onto media that is a separate identifiable object, meaning what is on the card. The original image is an accurate and complete replica of it. A working copy is anything you process, and the moment you process it, it becomes derivative evidence carrying its own integrity requirements. The working group's guidance on maintaining the integrity of imagery, current as of March 3, 2025, sets the sequence: images are stored in an unaltered state, review and processing are completed on a copy, and hashing is performed before and after any copy operation and again before storage.

Write protection is the mechanical half of the same idea, and here most photographers are wrong. The sliding tab on the side of a memory card does not write-protect anything. The card's circuitry does not detect the switch position; the tab is a flag the host is expected to honor voluntarily, and some devices ignore it. If write protection matters you need a hardware write blocker, which does not depend on anyone's good behavior. The federal tool testing program publishes a specification whose first requirement states the point: a hardware write blocker shall not, after receiving an operation of any category from the host nor at any time during its operation, transmit any modifying category operation to a protected storage device.

Copy the card before you do anything else to it

Card handling is the same discipline one step earlier. Copy, do not move: the card is the only copy until a hash-verified duplicate exists elsewhere, and only then does it get wiped for reuse. Do not review and delete frames in camera, because deleting mid-shoot fragments the card, breaks the sequential file numbering that is itself corroborating metadata, and destroys the only copy of a frame you cannot assess on a three-inch screen. Prefer several small cards to one large one. If the camera body has two slots, set them to write simultaneously rather than to overflow. Format cards in the camera that will use them, not on a computer.

29.3Never format a card until a hash-verified copy exists on two separate devices, and log the verification before you format.

What your custody log has to record

The log ties it together, and it is unglamorous on purpose. Federal guidance on integrating forensic techniques describes a defensible one: a record of every person who had physical custody, the actions they performed and at what time, secure storage when the item is not in use, examination performed only on a copy, and verification of both original and copy. The law enforcement examination guide adds the standard for your notes: detailed enough to allow complete duplication of your actions, with dates, times, results, and the irregularities you hit and what you did about them. One page per case does it.

A duplicate comes in without a hash, so hash it anyway

The legal posture is easier than experts make it. A forensically sound copy is a duplicate under Federal Rule of Evidence 1001(e), and under Rule 1003 a duplicate is admissible to the same extent as the original unless a genuine question is raised about the original's authenticity. The burden of raising that question sits with the opponent. Hashing is not a prerequisite to admission, and Chapter 31 works through why. Do it anyway. It converts a contestable question into a certification, and its absence, while not fatal, is an invitation.

29.4Keep one custody page per case, in ink, listing every access, copy, and process by date, and treat it as an exhibit from the day you open it.

Chapter 29 Checklist: The Second Chain

Chapter 30Storage That Outlives the Case

The photograph you cannot produce in year four is worth exactly what the photograph you never took is worth.

Storage is where careful imaging goes to die, and it dies quietly. The failure surfaces at the worst moment: a supplemental production request in a case that went dormant for three years and then woke up.

The three-two-one rule has a citation problem

The rule everyone recites is the three-two-one rule: keep at least three copies of the data, on at least two different types of storage media, with at least one copy offsite. It is good advice with a provenance problem, and I say so plainly because conceding is a credibility move. The rule is commonly attributed to a federal computer emergency readiness publication on data backup options, and separately to the photographer Peter Krogh and his digital asset management work from 2005. Neither attribution could be verified in the research behind this book, and the federal cybersecurity agency's current ransomware guidance does not use the label at all. It says to maintain offline, encrypted backups and test their integrity.

So the rule is a widely propagated maxim with a weak citation trail. That does not make it wrong. It makes it a rule you defend on arithmetic rather than authority. The largest published drive reliability dataset, covering more than 337,000 drives, put the annualized failure rate for 2025 at 1.36 percent, down from 1.55 percent the year before. Hold one drive five years at that rate and survival is about 93.4 percent; hold three and the chance at least one fails is about 18.6 percent. That is the argument for three copies, and for verification, because a copy you never check is a copy you are only assuming exists.

30.1Keep three copies on two kinds of media with one offsite, and justify the rule from failure rates, not from a citation.

The national archives will not take your raw file

The National Archives and Records Administration publishes tables of formats it accepts for permanent electronic records, updated August 2025. For born-digital photographs the preferred format is the Tagged Image File Format, versions 4, 5, and 6, and nothing else. Acceptable formats include the JPEG File Interchange Format, Adobe's Digital Negative version 1.6.0.0, Portable Network Graphics version 1.2, and JPEG 2000 Part 1. Proprietary camera raw files do not appear at all. The national archives will not take your camera's native raw file. The Library of Congress, in its Recommended Formats Statement for 2025 to 2026, puts Digital Negative and proprietary camera raw in the same acceptable tier, so Digital Negative's advantage is real at the archives and nonexistent at the Library.

Keep both. Keep the camera original, because that is your primary and what an opponent will ask for. Beside it, write an uncompressed archival master in the Tagged Image File Format, because that is the file a person can still open in twenty years without a manufacturer's decoder.

30.2Archive both the camera original and an uncompressed archival master, because the first proves provenance and the second guarantees readability.

Bit rot is measured, not folklore

Bit rot is treated as folklore and has measured numbers behind it. Drives also return bad data quietly. A consumer class drive is specified on its datasheet at fewer than 1 unrecoverable read error in 10 to the 14th power bits read, which is 12.5 terabytes. Read an 18 terabyte archive drive end to end and expect about 1.4 such errors per pass. An enterprise nearline drive at 1 in 10 to the 15th power gets one per 125 terabytes. Those are bounds, not measured rates, and they set the floor: a full restore of a multi-terabyte archive is statistically likely to hit at least one bad sector.

The best published study observed 1.53 million disk drives over 41 months and found more than 400,000 checksum mismatches: 0.86 percent of nearline drives developed them against 0.065 percent of enterprise drives. The finding that should change your architecture is that 8 percent of the corruptions discovered were discovered during rebuild. The moment an array reconstructs itself after one failure is the moment the second problem finds you.

A redundant array of independent disks protects against a device dying. It does not protect against deletion, ransomware, overwriting, or your own mistake, and without block checksums it cannot tell a good copy from a corrupted one. Redundancy is availability. It is not backup. The defense is a file system that checksums every block plus a scheduled pass, called a scrub, that reads all the data and repairs what it can from a replica. The documentation for the checksumming file systems in common use recommends a scrub period of a month or less, and they ship weekly and monthly timers.

How long each medium lasts sitting unpowered

Media lifetimes are where the marketing is worst and the federal guidance is best. The institute's 2022 publication on digital evidence preservation gives expected longevity for unused media, meaning shelf life sitting unpowered. A pre-recorded optical disc: under 10 years, not recommended. A recordable compact disc or digital versatile disc: under 30 years. Blu-ray: under 30 years. Tape: 20 years. A hard disk drive: under 2 years, not recommended for archival. A solid state drive: under 1 year, not recommended.

Read those last two again. The semiconductor standards body specifies powered-off retention of one year at 30 degrees Celsius for client-class solid state drives and three months at 40 degrees for enterprise drives, and those minimums worsen with temperature. An external drive in a desk drawer is not an archive. It is a copy with an expiration date nobody wrote on the label.

The one optical medium the federal guidance rates for the long term is the inorganic-layer disc sold as M-DISC, at 100 years or more. Its marketing claims a thousand years, which the same manufacturer contradicts in other markets by advertising a hundred years or more, so do not print the thousand. The supporting evidence is accelerated aging, not observed longevity, and it is contested: a 2009 Naval Air Warfare Center report run on behalf of the technology found the organic-dye comparison brands failing the aging series, while later French national metrology laboratory testing rated the same disc under 250 hours at 90 degrees Celsius. Cite the federal number.

Write-once media earns its place for what it does to testimony. A recordable optical disc is write-once by construction, the cheapest true write-once medium available to a solo expert and the one most intelligible to a jury. In the cloud, an object lock in compliance mode cannot be overwritten or deleted by any user, including the account root user, until its retention date passes. Write-once turns "I did not alter the file" from testimony into a property of the system. It also protects only the copy you locked and does not prove the file was correct when you wrote it.

30.3Put at least one copy on media that cannot be rewritten, and be ready to say why that beats your own logbook.

The archive workflow a solo expert can keep up

Ingest by copying, never moving, and hash everything with SHA-256 before and after the copy. Designate the original and never touch it again. Work only on copies. Package the case as a bag with a checksum manifest. Distribute that bag three ways: a working drive running a checksumming file system, a cloud object store with versioning and an object lock, and an offline copy on a different medium kept somewhere other than your office. Then verify on a schedule you can keep: scrub monthly, revalidate the manifest quarterly, restore the offsite copy once a year. Log each verification. Once built, it takes about two hours a year.

30.4Test a restore from the offsite copy once a year and log the result, because an unverified backup is a belief, not a copy.

How long you keep it is not a technical question

How long must you keep case imagery? I found no standard, no rule, and no government publication prescribing a retention period for an expert witness, and I will not invent one. That is a legal and professional responsibility question, not a technical one. It turns on your engagement agreement, the statute of limitations and the statute of repose for the underlying claim, any protective order or litigation hold, and your carrier's requirements. Those push toward longer than you expect. Get the number from retaining counsel and from your carrier, in writing, per engagement, and write it into the case manifest.

Chapter 30 Checklist: The Archive

Chapter 31Authentication

Authentication is not a finding that your photograph is true. It is a finding that a jury could reasonably decide it is what you say it is.

Federal Rule of Evidence 901(a) provides: "To satisfy the requirement of authenticating or identifying an item of evidence, the proponent must produce evidence sufficient to support a finding that the item is what the proponent claims it is."

Read the operative words. Evidence sufficient to support a finding. Not proof, not certainty, not a certificate. As Lorraine v. Markel American Insurance Co., 241 F.R.D. 534 (D. Md. 2007), put it, a court need not find that the evidence is necessarily what the proponent claims, only that a jury might.

The three illustrations that carry imagery

Rule 901(b) then gives a list, introduced as examples only and not a complete list. Three of them carry imagery. Rule 901(b)(1), testimony of a witness with knowledge, is one line: "Testimony that an item is what it is claimed to be." Rule 901(b)(4), distinctive characteristics and the like: "The appearance, contents, substance, internal patterns, or other distinctive characteristics of the item, taken together with all the circumstances." Rule 901(b)(9), evidence about a process or system: "Evidence describing a process or system and showing that it produces an accurate result."

The Advisory Committee Note to that ninth example says it is designed for situations in which the accuracy of a result is dependent upon a process or system which produces it, and offers X-rays as a familiar instance. Hold that comparison. Nobody saw the inside of the bone; the image comes in because the machine works.

All of this reaches video and digital stills through Rule 1001(c): a photograph "means a photographic image or its equivalent stored in any form." Medium-agnostic on its face, which is why raw files, recorder containers, and frame grabs are all governed by the photograph rules.

31.1Learn Rule 901(a) as a sentence you can recite, because everything else here is an application of it.

Pictorial testimony against the silent witness

Two theories carry imagery into evidence. The first is the pictorial testimony theory, sometimes called the illustrative theory. The photograph is not independent evidence; it is a nonverbal way of expressing a witness's testimony. In United States v. Rembert, 863 F.2d 1023 (D.C. Cir. 1988), the court stated the foundation directly: a sponsoring witness, whether or not he is the photographer, who has personal knowledge of the scene depicted testifies that the photograph fairly and accurately portrays that scene. He need not know anything about the camera and need not have been present when the image was made. Rembert also declined to confine the rule to the two named models, noting that the uses of photography have not stood still and neither should the law.

The second is the silent witness theory. The image is itself substantive evidence of what it portrays, independent of any person who perceived the event, and the foundation comes entirely from the imaging system. In Bergner v. State, 397 N.E.2d 1012 (Ind. Ct. App. 1979), the Indiana Court of Appeals described the photograph as speaking for itself, with no need for a witness to testify that it accurately represents what he or she observed, and required evidence as to how and when the camera was loaded, how frequently it was activated, when the photographs were taken, and the chain of custody of the film.

Your own photographs travel under the pictorial testimony theory. You stood at the scene, you perceived it, and you can testify that the images fairly and accurately depict what you saw. Surveillance video travels under the silent witness theory, because no human perceived the recorded event. That is the harder foundation, and it is why Chapter 25 spends its length on the native file and the recorder.

United States v. Taylor, 530 F.2d 639 (5th Cir. 1976), is the foundational federal silent witness decision. A bank camera was activated only after employees were locked in the vault, so no teller could testify that the film accurately depicted what they saw. The foundation accepted was testimony about how the film was installed, how the camera was activated, that the film was removed immediately after the robbery, the chain of its possession, and that it was properly developed. Process testimony, nothing else, and it sufficed.

United States v. Stearns, 550 F.2d 1167 (9th Cir. 1977), adds the other half: the contents of a photograph itself, together with such other circumstantial evidence as bears upon the issue, may serve to explain and authenticate it. Stearns also opens with the sentence every imaging expert should memorize: a photograph may be distorted, and thus inadmissible as a technically inaccurate representation of the scene photographed. Distortion is an authentication problem, which is the doctrinal home of every lens question in Chapter 8. A competent opponent attacks foundation rigor, not the theory, and that is where he starts.

31.2Know which theory each exhibit travels under and say so in the report, because an exhibit with the wrong foundation built for it has no foundation.

The self-authenticating shortcut added in 2017

Since December 1, 2017 there has been a shortcut, and it is the part most often stated backwards. Rule 902(13) makes self-authenticating "a record generated by an electronic process or system that produces an accurate result, as shown by a certification of a qualified person that complies with the certification requirements of Rule 902(11) or (12)," with the proponent also required to meet the notice requirements of Rule 902(11). Rule 902(14) makes self-authenticating "data copied from an electronic device, storage medium, or file, if authenticated by a process of digital identification, as shown by a certification of a qualified person that complies with the certification requirements of Rule 902(11) or (12)," again with the notice requirement.

Paragraph (13) is the self-authenticating analogue of Rule 901(b)(9), the process route. Paragraph (14) is the analogue of Rule 901(b)(4) applied to bit-for-bit copies. Neither displaces Rule 901. Each avoids the cost of a live witness for something nobody contests.

The Advisory Committee Note to paragraph (14) explains the mechanism. Data copied from electronic devices, storage media, and electronic files, it says, are ordinarily authenticated by hash value, a number produced by an algorithm based upon the digital contents of a drive, medium, or file. If the hash values for the original and copy are different, the copy is not identical to the original. If they are the same, it is highly improbable that they are not identical, and identical hash values reliably attest that they are exact duplicates. The amendment allows self-authentication by a certification of a qualified person that she checked the hash value of the proffered item and that it was identical to the original.

Now the limiting sentence in the same Note, which keeps this from becoming a trap: "The rule is flexible enough to allow certifications through processes other than comparison of hash value, including by other reliable means of identification provided by future technology."

That forecloses the strong reading. Rule 902(14) makes hash matching a sufficient shortcut. It does not make hashing a prerequisite to admission. A proponent who never computed a hash may still authenticate under 901(b)(1), 901(b)(4), or 901(b)(9), and no decision I have found requires a hash, a write blocker, or a working-copy protocol as a condition of admitting a photograph. The National Institute of Justice's courtroom guide to digital evidence, catalog number 211314, does not discuss hash values at all, and reports that courts have generally not been receptive to alteration claims absent specific evidence of alteration. Two more sentences close it out: a certification can only establish that the item is authentic, and the opponent remains free to object on other grounds.

31.3Hash and certify because it removes an argument cheaply, not because the rule requires it, and never testify that hashing is required.

Two gaps to concede before the other side finds them

First, authenticating a photograph does not authenticate the container it was found in: Griffin v. State, decided by Maryland's high court on April 28, 2011, held that pages printed from a social media profile were not properly authenticated, because someone other than the purported creator could have made them. Second, the case law has not caught up to the hardware. The framework plainly extends to dashboard cameras, body-worn cameras, and drone footage, but I know of no decision squarely so holding. Say that as analysis, not authority.

What the rules ask, and what they do not

The rules do not ask whether your photograph is beautiful, or expensive, or made with the equipment somebody in the courtroom happens to own. They ask whether you can produce evidence sufficient to support a finding that the file is what you say it is. Every habit in Part Five exists to make that answer boring.

31.4Prepare the authentication foundation for every exhibit before you write the report, not the night before the deposition.

Chapter 31 Checklist: Foundation

Chapter 32Detecting the Faked and Defending the Real

The most common image authentication exhibit in American civil litigation is a printout from a free website with the bright patches circled in red. It proves nothing. The man who wrote the software says so on his own site, and almost nobody who offers the exhibit has read it.

The technique is error level analysis, and the mechanism explains in one breath. The tool takes your JPEG, the compressed format most cameras produce by default, resaves it at a chosen quality to introduce a known error everywhere, subtracts the resave from the input, and amplifies what is left. Because JPEG quantizes each 8 by 8 block of pixels independently, an untouched file should respond uniformly. In the developer's words, "If the image is completely unmodified, then all 8x8 squares should have similar error potentials." Bright regions are read as content that went through fewer compression cycles, which is to say, as content pasted in later.

Now read the same author's own limitations, on the same page as the tutorial. "[Error level analysis] only identifies what regions have different compression levels. It does not identify sources." If a file has been resaved enough times, "the [error level analysis] will return a black image and no modifications can be identified using this algorithm." And the sentence that should end the practice: "With Photoshop, the simple act of saving the picture can auto-sharpen textures and edges, creating a higher error level potential. This artifact does not identify intentional modification; it identifies that an Adobe product was used." The site's own frequently asked questions add that "There is a difference between real and authentic. A real photo of a forged document or a staged situation will not appear unusual," and that the site itself "does not draw any conclusions" and works "like a microscope."

There is a second problem, and it is a citable absence. The Scientific Working Group on Digital Evidence, the live body whose documents govern this field, published version 2.0 of its best practices for image authentication on March 3, 2025. Error level analysis is not in it at all.

Then there is the record in public. In May 2013 an analyst published a claim that the World Press Photo of the Year was a significantly reworked digital composite, resting on error level analysis, a shadow argument, and a reading of the file's metadata. The organization retained two independent examiners, who reported on May 14, 2013: "It is clear that the published photo was retouched with respect to both global and local color and tone. Beyond this, however, we find no evidence of significant photo manipulation or compositing." They added that "the analysis purporting photo manipulation is deeply flawed." The photographer had developed one raw file at different densities to balance uneven light in an alleyway, which is the digital equivalent of dodging a print. That is the technique's most public outing.

32.1Do not put error level analysis in your report. When an opponent puts it in his, read his own tool's documentation into the record before you say a word about photographs.

What the honest techniques can and cannot show

The honest techniques are better, and every one is narrower than its advocates admit. Double compression analysis, published in 2009 as the JPEG ghost method, finds a second error minimum where an inserted region was originally saved at a different quality. It exceeds ninety percent detection only when the quality difference is at least 20 and the tampered region is at least 100 by 100 pixels. Shrink the region to 50 by 50 and drop the quality difference to 5 and accuracy collapses to 5.4 percent. Photo-response non-uniformity, the sensor fingerprint left by manufacturing variation in silicon, has real scientific standing: a 2009 test across 1,053,580 images from 6,896 individual cameras reported a false rejection rate below 2.38 percent. It also needs hundreds of images to build the reference, degrades under compression and cropping, can be copied into a foreign image by an adversary who understands it, and is attacked directly by the multi-frame fusion inside every modern phone.

Copy-move detection finds duplication inside one image and says nothing about content spliced in from a different image. Color filter array analysis, which looks for the periodic correlations demosaicing imposes, reaches 97 to 100 percent accuracy on lossless files and falls to 6 to 56 percent at a JPEG quality factor of 70. Real evidence is almost always JPEG.

The working group states the limit better than I can. "The state of the art in digital imagery is such that in a single image, manipulations can be performed which a trained forensic practitioner may not adequately detect." Authentication "should be performed on a series of images depicting the same or similar subjects, or on video." And: "Results should not be reported in terms of numerical probability without a proper scientific foundation and/or related research."

32.2Never offer a single algorithm as an authentication result, and never attach a number to your confidence that a file is genuine.

What a signed manifest binds to your pixels

Provenance is the industry's answer, and it is a real advance. The content provenance specification behind what is marketed as Content Credentials binds three things into a signed manifest: assertions, which are structured statements about the file such as capture metadata, actions taken, and hashes; a claim, which is a signed wrapper referencing those assertions; and the signature itself, carried in a certificate chain and timestamped by an independent authority. The binding to the pixels is a cryptographic hash over byte ranges of the file, using the Secure Hash Algorithm at 256, 384, or 512 bits. Cameras ship it: the Leica M11-P was first in the world, on October 26, 2023, Sony launched signing across several camera bodies on June 26, 2025, and the first consumer phone arrived in August 2025.

Every in-camera signing system so far has been broken

Now the history, because it is the part that keeps this honest. In November 2010 a security firm defeated Canon's Original Data Security Kit by dumping the camera's memory, extracting the embedded secret keys, and computing the value responsible for detecting modification. In April 2011 the same firm defeated Nikon's Image Authentication System, which hashed metadata and image data separately with the Secure Hash Algorithm at 160 bits and signed the results with a 1024-bit key stored in a maker note tag. The finding was blunt: "The private (should-be-secret) cryptographic key is handled inappropriately, and can be extracted from camera." Nine models were affected.

Fourteen years later, on September 5, 2025, Nikon suspended its content credentials service on the Z6III. The reason was not cryptography. The camera's multiple exposure mode would combine signed photographs with unsigned photographs into one output file and then sign the result. Genuine credentials, attached to imagery that was never authenticated.

A device that signs autonomously must hold the key, and a key inside a device anybody can buy is a key an adversary can buy. A signing pipeline can also be induced to sign the wrong thing. Provenance proves an origin claim, not truth.

32.3Treat a content credential as evidence of where a file came from, never as evidence that the scene was real.

No detector generalizes past the generator it learned

Artificial intelligence generated imagery is where practitioners most want a tool and least have one. The National Institute of Standards and Technology addressed it in its November 2024 report on reducing the risks posed by synthetic content. Detectors do not generalize: they perform better on images from the generator they were trained on, and cross-generator accuracy runs 50 to 70 percent, which at the low end is a coin flip. Watermarking is not a fix either. Researchers have "theoretically proven and empirically confirmed" that for any conceivable image watermarking scheme the mark can be removed by adding noise and then denoising, and most current schemes "have been consistently found vulnerable to removal." Overall: "none of these techniques offer comprehensive solutions on their own."

Authentication is a record-keeping result, not a laboratory one

So what can you say under oath? Say what your own record supports. Federal Rule of Evidence 901(b)(1) still carries most photographs into evidence through a witness with personal knowledge, and Rule 901(b)(9) carries the rest through evidence describing a process or system that produces an accurate result. Authentication of a photograph is not a laboratory result. It is a record-keeping result. Shoot raw, hash at acquisition, record the hash in a contemporaneous log, photograph a known time reference, keep originals untouched, work only on copies. Every technique in this chapter is a fallback for the case where somebody failed to do that.

Chapter 32 Checklist: Authentication and Fakery

Chapter 33What You Must Turn Over

Every photograph you take in a federal case has two possible futures. It becomes your exhibit, or it becomes theirs. There is no third future in which it stays private because you decided it was not important.

Start with the text, because the text is where the argument lives. Federal Rule of Civil Procedure 26(a)(2)(B) requires that an expert disclosure "must be accompanied by a written report," one "prepared and signed by the witness," from a witness "retained or specially employed to provide expert testimony in the case." The report must contain "(i) a complete statement of all opinions the witness will express and the basis and reasons for them; (ii) the facts or data considered by the witness in forming them; (iii) any exhibits that will be used to summarize or support them," and then qualifications with a ten-year publication list, a four-year list of prior testimony, and "a statement of the compensation to be paid for the study and testimony in the case."

Subparagraphs (ii) and (iii) are the whole imaging question. Subparagraph (iii) is the easy half: your photo boards, overlays, animations, and scan renderings are exhibits that will summarize or support your opinions, and they go over. Subparagraph (ii) is the half that catches people.

Considered is a wider net than relied upon

Are your photographs "facts or data considered by the witness"? Work it through rather than guessing. The 2010 Advisory Committee Note to Rule 26 says the refocus onto "facts or data" was meant "to limit the disclosure requirement to material of a factual nature by excluding theories or mental impressions of counsel." Then it says the opposite of what an expert hopes: "At the same time, the intention is that 'facts or data' be interpreted broadly to require disclosure of any material considered by the expert, from whatever source, that contains factual ingredients. The disclosure obligation extends to any facts or data 'considered' by the expert in forming the opinions to be expressed, not only those relied upon by the expert."

Read the last clause twice. Considered, not relied upon. A photograph of the accident site contains factual ingredients by definition; that is the entire reason you took it. If you shot two hundred and six frames at a site inspection, looked at all two hundred and six on the rear screen of the camera or on your screen, and kept six for the report, you considered two hundred and six. The six you kept are relied upon. The two hundred you set aside are still considered.

33.1Assume every frame you press the shutter on is discoverable, and shoot accordingly.

What the draft protection covers, and where it stops

Federal Rule of Civil Procedure 26(b)(4)(B) provides that "Rules 26(b)(3)(A) and (B) protect drafts of any report or disclosure required under Rule 26(a)(2), regardless of the form in which the draft is recorded." That is broad in one direction and narrow in another. Broad, because form does not matter: a draft is protected whether it is written, electronic, or otherwise, and the protection reaches drafts of supplementation too. Narrow, because it protects drafts of a report or disclosure. That and no more.

The Committee said where the shield stops, and it said it in language written for exactly this problem. "Rules 26(b)(4)(B) and (C) do not impede discovery about the opinions to be offered by the expert or the development, foundation, or basis of those opinions. For example, the expert's testing of material involved in litigation, and notes of any such testing, would not be exempted from discovery by this rule." And even where a party makes a showing of substantial need, the protection for mental impressions "does not extend to the expert's own development of the opinions to be presented; those are subject to probing in deposition or at trial."

Site photographs are not drafts of a report. Test photographs are not drafts of a report. Point clouds are not drafts of a report. They are the testing and the notes of the testing, which the Note names as unprotected. There is one closer question, and you should concede it before you are pushed: a preliminary version of an animation built to be embedded in the report may well be a draft of a Rule 26(a)(2)(B)(iii) disclosure. The site photographs from which that animation was built plainly are not.

Three carve-outs in the attorney communication shield

Federal Rule of Civil Procedure 26(b)(4)(C) protects communications between the party's attorney and a reporting expert, "regardless of the form of the communications," with three carve-outs. Communications are discoverable to the extent they "relate to compensation for the expert's study or testimony," "identify facts or data that the party's attorney provided and that the expert considered in forming the opinions to be expressed," or "identify assumptions that the party's attorney provided and that the expert relied on in forming the opinions to be expressed."

That middle exception is an imaging exception more often than lawyers realize. When counsel emails you six scene photographs taken by an adjuster, that email identifies facts or data the attorney provided and you considered. The strategy discussion around it is protected. The photographs, and the fact that they came from counsel, are not.

33.2Log the provenance of every image in your file: who made it, when you received it, and from whom. You will be asked, and "counsel sent it" is not an adequate answer eighteen months later.

The consulting role, and who pays for production

Rule 26(b)(4)(D) shields the non-testifying consultant: ordinarily a party "may not, by interrogatories or deposition, discover facts known or opinions held by an expert who has been retained or specially employed by another party in anticipation of litigation or to prepare for trial and who is not expected to be called as a witness at trial," absent exceptional circumstances. If your role is truly consulting, that is a different posture, and it is counsel's call, not yours. Rule 26(b)(4)(E) says the party seeking discovery must pay you "a reasonable fee for time spent in responding to discovery" unless manifest injustice would result. Producing three hundred raw files and a photograph log is compensable time. Say so in your engagement agreement before it happens.

A frame you deleted is a frame you considered

Deleted frames deserve their own paragraph, because they are the place where a disclosure question turns into a preservation question. A frame you deleted is a frame you considered. Deleting it does not narrow the disclosure obligation; it converts a production problem into the subject of the next chapter. The Committee itself described the behavior the draft protection was designed to make unnecessary, and the description reads as disapproval: experts "might adopt strategies that protect against discovery but also interfere with their effective work, such as not taking any notes, never preparing draft reports, or using sophisticated software to scrub their computers' memories of all remnants of such drafts."

33.3Cull by selecting, never by deleting. Move frames out of the report, not out of the file.

Unlitigated is not the same as settled

I have found no case law addressing discovery of an expert's photographs specifically. Not a circuit decision, not a district decision squarely on the point. What exists is the rule text and the 2010 Committee Note, and on the question of whether you must produce frames you took but did not rely on, the Note is close to dispositive on its face. Say that in a deposition in exactly those terms if you are asked. Do not tell anyone the law is settled. It is not settled; it is unlitigated, and those are different things.

Shoot as if every frame will be an exhibit, because it may be. Not the frame you love. The frame you nearly did not take, the one with your own shadow in the corner, the one where the flash misfired. Somebody will get all of them, and the only question you control is whether they are a set you are proud of or a set you have to explain.

Chapter 33 Checklist: Disclosure

Chapter 34Spoliation and the Deleted Frame

The photograph most likely to end your involvement in a case is not one you took badly. It is one you took, looked at, and erased.

What Federal Rule of Civil Procedure 37(e) actually asks of you

Federal Rule of Civil Procedure 37(e) governs, and it is short enough to hold in your head. "If electronically stored information that should have been preserved in the anticipation or conduct of litigation is lost because a party failed to take reasonable steps to preserve it, and it cannot be restored or replaced through additional discovery, the court: (1) upon finding prejudice to another party from loss of the information, may order measures no greater than necessary to cure the prejudice; or (2) only upon finding that the party acted with the intent to deprive another party of the information's use in the litigation may: (A) presume that the lost information was unfavorable to the party; (B) instruct the jury that it may or must presume the information was unfavorable to the party; or (C) dismiss the action or enter a default judgment."

Three features of that text run this chapter. It reaches electronically stored information, which your image files, raw files, and video exports plainly are. It asks whether reasonable steps were taken, not whether the outcome was perfect. And it splits remedies at intent.

The 2015 Advisory Committee Note tells you where the duty comes from. "Rule 37(e) is based on this common-law duty; it does not attempt to create a new duty to preserve. The rule does not apply when information is lost before a duty to preserve arises." The Note adds a warning aimed at judges that experts should read as a small mercy: "It is important not to be blinded to this reality by hindsight arising from familiarity with an action as it is actually filed."

On the standard itself, the Note is generous. "Due to the ever-increasing volume of electronically stored information and the multitude of devices that generate such information, perfection in preserving all relevant electronically stored information is often impossible." And: "This rule recognizes that 'reasonable steps' to preserve suffice; it does not call for perfection."

Routine, good-faith operation of an electronic information system is a relevant factor, "although the prospect of litigation may call for reasonable steps to preserve information by intervening in that routine operation."

Three places the routine deletion keeps running after your duty attaches

Intervening in that routine operation is your job. It has three faces.

The first is the memory card. Reformatting between jobs is a good habit that becomes a bad one the moment a duty attaches. My rule is that a card is not reformatted until the files are copied, hashed, and confirmed present in two locations, and that a card carrying imagery in a matter where litigation is anticipated is not reused until counsel says so.

34.1Never reformat a card until the offload has been hashed and verified in two locations, and never at the scene.

The second is the cloud. Consumer photo services empty a trash folder on a fixed schedule. Storage tiers purge automatically when they fill. Settings labeled "optimize storage" remove full-resolution files and leave you a thumbnail. Each is the routine operation of an electronic information system, and each keeps running after your duty attaches unless you turn it off. The Note contemplates that "a 'cloud' service may fail" outside your control, but it also says courts "may, however, need to assess the extent to which a party knew of and protected against such risks." You know about the thirty-day trash. That knowledge is the whole problem.

34.2Disable every automatic deletion, purge, and optimize-storage setting on any device or account that touches evidence, and record the date you did it.

The third is the hold letter. When counsel issues a litigation hold, most experts read it as instructions to the client. It reaches you. Your cards, working drives, offsite copies, synced accounts, and email attachments are all within its scope if they hold matter-related imagery. I will not quote you the sentence about litigation holds that every article on this subject repeats, because it comes from two district court opinions whose full text I could not verify in a published reporter. An expert who quotes an opinion he has not read is one question from a bad afternoon. What follows is verified: the rule, the 2015 Note, and one case.

The duty reaches evidence you inspected but never owned

That case is Silvestri v. General Motors Corp., 271 F.3d 583 (4th Cir. 2001), and it belongs in every expert's file. A driver struck a utility pole and the airbag did not deploy. His experts "inspected and photographed the vehicle and inspected the site, and each prepared a report of his findings." The vehicle then went to an insurer, to a repair shop, and out into the world. General Motors was not told about the accident for almost three years.

The court defined the wrong: "Spoliation refers to the destruction or material alteration of evidence or to the failure to preserve property for another's use as evidence in pending or reasonably foreseeable litigation." The duty "arises not only during litigation but also extends to that period before the litigation when a party reasonably should know that the evidence may be relevant to anticipated litigation." And it stated the obligation that reaches nearly every expert, because you rarely own the thing you examine: "If a party cannot fulfill this duty to preserve because he does not own or control the evidence, he still has an obligation to give the opposing party notice of access to the evidence or of the possible destruction of the evidence if the party anticipates litigation involving that evidence."

The photographs did not save the case. The court said the opposite: "To require General Motors to rely on the evidence collected by Silvestri's experts in lieu of what it could have collected would result in irreparable prejudice." Dismissal affirmed.

The expert's photographic record was excellent, and its excellence proved the party knew the vehicle mattered, examined it, and let it go without telling anyone. Thorough documentation is not a defense to spoliation. It can be the evidence of spoliation.

34.3When you inspect something you do not control, tell counsel in writing the same day that the item exists, where it is, and that it is at risk.

Cull by selection, never by deletion

Culling is not the problem. Culling silently is the problem.

Culling as a discipline means you shoot everything, preserve everything, and then select. The report carries six frames. The file carries two hundred and six. Your log states which frames were selected and why, and the unselected frames are produced with the rest. The selection becomes a demonstration of method.

Culling as a problem means you delete. Then the frames are electronically stored information that should have been preserved, whether they can be "restored or replaced through additional discovery" is somebody else's question, and the argument about your intent has begun in a proceeding where you are not the client and cannot control how it is litigated.

Rule 37(e)(2) does raise the bar for the severe sanctions. The Note is explicit that the rule "rejects cases such as Residential Funding Corp. v. DeGeorge Financial Corp., 306 F.3d 99 (2d Cir. 2002), that authorize the giving of adverse-inference instructions on a finding of negligence or gross negligence," because "[n]egligent or even grossly negligent behavior does not logically support that inference." Do not read that as comfort. The Note preserves a path under subdivision (e)(1) needing no intent finding at all: a court may allow "the parties to present evidence to the jury concerning the loss and likely relevance of information." Loss without bad intent is not costless.

Rule 37(e) "applies only to electronically stored information." A lost print, negative, or slide falls outside it and back onto the court's inherent authority and the common law, which is where Silvestri operates.

No published case yet reaches an expert's own deleted frames

I have found no authority applying any of this to an expert's own deleted frames, raw files, or point clouds. That is a real gap, and I state the conclusion as inference: the rule is medium-agnostic on its face, nothing in it exempts an expert's working files, and Silvestri shows the duty attaching to material an expert generated. The 2015 Note treats "a party's own information-retention protocols" as a relevant independent obligation, which cuts both ways. Better to have one and follow it than to improvise under oath.

34.4Write a one-page retention protocol, apply it to every matter identically, and never make an exception for a case that is going badly.

Chapter 34 Checklist: Preservation

Chapter 35Where You May Point the Camera

A camera confers no right of entry. Every place you want to photograph belongs to somebody, and the question is never whether you can get the angle. It is whose permission the angle requires, and in what form.

The rule that lists photograph among the things you may do

The ordinary vehicle is Federal Rule of Civil Procedure 34(a)(2), which lets a party serve a request "to permit entry onto designated land or other property possessed or controlled by the responding party, so that the requesting party may inspect, measure, survey, photograph, test, or sample the property or any designated object or operation on it." Read the verb list. "Photograph" is expressly enumerated, so photography on a Rule 34 inspection needs no separate authority. Note also "possessed or controlled": control is the touchstone, not ownership.

Rule 34 reaches only parties. For everyone else, Rule 34(c) builds the bridge: "As provided in Rule 45, a nonparty may be compelled to produce documents and tangible things or to permit an inspection." Rule 45(a)(1)(A)(iii) lets a subpoena command a person to "permit the inspection of premises," and Rule 45(d)(2)(A), captioned "Appearance Not Required," provides that such a person "need not appear in person at the place of production or inspection unless also commanded to appear for a deposition, hearing, or trial." That power dates to the 1991 amendments, not the 2013 ones.

Why an inspection is harder to win than documents

Getting an inspection ordered is harder than getting documents. The framework is relevance and proportionality under Rule 26(b)(1), overlaid with the balancing stated in Belcher v. Bassett Furniture Industries, Inc., 588 F.2d 904 (4th Cir. 1978): "Since entry upon a party's premises may entail greater burdens and risks than mere production of documents, a greater inquiry into the necessity for inspection would seem warranted," and "the degree to which the proposed inspection will aid in the search for truth must be balanced against the burdens and dangers created by the inspection." The order there let plaintiffs' expert roam five plants over five days and question employees without designated areas of inquiry. Reversed as improvidently granted. On the other side, New York State Association for Retarded Children, Inc. v. Carey, 706 F.2d 956 (2d Cir. 1983), affirmed an inspection expressly permitting experts to photograph conditions.

35.1Write the photography terms into the inspection stipulation yourself. Counsel negotiates access; only you know what you need to shoot.

Courts condition rather than deny, and Rule 26(c)(1) supplies the menu: terms of time, place, and expense, limits on scope, and, in subparagraph (E), "designating the persons who may be present while the discovery is conducted." A workable protocol fixes the duration, the areas in scope, the attendees, one designated photographer with an equipment list, what may and may not be imaged, and a bar on talking to the owner's employees. In Belcher, unsworn on-site questioning of employees was a principal ground for reversal, because it functions as an unrecorded deposition with none of the protections.

One step onto the neighbouring parcel is a trespass

Private property outside a discovery order runs on consent and trespass law. In the general common-law formulation, trespass is complete on intentional entry and liability does not depend on harm, so the expert who steps twenty feet onto a neighboring parcel for a cleaner angle has likely committed a tort even though nothing was touched. The elements vary by state, and the forum state's are the ones that control. There is no news-gathering escape hatch: Branzburg v. Hayes, 408 U.S. 665 (1972), holds that "newsmen have no constitutional right of access to the scenes of crime or disaster when the general public is excluded." And access obtained by concealment is fragile. In Food Lion, Inc. v. Capital Cities/ABC, Inc., 194 F.3d 505 (4th Cir. 1999), trespass liability was affirmed not because of the misrepresentation used to get through the door, but because of the wrongful acts committed after entry: "consent to enter is vitiated by a wrongful act that exceeds and abuses the privilege of entry." Read that distinction carefully, because most summaries get it backward.

35.2Never enter a parcel you were not granted, and never obtain access by concealing who retained you.

Getting a camera into a prison or a hospital

Custodial facilities run on equal access, not on any right to walk in with a camera. Pell v. Procunier, 417 U.S. 817 (1974), holds that "newsmen have no constitutional right of access to prisons or their inmates beyond that afforded the general public," and Houchins v. KQED, Inc., 438 U.S. 1 (1978), refused to compel a television station's entry to photograph jail conditions. Federal Bureau of Prisons media regulations, at title 28 of the Code of Federal Regulations, part 540, subpart E, give an inmate "the right not to be photographed and not to have his or her voice recorded by the media" and require a media representative "to obtain written permission from an inmate before photographing or recording the voice of an inmate." Read the qualifier. Those provisions govern news media, and a retained civil expert is not a media representative. Your route is a stipulation, a Rule 34(a)(2) request, a Rule 45 subpoena, or special arrangement with the warden's approval, each of which comes with an escort and equipment pre-approval.

A private hospital is private property, and entry turns on facility policy and owner consent. The federal medical privacy law, the Health Insurance Portability and Accountability Act, binds the provider and not the visitor, but the provider carries the enforcement risk and therefore enforces categorical no-photography rules. A New York hospital entered a two million two hundred thousand dollar resolution effective April 19, 2016 after a film crew was given access. Get two consents, facility and patient. Photograph the physical environment only, keep patients and staff out of frame, and get a protective order over the images.

What the criminal rule everyone cites does not reach

Now the courthouse, where I see competent people make the most avoidable mistake. Clear away the rule everyone cites. Federal Rule of Criminal Procedure 53 bars photographing in the courtroom during judicial proceedings and broadcasting from the courtroom. It is a criminal rule, and it does not by its own force govern civil proceedings, hallways, lobbies, or the building exterior. Those are governed by Judicial Conference policy, local rules, and federal property regulations, and it is those, not Rule 53, that stop you at the door.

Judicial Conference policy, in the Guide to Judiciary Policy, volume 10, chapter 4, last revised September 18, 2023, "does not allow either civil or criminal courtroom proceedings in the district courts to be broadcast, televised, recorded, or photographed for the purpose of public dissemination." A judge may nonetheless authorize camera or recording use "for the presentation of evidence; for the perpetuation of the record of the proceedings; for security purposes; for other purposes of judicial administration; for the photographing, recording, or broadcasting of appellate arguments; or consistent with pilot programs." The prohibition dates to 1972.

"For the presentation of evidence" is your hook, and it is the answer for the expert who wants to carry a laptop, projector, tablet, or camera into a federal courtroom. It is not self-executing. It is a request, and the vehicle is a pretrial order naming the equipment and the purpose.

Local rules are the trap, because several districts bar the device from the building, not merely from the courtroom. The joint local civil rule of the Southern and Eastern Districts of New York, effective January 2, 2025, provides that no one other than court officials on court business shall "bring any camera, transmitter, receiver, recording device, cellular telephone, computer or other electronic device into any courthouse." Read that again. Not the courtroom. The courthouse. The Northern District of Ohio prohibits photography, recording, and broadcasting "in the Court House" and authorizes officers "to seize and inspect devices." The District of Maryland permits devices after screening but requires them off in the courtroom. Three districts, three answers, none in the Federal Rules.

35.3Check the local rule and standing order for the specific courthouse, by name, before every appearance. Assume nothing from the district next door.

Photographing a federal lobby counts as a commercial purpose

One more provision catches experts working premises cases in federal buildings. Title 41 of the Code of Federal Regulations, section 102-74.420, provides that except where security rules or a federal court order or rule prohibit it, persons on federal property may photograph space occupied by a tenant agency "for non-commercial purposes only with the permission of the occupying agency concerned," such space "for commercial purposes only with written permission of an authorized official of the occupying agency concerned," and "building entrances, lobbies, foyers, corridors, or auditoriums for news purposes." An expert documenting the lobby floor of a federal building in a slip-and-fall case is photographing for commercial purposes. Written agency permission, on top of any court rule.

35.4Get written permission from the occupying agency before photographing inside a federal building, and carry the paper with you.

State courts vary too much to summarize, and Chandler v. Florida, 449 U.S. 560 (1981), only tells you a state may permit camera coverage, which is not your right to bring one. Never plan an exhibit around equipment you have not confirmed you may carry through the door.

Chapter 35 Checklist: Access and Permission

Part Six. The Verdict

Chapter 36Rule 403 and the Photograph That Wins the Battle

The photograph most likely to lose your case is the one everybody in the room admires.

One sentence decides more imaging disputes than the rest

Federal Rule of Evidence 403 is one sentence long and it decides more imaging disputes than every other rule combined. It reads: "The court may exclude relevant evidence if its probative value is substantially outweighed by a danger of one or more of the following: unfair prejudice, confusing the issues, misleading the jury, undue delay, wasting time, or needlessly presenting cumulative evidence." That text was enacted in 1975 and restyled in 2011, the Committee Note stating that the changes were "intended to be stylistic only."

Read the balance carefully, because experts state it backwards under oath more often than they state it correctly. The proponent does not have to prove that probative value outweighs prejudice. The opponent has to show that probative value is substantially outweighed. Rule 403 is a thumb on the scale in favor of admission. Which means that when your photograph is excluded under this rule, it was not a close call. It failed a test designed to let evidence in.

Rule 403 does not ask whether your photograph is accurate. It assumes accuracy and asks what the photograph does to the jury beyond proving the fact it was offered to prove. An image can be perfectly authenticated, perfectly measured, perfectly relevant, and still come out, because its effect exceeds its function.

Gruesome will be allowed, showmanship will not

Start with gruesome injury imagery in a civil case, where the law is more permissive than most experts expect. In Jenkins v. Associated Transport, Inc., 330 F.2d 706 (6th Cir. 1964), a civil wrongful death and personal injury case, colored photographs of the injured plaintiff's condition were admitted and the admission was affirmed. The court warned in the same breath that it "would not condone, nor should a trial judge, the use of lurid photographs, paraphernalia, or other trappings of showmanship to enhance a plaintiff's verdict," and then held that proper trial conduct was not offended. Two details are worth carrying: the sponsoring witness was the treating physician, not the photographer, and uncertainty about the exact dates the photographs were taken was not fatal because the doctor identified them as fair representations of the plaintiff's condition.

So the line is not gruesomeness. The line is showmanship. A court will let a jury see an injury. It will not let a party stage one.

Why the stills came in and the video stayed out

The sharpest lesson in this area comes from a single opinion that split the difference between two media. In Thomas v. C.G. Tate Construction Co., 465 F. Supp. 566 (D.S.C. 1979), the plaintiff offered a 27 minute audio and video recording of burn treatment: bandage removal, whirlpool therapy, rebandaging, with visible expressions and contortions of pain, groaning, and statements on the soundtrack. The court excluded it, finding "that the probative value is substantially outweighed and over-shadowed by the danger of unfair prejudice," and relying on the recording's "dominating effect" that "will distract the jury from its proper consideration of other issues." Then the court admitted 19 still photographs of the same injuries, reasoning that stills lack the dramatic, dominating effect of moving video.

Same injury. Same plaintiff. Same day. One medium in, one medium out.

36.1When the injury is severe, produce the still photographs first and treat moving video as a separate, harder decision that counsel makes with a written record of why.

The four concerns a day-in-the-life film has to survive

Day-in-the-life films sit at the center of this fight and the federal authority runs in both directions. In Bannister v. Town of Noble, Oklahoma, 812 F.2d 1265 (10th Cir. 1987), the Tenth Circuit affirmed admission of a videotape showing the plaintiff getting around school, entering his car, pumping gasoline, and doing tasks at home. The court acknowledged that "such films purport to show how an injury has affected the daily routine of its victim" and are often wanted because films "illustrate, better than words, the impact the injury had had on the plaintiff's life," while stating plainly that admission "raises obvious dangers of prejudice to the opposing party." The standard of review is abuse of discretion, which is why both admissions and exclusions get affirmed.

The Tenth Circuit gathered four concerns, and they are your build specification whether you are making the film or attacking one. Accuracy and fairness of foundation, drawn from Sanchez v. Denver & Rio Grande Western Railroad Co., 538 F.2d 304 (10th Cir. 1976), which held that motion pictures "must be premised by a foundation of accuracy and fairness" and that where they purport to represent a reenactment of human conduct "the court should scrutinize the foundation with great care as to detail." Self-serving behavior, because a plaintiff who knows he is being recorded for litigation is "likely to cause self-serving behavior, consciously or otherwise." The dominating nature of film. And the loss of cross-examination, because a film cannot be questioned.

That last concern produced the strongest federal appellate authority on the subject, and it is a reversal. In Foster v. Crawford Shipping Co., 496 F.2d 788 (3d Cir. 1974), plaintiff's counsel took a videotape crew to an incompetent plaintiff's home on the evening of the second trial day, without notice to the court or opposing counsel, and recorded a two minute segment showing the man "in an apparently uncommunicative and partially catatonic state, responding to interrogation with grunts and growls." The Third Circuit held that "the ruling admitting the videotape was error," and that any benefit to the factfinding process "was far outweighed by the prejudice of admitting what amounted to ex parte testimony from the absent incompetent." Reversed, new trial on damages.

Bolstridge v. Central Maine Power Co., 621 F. Supp. 1202 (D. Me. 1985), excluded a day-in-the-life tape as cumulative, because the plaintiff could testify and demonstrate her activities in open court, and added the editing problem in a sentence you should keep: "Almost always an edited tape necessarily raises issues as to every sequence portrayed of whether the event shown is fairly representational of fact, after the editing process." The formulation both Bolstridge and Bannister carry forward came from Haley v. Byers Transportation Co., 414 S.W.2d 777 (Mo. 1967), where six reels of film showing home rehabilitation were excluded because "the very obvious impact of these films would have been to create a sympathy for the plaintiff out of proportion to the real relevancy of the evidence."

There is also a sequencing safeguard that experts almost never think about because it is a trial management decision, not an imaging decision. The district court in Grimes v. Employers Mutual Liability Insurance Co. of Wisconsin, 73 F.R.D. 607 (D. Alaska 1977), admitted a day-in-the-life motion picture. The opinion itself is hard to obtain, so this account comes from the later courts that quote it, and those courts emphasized that liability had been established before the jury saw the film, so the film could not prejudice the defendant on liability. The Thomas court read Grimes that way expressly and said it doubted the Grimes court would have reached the same result with liability still open.

Offer the image that makes the point, not the flinch

You will frequently have a choice between two accurate images of the same fact. One of them makes the jury flinch. The other makes the point. The flinch buys you nothing that the second image does not already buy, and it costs you two things: the risk of exclusion, and, if it comes in, a preserved appellate issue that survives the verdict you just won. A verdict reversed on an evidentiary ruling about your exhibit is a verdict you helped lose.

36.2Where two accurate images prove the same fact, offer the restrained one and keep the other in the file, produced and available.

36.3Never let volume do the work of proof. Rule 403 names "needlessly presenting cumulative evidence" for a reason, and forty photographs of one wound is the reason.

Then say it out loud. When counsel asks on direct why you selected the images you selected, the answer is that you chose the frames that show the condition and its measurements, that you had more graphic frames available, that you produced all of them, and that you did not offer the worst ones because they added nothing the jury needed to decide the case. That answer is true, it is checkable, and it converts restraint from a concession into a demonstration of method. Opposing counsel is left cross-examining you about your judgment in not being inflammatory.

Run the same rule against the other side's exhibits

The rule cuts the other way too, and you should be able to run it in both directions. When the other side offers an image whose only work is emotional, your report should say what fact it proves, what the less prejudicial version would have been, and what live testimony already covers the same ground. That is the Bolstridge argument, the Thomas argument, and the Haley argument in one paragraph, and it is far more persuasive than an objection that the picture is upsetting.

Chapter 36 Checklist: The Restrained Exhibit

Chapter 37Illustrative Aids Under Rule 107

Everything you have been calling a demonstrative exhibit for the last thirty years has a new name in federal court, and the name change came with rules attached.

What the rule says, and when it took effect

Federal Rule of Evidence 107 took effect December 1, 2024. It is the newest material in this book, it is the least likely thing on your shelf to be reflected in the treatise you learned from, and any chapter, checklist, or expert report written from pre-2024 sources is out of date on this point. Learn it now, because the first time you are asked about it under oath will not be the time to work it out.

The rule has four subsections. Subsection (a), on permitted uses, provides that "the court may allow a party to present an illustrative aid to help the trier of fact understand the evidence or argument if the aid's utility in assisting comprehension is not substantially outweighed by the danger of unfair prejudice, confusing the issues, misleading the jury, undue delay, or wasting time." Subsection (b), on use in jury deliberations, provides that "an illustrative aid is not evidence and must not be provided to the jury during deliberations unless: (1) all parties consent; or (2) the court, for good cause, orders otherwise." Subsection (c), on the record, provides that "when practicable, an illustrative aid used at trial must be entered into the record." Subsection (d) routes summaries of voluminous materials admitted as evidence to Federal Rule of Evidence 1006.

Notice that the balance in (a) is not the Rule 403 balance. It weighs the aid's utility in assisting comprehension, not its probative value, against the same list of dangers. Your exhibit is being judged on whether it helps the jury understand, and that is a different question from whether it proves anything.

The 2024 Committee Note supplies the definition: "An illustrative aid is any presentation offered not as evidence but rather to assist the trier of fact in understanding evidence or argument." On the jury room, the Note explains that letting an aid into deliberations "runs the risk that the jury may unduly emphasize the testimony of a witness with whom it was used, or otherwise misinterpret" its purpose, while trial courts retain discretion to permit access for good cause. On preservation, the Note is direct: "While an illustrative aid is not evidence, if it is used at trial it must be marked as an exhibit and made part of the record, unless that is impracticable under the circumstances."

Stop calling it a demonstrative in federal court

The Advisory Committee chose "illustrative aid" over "demonstrative evidence" because "demonstrative" had become ambiguous, used both for illustrative aids and for real or substantive exhibits such as a physical object. Federal vocabulary has shifted. Stop treating "demonstrative" as a term of art in federal court, and stop writing it into your reports as though it carried a legal meaning that everyone shares.

37.1Purge "demonstrative evidence" from your federal report template and replace it with either "illustrative aid under Federal Rule of Evidence 107" or "substantive evidence," chosen deliberately for each exhibit.

The older cases sort your exhibits into three categories

The rule is new but the reasoning behind it is not, and the pre-2024 cases still explain why the lines fall where they do. United States v. Bray, 139 F.3d 1104 (6th Cir. 1998), is the most useful single federal citation because it identifies three categories rather than two. Rule 1006 primary evidence summaries, which "must fairly represent and be taken from underlying documentary proof which is too voluminous for convenient in-court examination" and where "the summary, and not the underlying documents, is the evidence to be considered by the factfinder." Pedagogical devices under Federal Rule of Evidence 611(a), which summarize or illustrate admitted evidence, are not themselves admitted, may reflect the proponent's inferences and conclusions, and are "more akin to argument than evidence." And a middle category, the secondary evidence summary, "admitted in evidence not in lieu of the evidence they summarize but in addition thereto," about which the jury should be instructed that the summary "is not independent evidence of its subject matter, and is only as valid and reliable as the underlying evidence it summarizes."

That middle category is where a great deal of expert imagery lives. A photo board compiling admitted images, a synchronized multi-camera timeline, a frame grab comparison sheet, a scan-to-photograph overlay: none of these is a pure aid and none is a pure summary. Bray gives you the vocabulary to describe what you built and the instruction counsel should be asking for.

United States v. Janati, 374 F.3d 263 (4th Cir. 2004), adds the instruction duty, stating that whenever pedagogical charts are used "the court should make clear to the jury that the charts are not evidence themselves, but are displayed to assist the jury's understanding of the evidence." United States v. Wood, 943 F.2d 1048 (9th Cir. 1991), states that such devices "should be used only as a testimonial aid, and should not be admitted into evidence or otherwise be used by the jury during deliberations." And Pierce v. Ramsey Winch Co., 753 F.2d 416 (5th Cir. 1985), holds that pedagogical charts "are not themselves evidence and, absent the consent of all parties, they should not be sent to the juryroom." That carve-out, absent the consent of all parties, is the direct ancestor of Rule 107(b)(1).

What one exemplar in the jury room cost a party

In Baugh ex rel. Baugh v. Cuprum S.A. de C.V., No. 12-2019 (7th Cir. Sept. 13, 2013), a products liability case about a ladder, an exemplar ladder was used at trial by the defense expert to illustrate his oral testimony and was expressly treated as demonstrative, not evidence. Over objection, the district court sent it to the jury room. The jury returned a defense verdict shortly afterward. The Seventh Circuit held that "the general rule is that materials not admitted into evidence simply should not be sent to the jury for use in its deliberations," that "demonstrative exhibits that are not admitted into evidence should not go to the jury during deliberation, at least not without consent of all parties," and that sending the ladder back was an abuse of discretion. Finding "a reasonable possibility that the unadmitted materials altered the jury's verdict," and noting that the jury reached its verdict shortly after it had the chance to "examine, step on, and manipulate the exemplar ladder," the court reversed and remanded for a new trial.

An expert's exemplar went into the jury room and cost a party its verdict. Your animation, your scale model, your overlay, and your scan-derived exemplar are all in the same position.

How you put an animation into the record

Subsection (c) is the sleeper, and it is where this book's central argument pays off. An illustrative aid that is never admitted must still be marked as an exhibit and made part of the record. For a photograph on a board that is easy. For an animation, an interactive model, or a synchronized video timeline, "entered into the record" cannot sensibly mean a still frame, because an appellate court cannot evaluate what the jury saw without knowing how it was shown. Lodge the actual media file, and lodge with it a description of playback conditions: resolution, aspect ratio, frame rate, playback speed, screen size, and room lighting. Nothing in the rule spells that out. That is the point. Where the rule does not specify, your documented method is what fills the gap, exactly as it does in the absence of a governing imaging standard.

37.2For every dynamic aid, prepare a one page playback specification and lodge it with the media file, so the record shows what the jury saw.

37.3Ask counsel, before trial, whether the aid will be offered as evidence or used only illustratively, and build the labeling and the documentation to match that answer.

Two opposite design pressures, and you must pick one first

First, an aid that will not go to the jury room has to do its work while you are on the stand, which argues for simplicity, large elements, and one idea per frame. Second, an exhibit you want in the jury room has to be offered and admitted as evidence, which means it must clear authentication and Rule 403, and which means the assumptions embedded in it are all fair game. Those are opposite design pressures. Decide which one you are under before you build, not after.

Chapter 37 Checklist: Aids and Evidence

Chapter 38Animation, Simulation, and the Line Between Them

An animation is a picture of what you already concluded. A simulation is the thing that did the concluding.

That single distinction decides which body of law lands on your exhibit, and it decides how much of your work becomes discoverable. Get it right and you have an illustrative aid with a modest foundation. Get it wrong, or label it dishonestly, and you have offered substantive scientific proof without the record to support it.

Where the courts draw the line, in their own words

Take the definitions from the verified cases rather than from a vendor's brochure. The Supreme Court of Pennsylvania, in Commonwealth v. Serge, No. 150 MAP 2004, decided April 26, 2006, affirming the Superior Court decision reported at 837 A.2d 1255 (Pa. Super. 2003), described a computer generated animation as "a drawing... which, when assembled frame-by frame, produce[s] the image of motion," and as "merely a graphic representation depicting the previously formed opinion of a witness." On the other side of the line, "computer-generated simulations do not depict witness opinion; rather, the computer program... draws a conclusion... the end product of a simulation represents the computer program's conclusion." The trial court's jury instruction, quoted in the opinion, put simulation in plain language: "data is entered into a computer, which is preprogrammed to perform certain calculations by applying... the laws of physics, mathematical formulas... in order for the computer itself to draw conclusions."

People v. Cauley, 32 P.3d 602 (Colo. App. 2001), states the reason the line falls where it does, in one clause worth memorizing. Simulations are "recreations or experiments based upon scientific principles and data" in which "data is entered into a computer, which is programmed to analyze and draw conclusions from it," while "an animation is distinguishable from a simulation because it is demonstrative, rather than scientific, and the validity of the conclusion does not depend on proper application of scientific principles."

The validity of the conclusion does not depend on the software. That is the whole test. If your opinion would be identical had you drawn it on a whiteboard, you have an animation. If the software produced a number you are relying on, you have a simulation, whatever you call it in your report.

The two labels do not carry the same burden

The consequences are asymmetric and severe. An animation, in federal court after December 1, 2024, travels as an illustrative aid under Federal Rule of Evidence 107. Its foundation, as the verified state cases lay it out, is authentication as a fair and accurate representation of the testimony it illustrates, relevance, and a Rule 403 balance with attention to inflammatory content. A simulation is substantive proof, which pulls Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), down on your head, along with everything those authorities carry: testability, peer review, known or potential rate of error and the existence of standards controlling the technique's operation, and general acceptance, all of them non-exclusive factors under a "flexible" inquiry. Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999), removed any argument that a physics engine escapes because it is engineering rather than science: the gatekeeping obligation "applies not only to testimony based on 'scientific' knowledge, but also to testimony based on 'technical' and 'other specialized' knowledge."

38.1Decide the label by asking one question: would my opinion change if the software were removed? If yes, it is a simulation. Write that down before you build anything.

What a simulation foundation has to establish

The best verified statement of what a simulation foundation requires comes from a case about a chart, not a movie. Bray v. Bi-State Development Corp., 949 S.W.2d 93 (Mo. App. E.D. 1997), concerned a computer generated chart of light intensity levels, the output of an engineering calculation program. The court reasoned that the function of such programs "is to perform rapidly and accurately an extensive series of computations," and that "just as experts may base their testimony on calculations performed by hand, they may perform the same calculations using a computer." The three part foundation it adopted requires that the computer is functioning properly, that the input and underlying equations are sufficiently complete and accurate and disclosed to the opposing party, and that the program is generally accepted by the appropriate community of scientists. The expert there testified to the software's reliability, to industry reliance on it, and, most usefully, to verification against actual field measurements.

Read element two again. Disclosure of inputs and equations is written into the foundation itself. Read element three with a caveat: Missouri applied a general acceptance test, and in a Daubert jurisdiction general acceptance is one factor rather than the test. And note what none of the verified cases supplies: not one of them states a known or potential error rate requirement for a computer simulation in so many words. That factor applies through Rule 702, and the 2023 Committee Note reinforces it, but do not attribute an error rate holding to any of these decisions.

The generosity shown to animation comes with conditions

Now the animation side, where the federal courts have been generous and the generosity has conditions. Hinkle v. City of Clarksburg, 81 F.3d 416 (4th Cir. 1996), holds that "video taped evidence purporting to recreate events at issue must be substantially similar to the actual events to be admissible," moderating that requirement where the events themselves were disputed, and then draws the distinction that matters: "the difference between a jury believing that they are seeing a repeat of the actual event and a jury understanding that they are seeing an illustration of someone else's opinion of what happened." The instruction given told the jury the animation "is not meant to be a recreation of the events, but rather it consists of a computer picture to help you understand" the expert's opinion. The court added that there was no reason for the jury "to credit the illustration any more than they credit the underlying opinion," and encouraged trial judges "to first examine proposed videotaped simulation evidence outside the presence of the jury to assess its foundation, relevance, and potential for undue prejudice."

That sentence is your best argument for a pretrial hearing, in either direction.

Robinson v. Missouri Pacific Railroad, 16 F.3d 1083 (10th Cir. 1994), is worth knowing because the technique was physical rather than digital: the expert built a scale model of a nighttime grade crossing from physical evidence and photographs, then shot frame by frame video by incrementally moving the train and the car. Affirmed, given the exhibit's limited, solely illustrative purpose, with the trial court emphasizing to the jury that it was not a recreation. The animation logic is medium independent. Stop motion, computer graphics, and a hand drawn sequence are governed the same way.

Datskow v. Teledyne Continental Motors, 826 F. Supp. 677 (W.D.N.Y. 1993), contributes two details you will use. The animation illustrated an expert's theory of the origin and spread of an engine fire and was characterized as illustrative, not a recreation. The court ordered the tape played with the volume turned off, because radio communications on the soundtrack would lend a false sense of authenticity. That is a ruling about an incidental production choice creating verisimilitude nobody had earned, and it should govern your sound design, your motion blur, your lens flare, your smoke, and your skid noise. The second detail deserves skepticism rather than citation: the court reasoned that jurors are fairly sophisticated media consumers and that proper instruction mitigates overvaluation. That was a 1993 assumption about media literacy. Applying it without comment to a photorealistic exhibit rendered today is not an argument, it is a quotation.

Finding the simulation wearing an animation label

Attacking the other side's exhibit is the mirror image of building your own. Ask what the software computed. Ask whether any number in the expert's opinion originated inside the animation package rather than outside it. Ask whether the timing was chosen or derived. Ask for the project file, the inputs, the version, and the render settings. If the answer to any of those shows the program doing analytical work, the exhibit is a simulation wearing an animation label, and the foundation it received was the wrong one.

38.2Ask the opposing animator one question first: which numbers in this exhibit came out of the software rather than into it?

38.3Never let your animation depict a fact you did not measure or an assumption you did not disclose, because the exhibit will be read as though every frame is evidence.

Ask for the limiting instruction in your own case too

Ask also for the limiting instruction, in your own case as well as against. All four verified animation decisions involved one. Serge's cautionary instruction remains the most quotable sentence in the field: the animation "is a demonstrative exhibit, not substantive evidence," offered "solely as an illustration" of a party's version of events, and "you should not confuse art with reality."

Chapter 38 Checklist: Labeling the Exhibit

Chapter 39Print, Projection, and the Juror in the Back Row

Your exhibit is not the file. Your exhibit is what reaches the eye of the juror farthest from it.

The one number every resolution rule comes from

Everything here comes from one number, so derive it once and you will never look it up again. The standard reference for human acuity, 20/20 vision, is defined by a letter on an eye chart subtending a visual angle of 5 arcminutes, one arcminute being one sixtieth of a degree. By the design of the letters, the critical gap the eye must resolve is one fifth of that, or 1 arcminute. One arcminute at a viewing distance D subtends a linear size of D multiplied by the tangent of one sixtieth of a degree, which is D multiplied by 0.00029089, or D divided by 3438.

Put one pixel on each resolvable element and you have the whole rule. Required resolution in pixels per inch equals 3438 divided by the viewing distance in inches, which is 286.5 divided by the viewing distance in feet.

That formula explains every printing rule of thumb you have been given. At 12 inches it returns 286 pixels per inch, which is where "300 for close viewing" comes from. At 24 inches it returns 143, which is where "150 for large prints viewed at a distance" comes from. Those are not conventions. They are 20/20 acuity at one foot and at two feet. Many jurors see better than 20/20, which raises the requirement by about a third, and that is the honest reason to round 286 up rather than down.

39.1Compute the pixels per inch your exhibit needs from the distance of the closest realistic viewer, not from a rule of thumb and not from the average viewer.

What the board everyone wants would really cost

A 40 by 60 inch foam-core exhibit at 300 pixels per inch requires 12,000 by 18,000 pixels, which is 216 megapixels. No conventional single frame camera delivers that. At 150 pixels per inch it requires 6,000 by 9,000 pixels, or 54 megapixels, which is still beyond most working camera bodies. At a viewing distance of 6 feet the formula returns 48 pixels per inch, and the same board needs only 2,880 by 4,320 pixels, or 12.4 megapixels, which any modern camera produces.

So the board is fine, until a juror walks up to it. At 2 feet the requirement is 143 pixels per inch, which is 5,720 by 8,580 pixels, or 49 megapixels. Boards get approached. Design for the closest viewer or limit the approach, and understand that your three options are a lower resolution matched to a real viewing distance, a higher resolution or pixel shift body, or stitching. Stitching is a process, and a process gets disclosed and documented like every other process in this book.

For calibration: a 24 megapixel camera produces a 6,000 by 4,000 pixel file, which prints 24 by 36 inches at 167 pixels per inch. It cannot reach 300 at that size without interpolation, and interpolation is not new information. It is arithmetic that invents pixels.

The projector, not your file, sets the ceiling

The projector, not the file, is the ceiling. A high definition projector, meaning 1920 by 1080 pixels, delivers 2,073,600 pixels, which is 2.07 megapixels, no matter what you feed it. A 45 megapixel file is downsampled roughly 22 to 1 to get there. A 24 megapixel file is downsampled about 11.6 to 1.

It is worse than that, because your photograph does not fit the screen. A camera file is 3:2. A high definition screen is 16:9. Letterbox the full frame at full height and it occupies 1,620 by 1,080 pixels, which is 1.75 megapixels, or roughly eighty-four percent of the projector's pixels, with black bars down each side. On a legacy 1024 by 768 courtroom projector, a 3:2 image fits full width at 1,024 by 683 pixels, which is 0.70 megapixels. An ultra high definition projector delivers 3840 by 2160, or 8.29 megapixels, and many courtrooms do not have one.

39.2If a detail must be seen on a screen, prepare a cropped detail view as a separate, disclosed exhibit. Do not expect a juror to find it inside a full frame downsampled to two megapixels.

What fails in the back row is element size, not resolution

Pixels stop being visible when the image height in inches is at most 0.314 times the viewing distance in inches, which at 20 feet permits an image 75 inches tall, at 15 feet 56 inches, and at 10 feet 38 inches. At ordinary courtroom distances, high definition resolution is usually not what is failing you. Element size and contrast are.

So compute the elements. Minimum element height equals the viewing distance multiplied by the angle in arcminutes, divided by 3438. At 5 arcminutes, the recognition threshold, a character at 20 feet must be 0.35 inches tall to be recognized at all. Threshold is not a design target. A juror reading a dimension label for two seconds while listening to your testimony needs margin, and a defensible target is 15 to 20 arcminutes, or 1.05 to 1.40 inches of character height at 20 feet.

Work it backwards and the practical lesson appears. Juror at 20 feet, target 15 arcminutes, element height 1.05 inches. Measure the smallest text in your actual file as a fraction of image height, because that is a property of your exhibit and not a standard value. At two percent of image height, the required image height is 1.05 divided by 0.02, which is 52.5 inches, and on a 16:9 screen that is a 107 inch diagonal. At one percent it doubles to 105 inches, a 214 inch diagonal, which no courtroom has. The fix is not a bigger screen. The fix is a bigger element.

39.3Design exhibits so the smallest element occupies at least two percent of image height, and measure that fraction in your own file rather than assuming it.

Aspect ratios throw away evidence without telling you

Aspect ratios cost you evidence silently. The common ratios are 4:3 at 1.333, native to legacy projectors and most phone cameras; 3:2 at 1.500, native to every dedicated camera; 16:10 at 1.600; and 16:9 at 1.778. To make a 3:2 photograph fill a 16:9 frame you must crop the height from 4,000 pixels to 3,375, discarding 625 pixels, which is roughly sixteen percent of the image height. To make a 4:3 image fill 16:9 you discard twenty-five percent. That removed content is evidence you photographed and are now choosing not to show.

Letterbox the full frame. Black bars are ugly and forensically correct. If a crop is necessary for a detail view, do it on a working copy, preserve and produce the full frame alongside it, label the exhibit as a detail, document the crop coordinates, and be ready to put the full frame on the screen during cross. Print sizes carry the same trap: 24 by 36 and 40 by 60 are 3:2 and match a camera file exactly, 18 by 24 and 30 by 40 are 4:3 and force you to crop eleven percent off the width of a 6,000 by 4,000 file, and 11 by 17 matches nothing.

Nobody is going to turn the lights off for you

Courtrooms cannot be darkened, because the judge, the reporter, and the jury all have to read and take notes. Ambient light lands on the screen and raises the black level, which compresses the bottom of the tone curve, and shadow detail goes first. Published projector industry guidance puts the realistic floor for a lit room with a 100 to 150 inch image at about 4,000 lumens; a 2,000 lumen portable business projector washes out. A large emissive flat panel display, 75 to 98 inches, is more legible under courtroom light than any projector. Ask the courtroom deputy what is installed before you assume either.

Two audiovisual industry standards govern this territory and neither can be quoted here. The standard jointly issued by the American National Standards Institute and the Audiovisual and Integrated Experience Association, designated V201.01:2021, "Image System Contrast Ratio," defines four contrast ratios based on content viewing requirements, and those four values sit behind a paywall. Its companion, designated 202.01:2016, "Display Image Size for 2D Content in Audiovisual Systems," published May 2016, sets required display size and viewing positions for two stated viewing needs, Basic Decision Making and Analytical Decision Making, and its thresholds are paywalled as well. Do not print figures attributed to either document that you have not read in the document itself. The widely repeated rule placing the farthest viewer at four, six, or eight times the image height traces to no authoritative source at all. Treat it as trade folklore and use the acuity arithmetic above, which you can derive on the stand with a pencil.

That is the Chapter 4 argument in a different costume. Where the standard is unavailable, the documented method is what you offer, and first principles are more defensible than a number you cannot source.

Brightening an image for projection is an alteration

Brightening an image for projection is an alteration. So is a contrast adjustment, a gamma change, a crop, and a resize. Never adjust the original; work on a copy. Record the source filename and its hash, the software and version, the ordered list of operations, and the numeric parameter of each, in the form "black point 0 to 18, white point 255 to 255, gamma 1.00 to 1.35, no local adjustments, no cloning, no sharpening." Then apply the reproducibility test: could a comparably trained person, given your original and your notes, produce your exhibit? Bring the original to court. And never run content aware fill, generative fill, artificial intelligence upscaling, or artificial intelligence denoising on an evidentiary image, because those tools synthesize pixels that were never recorded.

39.4Disclose every display adjustment in the report, name the numeric parameters, and bring the unadjusted original to court.

Chapter 39 Checklist: The Back Row

Chapter 40Testifying About Your Own Images

Every photograph you have ever taken is an unanswered question until you answer it out loud.

Foundation is the first thing you can lose

Foundation is the first thing that happens and the fastest thing to lose. Under Federal Rule of Evidence 901(a), the proponent must produce evidence "sufficient to support a finding that the item is what the proponent claims it is." That is a low bar and it is not a technical one. If you took the photograph and you were there, you say what the scene was, when you were there, that you made the image, and that it fairly and accurately depicts what you observed. If nobody was there, as with a time lapse camera or an unattended recorder, the foundation comes from the process instead: how the device was set up, how it was activated, what interval it recorded at, what date it carried, and who held the medium afterward.

Answer those questions in the order the rule cares about, and answer them in short sentences. Foundation testimony is not where you show range.

How to explain a method to twelve strangers

Use the three step move that runs through this entire book: name the term, gloss it in the same sentence, and state the consequence. "Depth of field is the band of distance in front of the lens that comes out sharp, and I set mine so that the fracture surface and the scale beside it were both inside that band, which is why you can measure from this photograph." That sentence contains a technical concept, a plain language definition, and the reason a juror should care. Nobody has to look anything up.

Explain the instrument before you explain the result, because a juror who understands the tool will follow the number. And name the units every time. A dimension without a unit is a rumor.

40.1Rehearse every technical explanation aloud in under twenty-five words, and if it will not fit, you have not understood it well enough to testify about it.

Concede on purpose, and concede first

Conceding is the part experts get wrong, and it is the fastest credibility move available to you. The 2023 Advisory Committee Note to Federal Rule of Evidence 702, as amended effective December 1, 2023, is addressed to this audience by name. It states that forensic experts "should avoid assertions of absolute or one hundred percent certainty, or to a reasonable degree of scientific certainty, if the methodology is subjective and thus potentially subject to error," that the judge should where possible receive an estimate of the known or potential rate of error, and that opinion testimony about feature correspondence "must be limited to those inferences that can reasonably be drawn from a reliable application of the principles and methods." Photographic comparison, image authentication, and reverse projection are subjective or feature comparison methods sitting squarely inside that paragraph.

So concede on purpose, early, in your own words. A photograph shows the scene under stated conditions through a stated instrument, and it does not show what any witness perceived. A color rendering is as good as the reference target in the frame and no better. A measurement from an image carries the uncertainty of the pixel size, and there are variables you could not quantify at all. Say those things on direct, before the cross, and they stop being weapons.

The lens, the light, the clock, and the file

A competent opponent will ask them about every photograph you took, and they are always the same four: the lens, the light, the clock, and the file.

The lens question is about geometry. Expect "what focal length was that?" followed by "and doesn't a wide angle lens exaggerate distance?" The answer that holds names the focal length and the sensor format together, because one is meaningless without the other, states whether in-camera or in-software distortion correction was applied, and states what you did about it before measuring. In-camera and in-software correction silently change the projection model, so a photogrammetric solution built on the nominal focal length is wrong either way, and the expert who volunteers that before being led to it owns the exchange. If you chose a focal length to approximate human viewing, say why. If you chose a long lens knowing it compresses apparent distance, say that too, and say what the frame was for.

The light question is about honesty. Expect "was that how it looked?" The answer that holds describes the ambient conditions you metered on site, the lights you brought and where they were, whether flash was direct or off camera, whether a color reference target is in the frame, and what your exposure settings were. Then the concession that does more work than any of it: the photograph is evidence of the scene, not evidence of what a person could see, because a camera and a retina do not share a response curve, an integration time, or, at night, even a photoreceptor class.

The clock question is about trust. Expect "how do you know when that was taken?" The answer that holds is that you verified the camera clock against a known reference before the first frame, that you recorded the reference and the offset if there was one, that you noted the time zone and the daylight saving state, and that the frames run in a sequence consistent with your photograph log. The wrong answer is that the camera said so. A camera clock is a free running oscillator that somebody set once, and every opponent knows it.

40.2Verify and record the clock before the first frame of every job, because the clock is the cheapest thing to fix in advance and the most expensive thing to fix later.

The file question is about integrity. Expect "is that the original?" The answer that holds is that originals were offloaded by copying rather than by sharing, that a hash was recorded contemporaneously at acquisition, that all work was done on verified duplicates, that every processing step is documented well enough for a comparably trained person to reproduce it, and that nothing was deleted. If frames were culled, say how many, say why, and produce them. Ordinary handling strips image metadata, so if a file passed through email, a text thread, or a cloud library, say that too, and produce the untouched original beside it.

Where no standard exists, your written method is the standard

That last exchange is where this whole book pays off. In a field with no governing standard, Federal Rule of Evidence 702 does not ask whether a body has blessed your equipment. It asks, in subsection (c), whether the testimony is the product of reliable principles and methods, and in subsection (d), whether your opinion reflects a reliable application of those principles and methods to the facts of this case, with the proponent demonstrating both to the court by a preponderance of the evidence. A reliable application is a described application. Your documented method is not a substitute for the standard that does not exist. In this field, it is the standard, and you are the one who wrote it.

Which is why the answer to almost every hard question from the stand takes the same shape. Here is what I did. Here is why I did it that way. Here is what I wrote down at the time. Here is what my method cannot tell you.

The craft of testifying, the pauses, the pace, the management of a hostile examiner, the preparation sessions with counsel, belongs to a different book in this series, Being a Great Expert Witness. This one ends where the image reaches the courtroom.

Everything in these forty chapters reduces to a single instruction, and you have read it in a dozen forms. Decide what question the frame must answer. Capture it to the higher standard. Preserve the original. Write down what you did.

Do that, and the photograph will still be true eighteen months from now, in a room full of people who were not there.

Chapter 40 Checklist: The Stand

Appendices

Appendix A. The Master Field Checklist

Every checklist in this book, consolidated in the order a case runs. Five hundred ninety-eight items.

Print it, work it, and put the worked copy in the file. A checklist you filled out at the scene is a contemporaneous record. A checklist you filled out afterward is a memory, and a competent opponent will make you say which one it was.

Part One. The Frame

1. The Two Kinds of Image

2. The Instrument

3. Eye Versus Sensor

4. Standards and Authority

5. The Kit

Part Two. The Craft

6. Exposure

7. Focus and Depth

8. The Lens

9. Light

10. Beyond Visible Light

11. Color You Can Defend

12. Scale and Geometry

13. Format and Artifacts

Part Three. The Scene

14. Before You Leave

15. Overall, Midrange, Close-Up

16. Injuries and People

17. Parts, Machines, and Failure Surfaces

18. Difficult Scenes

19. Night and Visibility

20. Aerial Documentation

21. Three-Dimensional Capture

Part Four. The Record

22. Video Fundamentals

23. Rolling Shutter and Fields

24. High-Speed Capture

25. Surveillance Video Collection

26. Recording the Witness

27. Testing and Reenactment

Part Five. The Chain

28. The File's Own Testimony

29. The Second Chain

30. The Archive

31. Foundation

32. Authentication and Fakery

33. Disclosure

34. Preservation

35. Access and Permission

Part Six. The Verdict

36. The Restrained Exhibit

37. Aids and Evidence

38. Labeling the Exhibit

39. The Back Row

40. The Stand

Appendix B. Camera Settings Quick Reference

Ten scenarios, one page each. Every value below traces to a chapter. Where a chapter gives a verified published number, this appendix gives that number and names it as verified. Where no chapter supplies one, the entry is marked as a starting point, which means a defensible place to begin and a value you adjust on site and record in the log. A starting point is not a rule, and testifying to one as though it were a rule is how a good method gets impeached.

Two conventions run through all ten tables. The sensitivity setting is the control marked ISO on every camera. Raw means the camera's sensor-native raw file, confirmed lossless, and JPEG means the compressed format most cameras produce by default.

B.1 General Scene Documentation in Daylight

B.2 Interior Scene With Mixed Light

B.3 A Part or Fracture Surface on a Bench

B.4 Tool Marks and Impressions With Oblique Light

B.5 An Injury on Skin

B.6 A Large Dark Space Requiring Painting With Light

B.7 Night Visibility Documentation

B.8 Overlapping Capture for Photogrammetry

B.9 Video of a Walkthrough

B.10 High-Speed Video of an Impact Event

Appendix C. Equipment by Budget Tier

Buy the body last. That is how Chapter 5 ends and it is how this appendix begins, because the order in which you spend money is the only part of any equipment list that has ever mattered in a deposition. Nobody has ever asked me what body I shot with in a way that changed an answer. People ask about color and about size in every case, and the flat pieces of plastic that answer those two questions cost less than a decent zoom.

Read the tiers as a sequence rather than as a menu. The first tier is what you already own, and it is larger than most experts think, because a notebook, a pencil, a verified clock reference, and free hashing tools are equipment. The second tier buys the ability to state your settings. The third buys redundancy, which is the only thing that protects a scene that will not exist next month. The fourth buys specialization, and specialization is worth buying only when a recurring case type demands it.

Every price printed here comes from the chapters and is a July 2026 snapshot, several of them promotional. Bodies are superseded every two to four years. Lenses, tripods, flashes, scales, and color targets change far more slowly, and that is where your money should go. A $4,000 body operated on automatic with no record of what it did is weaker evidence than a $680 body operated in manual with a written procedure and a log. The expensive camera takes a better picture. The documented camera survives cross-examination.

One warning before the tables. Every item below answers the same question: does it make the image more defensible, or only prettier? Most photography money buys the second. Almost none of what follows does.

Tier 1. What You Already Own

The phone's limit is not resolution. A phone does not photograph the scene; it computes an image a trained model believes the scene should look like. Treat every phone image as a composite until you can prove otherwise, and never base a measurement or a color opinion on one.

Tier 2. Entry, Under $1,000

Tier 3. Working, $1,000 to $3,000

This is the book's default recommendation. The defining feature is not the sensor.

Tier 4. Full Capability, Above $3,000

What Not to Buy First

This is the useful part of the appendix, and it is short.

Do not buy the body first. The four things anyone will ever ask you about are the scale, the color target, the macro lens, and the second card slot. Buy those, then buy the body with what is left.

Do not buy a long lens for scene work. Anything longer than about 135 millimeters equivalent misrepresents spatial relationships, and a following vehicle photographed with a long lens looks dangerously close. That exhibit is answered from the same position with a normal lens, for free.

Do not buy an ultrawide lens because it fits more in. A 15 millimeter prime measured 6.12 percent barrel distortion uncorrected, and switching the camera's correction on moved its field of view by about 4 degrees and its effective focal length by a full millimeter, silently.

Do not buy a zoom for metrology. Use a prime, calibrate that specific body and lens at that specific focal length and focus distance, and keep the calibration date on the lid of the case.

Do not buy a shotgun microphone for indoor interviews. Published measurements of a modern shotgun show a directivity index flat at 6 decibels from 63 hertz through 1 kilohertz, which is what a plain hypercardioid delivers, and a small room's reverberant energy lives in exactly that region. Indoors, a shotgun is a boom pole holding a hypercardioid.

Do not buy a thermal camera expecting an infrared photograph. It records emitted long-wave radiation, not reflected light, its detector may be 320 by 240 pixels for the entire scene, and it reports a temperature only after you assume an emissivity and a background temperature.

Do not buy a phone-based or tablet-based scanner as your dimensional instrument. Use it to preserve a perishable scene when the alternative is nothing, say so in exactly those words, and do not let it carry a dimensional opinion a survey instrument would obviously have supported.

Do not buy a solid state drive or a shelved hard drive as your archive. Federal guidance gives expected shelf life for unused media as under 1 year for a solid state drive and under 2 years for a hard disk drive, neither recommended for archival.

Do not buy extra megapixels to fix an unreadable courtroom exhibit. A high-definition projector delivers 2,073,600 pixels regardless of what you feed it. The fix is a bigger element, not a bigger file.

Appendix D. Templates

Five forms, written to be copied or printed. Fill them in ink, on site, before you move the camera. A form completed afterward reads like a form completed afterward.

D.1 Photograph Log

Use this from frame one of every session, not from the frame where the case got interesting. It is the document you will be handed on the stand, and a gap in it invites a question you cannot answer well.

Case name: ______________________________________________ File number: ______________________

Date: ____________________ Location: ___________________________________________________

Photographer: ___________________________________________ Session start: ____________________

Camera body make and model: ______________________________ Body serial number: ______________

Clock reference photographed at start: ___________________ Reference time shown: ____________

Camera time recorded: ____________________________________ Signed offset: __________________

Time zone standing in: ___________________________________ Daylight saving state: ___________

Location tagging policy for this session: on / off, and reason: _________________________________

Hazards, protective equipment, and access method: _______________________________________________

Anything you could not photograph, and why: ____________________________________________________

Clock reference photographed at end: _____________ Reference time: _____________ Camera time: _____________

Session end: ____________________ Total frames captured: ____________________

Frames deleted in the field: none. If any, explain in full: ____________________________________

Signature: ______________________________________________ Date: ______________________________

D.2 Chain of Custody Record for Digital Image Files

Use one page per case, opened the day you first touch a card or receive a file, and treat it as an exhibit from that day forward. It records the second chain of custody, the one for the data, which exists in parallel with the chain for the physical card.

Case name: ______________________________________________ File number: ______________________

Media identifier: _______________________________________ Capacity: ________________________

Source: camera card / drive received from another party / export from a recorder / other: ______

If received from another party, from whom, in what packaging, and in what condition: ___________

Write blocker used on ingest: yes / no. Make, model, serial number: ___________________________

Ingest method: copy only. Files moved, cut, emailed, texted, or shared at any point: yes / no ___

Primary algorithm: the Secure Hash Algorithm at 256 bits. Second algorithm used: ______________

Original set designated and marked read-only on: ____________________ By: _____________________

Working copy created on: ____________________ Verified against original by hash: yes / no _____

Neural denoise, upscale, or other machine processing applied to any file: yes / no. If yes, list the file, the tool, the version, and the date: _________________________________________________

Copies held at, with medium named for each: 1) ________________ 2) ________________ 3) __________

Write-once or locked immutable copy held at: ___________________________________________________

Card retired to labeled sleeve on: ____________________ Card released for reuse by counsel on: ____________________

Verification schedule: manifest revalidated (quarterly) ____________ offsite restore tested (annually) ____________ Result logged: ____________

Irregularities encountered and what was done about them, recorded on the day they happened: _________________________________________________________________________________________________

Retention period obtained in writing from retaining counsel and carrier: ________________________

Signature: ______________________________________________ Date: ______________________________

D.3 Scene Documentation Plan

Fill this out before you leave the office, and carry it on paper to the scene so you can shoot from it. It is one page with four things on it, and the first one generates the other three.

Case name: ______________________________________________ File number: ______________________

Site: ___________________________________________________ Inspection date and time window: ____

1. The question the imagery must answer, in one sentence.

_________________________________________________________________________________________________

_________________________________________________________________________________________________

2. Shot list, built backward from that sentence. Mark each item O for overall, M for midrange, C for close-up, and S for any item requiring a scale.

Comparison items, adjacent conditions, and absences to photograph: ______________________________

Items you believe are irrelevant and will photograph anyway: ____________________________________

3. Equipment the shot list requires. Bodies ______ Lenses ______ Tripod ______ Flash and triggers ______ Scales ______ Color target ______ Gray card ______ Cards ______ Batteries charged ______ Write blocker ______ Laptop ______ Protective equipment ______ Second, disposable kit for a hostile scene ______ The plan itself, on paper ______

4. Constraints already known.

Access authority: order / stipulation / subpoena / owner consent. Cite it: ______________________

Restrictions on what may be photographed (faces, security systems, other tenants, proprietary processes): _____________________________________________________________________________________

Who may attend, who may operate equipment, may you measure, may you test, may you remove anything: _________________________________________________________________________________________________

Escort, site orientation, and protective equipment requirements: ________________________________

Time window, and what happens if it runs out: ___________________________________________________

Sunrise ________ Sunset ________ Sun position at the hour that matters ________ Moon phase ______

Weather forecast ________ Tide and current ________ Plant shutdown window ________

Protective order and retention terms confirmed in writing with counsel: yes / no. Date: _________

Location tagging policy for this inspection: ____________________________________________________

Second visit available: assume no. If a second visit is truly available, say why: _______________

Rough complete pass in all three tiers to be shot first, before any light is set up: yes

Prepared by: ____________________________________________ Date: ______________________________

D.4 Surveillance Video Collection Record

Use this on site, at the recorder, while you are still standing in the building. Every field on it captures something that is not recoverable after you leave.

Case name: ______________________________________________ File number: ______________________

Legal authority for the collection: ____________________________________________________________

Incident date, time, and location: _____________________________________________________________

Site address: ___________________________________________ Contact and title: _________________

Arrival time: ____________ Recorder located at: ______________________________________________

Photographs taken of the unit, the cameras, and all connections as found, before touching anything: frames ____________ to ____________

Clock. Photograph of the recorder's on-screen clock together with a known-good time source in one frame: frame ____________

Recorder clock displayed: ____________ Known-good time displayed: ____________

Signed offset, with direction stated in words: __________________________________________________

The time and date on the recorder were not changed. Confirmed: yes / no ________________________

The unit.

Make ____________ Model ____________ Serial number ____________ Firmware version ______________

Credentials used ____________________ Channels capable ________ Channels connected ___________

Storage capacity ____________ Used space ____________ Earliest recorded date ________________

Overwrite settings ____________________ Deletion or retention schedule ________________________

Per camera.

Export.

Native or proprietary export completed: yes / no. Format: ____________________________________

Viewer application and codec taken from the device: yes / no. Version: ________________________

Open-format export taken second, as a backup and not as a replacement: yes / no ________________

Time margin acquired on both sides of the incident window: from ____________ to ________________

Destination media identifier: __________________________________________________________________

Acquisition hash values recorded on this page: yes / no ________________________________________

Playback verified on a portable computer on scene, with dates and times displaying correctly: yes / no. Verified by: ____________________ Time: ____________

Encoder tag in the file metadata (a tag naming a general-purpose media library means the file was produced by software, not by the recorder): ____________________________________________________

System restored to its original state: yes / no Exit time: ____________

Chain of custody initiated on: ____________________ Signature: _______________________________

D.5 Consent to Record an Interview

Use this before any recorded interview, and read the announcement aloud on the recording itself so the consent exists inside the evidence rather than only on paper. The form is a starting point only, and retaining counsel must approve it for the jurisdiction before you use it, because consent law varies by state and this book is not legal advice.

Notice. This form is a starting point. Retaining counsel must review and approve it for every jurisdiction the conversation touches before it is used. A minority of states require the consent of every party, several prohibit secret or hidden-device recording rather than requiring consent, one requires notice rather than consent for in-person conversations, and at least one has no wiretapping statute at all. Apply the most restrictive rule among every jurisdiction the conversation touches, and confirm that rule with counsel in writing before the interview.

Case name: ______________________________________________ File number: ______________________

Date: ____________________ Start time: ____________________ End time: ______________________

Location of the interviewer: ___________________________________________________________________

Location of the person being recorded, as stated by that person: _______________________________

(Ask the witness where he or she is physically located, and record the answer. Area codes prove nothing.)

Every jurisdiction this conversation touches: __________________________________________________

Rule applied, and confirmed in writing by counsel on: __________________________________________

Statement to be read aloud on the recording, before any substantive question.

"My name is ______________________________. Today is ______________________. The time is ______________. I am recording this conversation, in audio and video, on two devices. This recording is being made in connection with ______________________________. Do I have your permission to record this conversation?"

Answer given, verbatim: ________________________________________________________________________

Recording device 1: make, model, media identifier: _____________________________________________

Recording device 2, independent power and independent media: ___________________________________

Sample rate 48 kilohertz, 24 bits or 32-bit float, uncompressed. Confirmed: yes / no ___________

Everyone present in the room or on the line, by name and role: _________________________________

Is this person represented by counsel in this matter, to your knowledge: yes / no / unknown ____

If yes, stop and call retaining counsel before proceeding: confirmed ___________________________

Consent of the person being recorded.

I have been told, before the recording began, that this conversation is being recorded in audio and video. I understand who is making the recording and in what matter. I understand that I may stop the recording at any time, may decline to answer any question, and may ask that any portion be stopped or that the recording be ended.

Printed name: ___________________________________________________________________________________

Signature: ______________________________________________ Date: ______________________________

Interviewer signature: __________________________________ Date: ______________________________

Retaining counsel approved this form for this jurisdiction on: ________________________________

Appendix E. All 145 Tips in One Table

Every numbered tip in the book, in order. The number is chapter and tip, so 12.3 is the third tip in Chapter 12. Read the chapter for the reasoning. Use the table to refresh before a site visit or a deposition.

Appendix F. Glossary

Every entry is written for a reader who has never held a camera, and every definition matches how the book uses the term.

Adaptation. The slow shift in the eye's sensitivity as light levels change. Full dark adaptation takes roughly forty minutes; bleaching fifty percent of the rod pigment raises the visual threshold by ten log units.

Airy disk. The blur spot a point of light becomes after passing through an aperture. Its diameter is 2.44 times the wavelength times the f-number, which is why small apertures cost resolution.

Alternate light source. A narrowband lamp that makes a subject fluoresce, viewed through a barrier filter that blocks the excitation light. Record the wavelength and the filter combination for every frame.

Animation. A moving graphic that depicts an opinion a witness already formed. If your opinion would be identical without the software, you have an animation, and it travels as an illustrative aid.

Aperture. The adjustable hole inside a lens, named by the f-number. Each step on the scale halves or doubles the light. Smaller number, bigger hole.

Arcminute. One sixtieth of a degree. The unit of visual angle behind both the 20/20 acuity standard and the perpendicularity tolerance on a forensic scale.

Aspect ratio. The proportion of width to height. Camera files are usually 3 by 2; screens are usually 16 by 9, so filling a screen with a photograph discards about sixteen percent of its height.

Authentication. Producing evidence sufficient to support a finding that an item is what you say it is. It is not a finding that the image is true, and it does not by itself get the exhibit admitted.

Available light. The light already present at a scene. The published trigger for available light technique is that some existing light is present and a single flash will not produce a proper exposure.

Banding. Visible steps in what should be a smooth gradient, produced by compression. It can read as the boundary of a shallow dent that is not there.

Barrel and pincushion distortion. Lens errors in which straight lines bow outward or inward. Barrel lives at the wide end, pincushion at the long end, and a mixture of both is called mustache distortion.

Base sensitivity. The lowest sensitivity setting the camera offers, where read noise is highest in electrons but the signal is largest. Use it whenever a tripod and time allow.

Bidirectional frame. A video frame encoded by interpolating between the nearest self-contained and predicted frames. It is instructions, not a picture, and cannot be verified in isolation.

Bit depth. How many tonal levels a file records per channel. Baseline JPEG is always 8 bits, or 256 levels. A 14-bit raw carries 16,384.

Bit rot. The quiet corruption of stored data over time. Consumer drives are specified at fewer than one unrecoverable read error per 12.5 terabytes read, so a full restore of a large archive is likely to hit one.

Blocking. Compression damage in which the 8 by 8 pixel blocks become visible. The tell is regularity, because real physical features do not align to a perfect lattice.

Bracketing. Shooting the metered exposure plus frames one and two stops under and over. It is not a hedge; it is a method, and you keep every frame including the ones you would never show.

Chain of custody. The documented record of who held an item, when, and what they did to it. In digital work there are two chains, one for the physical card and one for the file.

Chroma subsampling. Sharing one color measurement across a block of pixels. The common scheme discards three-quarters of the color measurements before compression even runs.

Circle of confusion. The blur diameter a point is allowed to reach before a viewer calls it unsharp. It is a viewing assumption, conventionally the frame diagonal divided by 1500, not a property of your camera.

Codec. The scheme that encodes and decodes video. An all-intra codec encodes every frame independently, which is what you want for anything you created.

Color filter array. The grid of colored filters over the sensor, green over half and red and blue over a quarter each. It is why roughly two-thirds of the color in any digital photograph is interpolated.

Color reference chart. A card of printed patches of known color, photographed in the subject's light and plane, that lets an opposing expert re-derive your color correction from your own image.

Color temperature. A one-number description of a light's color. A light that is not a blackbody radiator has only a correlated color temperature, and two lights sharing one can still look plainly different.

Content credential. A cryptographically signed manifest binding capture information and actions to a file. It is evidence of where a file came from, never evidence that the scene was real.

Contrast sensitivity. The smallest luminance difference the eye can detect. It peaks near half a percent in daylight and around eight percent at night, a factor of sixteen.

Crop factor. The ratio of the full frame diagonal to your sensor's. It tells you how a lens frames, not what focal length it is, and every calculation uses the real focal length.

Cross-polarization. A polarizing filter on each light and another on the lens, rotated until glare extinguishes. The single most useful trick in injury work, because it removes sheen and leaves color and texture.

Culling. Selecting which frames go into the report. Culling by selection is a discipline; culling by deletion is a preservation problem.

Day-in-the-life film. A recording of a plaintiff's daily activities offered to show the impact of an injury. It raises objections about staging, self-serving behavior, cumulativeness, and its dominating effect.

Deinterlacing. Turning interlaced fields into progressive frames. Every method destroys information, and motion-adaptive deinterlacing is the worst forensically because you cannot tell recorded pixels from invented ones.

Demosaicing. Reconstructing full color at every pixel from the single color each photosite measured. A raw file preserves the measured mosaic; a JPEG contains only the camera's one-time guess.

Depth of field. The range of subject distances that render acceptably sharp. It scales with the square of subject distance and inversely with the square of focal length, which is why macro work has millimeters of it.

Diffraction. The spreading of light as it passes an aperture, which softens the image as you stop down. It sets the practical aperture floor at f/16 on full frame, f/11 on crop, f/8 on Four Thirds.

Digital Negative. An archival container for raw data. It is not automatically lossless, and a file written by a neural denoising tool arrives in the same wrapper, so treat it as processed rather than original.

Distortion correction. Software or in-camera warping that straightens lens curvature. It is a geometric warp that resamples the image and changes the projection model, and many raw converters apply it on open.

Documentation image and examination quality image. The two published categories. Documentation images record appearance, condition, or location. Examination quality images are intended for scientific analysis and require a lossless format at the highest resolution.

Drop-frame timecode. A numbering scheme that skips certain frame numbers, not frames, so that elapsed timecode tracks real time on the 1000/1001 rates. Every recorded frame is still present in the file.

Duplicate. Under the evidence rules, a counterpart produced by a process that accurately reproduces the original. A forensically sound copy is a duplicate and is admissible to the same extent as the original.

Dynamic range. The span from the darkest usable signal to saturation, always measured against a chosen noise criterion. On one measured sensor the two published criteria give 12.37 stops and 9.00 stops.

Effective f-number. The working aperture at high magnification, equal to the marked f-number times one plus the magnification. At life size a marked f/16 is f/32.

Electronically stored information. The category the federal preservation rule reaches. Image files, raw files, video exports, and point clouds are all inside it; prints, negatives, and slides are not.

Emissivity. How efficiently a surface emits thermal radiation. A thermal camera converts radiance to temperature only by assuming it, which is why a shiny metal reading is confidently wrong.

Error level analysis. A resave-and-subtract technique that highlights regions with different compression histories. Its own author says a bright region may only mean an Adobe product touched the file. Keep it out of your report.

Exposure value. A single number combining aperture and shutter time at a reference sensitivity. Full sunlight is 15, offices 7 to 8, home interiors 5 to 7, night vehicle traffic 5.

Field. Half of an interlaced video frame, either the odd lines or the even lines, captured at a different instant from the other half. For timing analysis you want fields, not frames.

Focal length. The optical property that sets angle of view for a given sensor. The number printed on the barrel is nominal and is not a measurement, which is why metrology requires calibration.

Focus stacking. Merging frames shot at stepped focus distances into one image with extended depth. It is a computational composite: retain every source frame and label it as a stack in the report.

Foundation. The showing a proponent makes before an exhibit is received. For your own photographs it is percipient; for imagery nobody watched it is a description of the process that produced it.

Four cardinal points protocol. Photographing an impression with oblique light from twelve, three, six, and nine o'clock. It exists because a striation running along the light direction casts no shadow across itself.

Frame rate. How many frames a video records per second. Write it as a fraction, because the real rates are often 24000/1001, 30000/1001, and 60000/1001.

Global shutter. A sensor that exposes every row at the same instant. It produces no skew and no length error, and a flash synchronization claim at a very fast shutter speed is its diagnostic signature.

Gray card. A neutral card serving two separate purposes that practitioners conflate. Spectral neutrality is what matters for white balance; the 18 percent reflectance value is what matters for exposure.

Ground control point. A surveyed point of known position placed in a photogrammetric scene. A subset must be withheld from the solution and used only to measure residual error.

Ground sampling distance. The real-world size a single pixel covers. It is a statement of resolution, never of accuracy, and conflating the two is a cheap win for your opponent.

Group of pictures. The repeating structure of self-contained, predicted, and bidirectional frames in compressed video. Longer groups compress better and capture quick transitions less effectively.

Guide number. A flash's power expressed as f-number times distance. It means nothing without a stated sensitivity and a stated zoom head position, because narrowing the beam inflates it.

Hash. A fixed-length fingerprint computed from every bit of a file. Change one bit and the output changes completely. It is not encryption and you cannot recover the file from it.

High Efficiency Image File Format. A newer image format roughly twice as efficient as JPEG. Its damage does not show the tell-tale 8 by 8 grid; it smooths and substitutes detail, which looks clean.

Hyperfocal distance. The focus distance at which sharpness extends from half that distance out to infinity. A 35 millimeter lens at f/8 has a hyperfocal distance of 5.14 meters on full frame.

Illustrative aid. Under the new federal rule, any presentation offered not as evidence but to help the trier of fact understand evidence or argument. It stays out of the jury room absent consent or good cause.

Image metadata. The block of structured text a camera writes at capture, formally the Exchangeable Image File Format. It is unsigned, plaintext, writable by free software, and corroborative rather than conclusive.

Image stabilization. A system that counteracts camera shake, worth 2 to 5.5 stops. It does nothing at all about a subject that is moving.

Infrared imaging. Photography of reflected near infrared light, practically 700 to 900 nanometers. It carries no color data and should be converted to black and white after capture.

Interlacing. Recording alternate lines at alternate instants. A 1080i frame is not a frame; it is two fields captured 16.683 milliseconds apart at the common rate.

Intra frame. A video frame containing all newly encoded information, meaning a complete picture. Only these can be verified independently of the frames around them.

Inverse square law. Illumination falls as the square of distance from a point source. Doubling the distance costs exactly two stops, so a foreground at 1.5 meters and a background at 6 meters differ by four.

JPEG. The compressed format most cameras produce by default. The white balance is baked into the pixel values, the bit depth is 8, and the camera's interpretation cannot be re-made.

Litigation hold. Counsel's instruction to preserve material once litigation is anticipated. It reaches your cards, working drives, offsite copies, synced accounts, and email attachments.

Location tagging. Coordinates and sometimes a heading written into the file, describing where the photographer stood, not where the subject is. It carries a second, satellite-derived clock you can check the camera's against.

Lossless and lossy compression. Lossless compression discards nothing. Lossy compression discards data permanently and may be acceptable for documentation images, but is not recommended for images that will be analyzed or compared.

Luminance. The brightness of a surface as seen from a viewpoint, measured in candelas per square meter. Cone vision carries above about 0.03 of those units; below it rods take over.

Macro lens. A lens that reaches life-size magnification, meaning the image on the sensor is as large as the object. Buy one by its working distance at maximum magnification, measured by you.

Microphone type and pattern. Condenser capsules are sensitive with extended high-frequency response; dynamic capsules are tougher and less sensitive. Directional patterns buy working distance: an omnidirectional scores 1.00, a hypercardioid 2.00.

Mosquito noise. Shimmering compression artifacts around edges. It is one of the artifacts the published compression guidance does not name, which is a gap worth stating in your report.

Motion blur. Smear from subject or camera movement during the exposure. Blur on the sensor equals subject speed times exposure time times magnification, and the numbers are worse than intuition suggests.

Native file. The file a recorder or camera wrote, in its own format, with its own timing and overlay. It sits at the top of the quality hierarchy, above an export and far above a screen recording.

Noise. Randomness in the recorded signal. Read noise falls as you raise the sensitivity setting; photon shot noise rises as the square root of the signal and is cured only by more light.

Normal lens. Conventionally a lens whose focal length equals the frame diagonal, 43.3 millimeters on full frame. It reproduces the geometry of central, attentive looking, not the field of human vision.

Oblique light. Light placed low enough to cast a shadow into the relief of a surface. The published standard states that criterion and prescribes no angle, because the correct angle depends on the depth of the relief.

Original and working copy. The original is an accurate and complete replica of what the camera wrote. Everything you process is a working copy, and processed output is derivative evidence with its own integrity requirements.

Overlap. The fraction of content shared between adjacent frames. Roughly a third for scene coverage; at least 75 percent frontal and 60 percent side for photogrammetry, rising to 90 percent on large vertical objects.

Painting with light. Building one long exposure out of many flash pops fired from inside a large dark scene, with the lens covered between pops so your movement never records.

Perspective compression. The flattening of apparent depth when you photograph from far away with a long lens. Perspective is caused by distance, not by the lens, but focal length dictates where you must stand.

Photogrammetry. Deriving measurements from photographs. Close-range vehicle crush measurement has been reported at 0.1 centimeters of difference from a total station baseline, at least as good as a tape measure in a trained hand.

Photo-response non-uniformity. The sensor fingerprint left by manufacturing variation in silicon. It has real scientific standing, needs hundreds of images to build a reference, and degrades under compression and cropping.

Photopic and scotopic vision. Daylight vision carried by cones and night vision carried by rods. Scotopic vision is achromatic, resolves about a tenth as finely, and is sharpest 5 to 15 degrees off the line of sight.

Pictorial testimony theory. Authentication through a sponsoring witness with personal knowledge who testifies that the photograph fairly and accurately portrays the scene. Your own photographs travel this way.

Pixel pitch. The center-to-center spacing of photosites. Light gathering scales with its square, which is why a 6.00 micrometer full frame pixel gathers roughly 24 times the light of a 1.22 micrometer phone pixel.

Point cloud. The set of three-dimensional points a scanner or a photogrammetric solution produces. Archive it in the open interchange format specified by ASTM International standard E2807.

Preimage resistance and collision resistance. Collision resistance asks whether an attacker can find two files of his own choosing with the same hash. Preimage resistance asks whether he can build a file matching a hash you already recorded. Evidence integrity rests on the second.

Presentation timestamp. The per-frame time value stored throughout a video container. Compute the difference between consecutive timestamps rather than trusting a nominal rate.

Quantization table. The divisor grid a compressor uses to decide what detail to discard. It is a written record of what the camera threw away, and every camera uses its own.

Raw file. A file preserving the sensor's measured mosaic with the camera's interpretation choices stored as metadata. White balance stays a reversible parameter, which supports the testimony that you changed nothing.

Readout time. How long a rolling-shutter sensor takes to scan from top to bottom. Published oscilloscope measurements range from 3.7 milliseconds on a stacked body to about 71 milliseconds in a silent full-frame mode.

Reciprocal rule. Use a shutter speed numerically closest to the 35 millimeter equivalent focal length when handholding. It was derived for small prints, so double it for files that will be enlarged.

Registration. Aligning multiple scanner setups into one coordinate frame. It is a separate error budget that compounds across setups, and its statistics belong in the report.

Retroreflective material. Tape and sheeting that returns light toward its source. It ages measurably, so an exhibit shot against fresh tape is not an exhibit of the trailer as it existed.

Reverse projection photogrammetry. Re-photographing a scene with a reference of known height placed at the subject's position. It is not the same as scaling within a single image from objects of known size.

Ringing. Faint light and dark bands parallel to a high-contrast edge, produced when compression drives high-frequency coefficients to zero. At magnification it is indistinguishable from a hairline crack.

Rod and cone. The two photoreceptor classes. Rods carry night vision, peak near 507 nanometers, integrate over about 100 milliseconds, and see no color; cones carry daylight vision and integrate over about 10 to 15 milliseconds.

Rolling shutter. A sensor that reads out one row at a time from top to bottom, so every row in a frame has a different exposure start time. Vertical edges lean, and lengths along the scan direction stretch or compress.

Sample aspect ratio. The shape of the stored pixel. Legacy surveillance formats do not use square pixels, and displaying a 720 by 480 frame naively makes it 11.11 percent too wide, horizontally only.

Scale. A rigid reference of known dimension placed in the plane of the evidence. The standard forensic pattern carries graduated legs, a tonal reference, and three circles used to detect and correct off-axis perspective.

Scrub. A scheduled pass in which a checksumming file system reads all stored data and repairs what it can from a replica. Recommended monthly or more often.

Sensor plane. The plane of the imaging chip. Held parallel to the subject plane, measurement is direct; tilted by 20 degrees, dimensions along the tilt read 6.0 percent short and shapes change.

Shutter angle. A cinema convention expressing exposure time as a fraction of the frame interval. The 180 degree convention gives one over twice the frame rate, and it is an aesthetic standard, not a measurement standard.

Silent witness theory. Authentication of an image as substantive evidence through the imaging system itself, with no human who perceived the event. This is how surveillance video travels.

Simulation. A computation that draws its own conclusion from entered data. It is substantive proof and pulls the reliability rule and its case law onto your exhibit, whatever you called it in the report.

Skew. The lean of vertical edges in a rolling-shutter frame of a moving subject. The object's own size cancels out, so any known vertical in the scene lets you solve for readout time or speed.

Specular reflection. A mirror image of the light source on a glossy, wet, chromed, or oily surface. Those pixels clip to maximum on every channel and hold nothing recoverable.

Spoliation. The destruction or material alteration of evidence, or the failure to preserve it for another's use, in pending or reasonably foreseeable litigation. Thorough documentation is not a defense to it.

Stop. A factor of two in light. Aperture steps, shutter steps, and sensitivity steps are all interchangeable in this currency, which is what makes exposure arithmetic rather than guesswork.

Structure from motion. Recovering camera positions and a three-dimensional point set from overlapping photographs, by detecting features, matching them, solving poses in a bundle adjustment, and fusing depth.

Substantially similar conditions. The foundational showing required before a staged test or demonstration is admitted. The label you attach does not control; what controls is whether the demonstration looks close enough to the event to mislead.

Sunny 16. A field exposure rule needing no batteries: in direct sunlight, set f/16 and a shutter speed equal to the reciprocal of the sensitivity. It computes to exposure value 15, which is the published value for full sun.

Tagged Image File Format. An uncompressed image format. It is the only preferred format the national archives lists for born-digital photographs, and it is what a person can still open in twenty years without a manufacturer's decoder.

Terrestrial laser scanning. Sweeping a beam over a scene and recording range and angle for millions of points. Read the specification with its conditions: reflectance, confidence level, and single-shot measurement.

Thermal imaging. Recording long-wave radiation a subject emits, typically 7.5 to 13 micrometers. It is a different technology from infrared photography, glass is opaque to it, and it is not a temperature measurement without more work.

Three-two-one rule. Three copies, on two kinds of media, with one offsite. Its citation trail is weak, so defend it from published failure rates rather than from authority.

Timecode. A frame-accurate time label carried with video. Genlock locks the rate; jam sync only sets the value, so a jam-synced pair holds the same number and drifts apart all afternoon.

Ultraviolet imaging. Photography of reflected ultraviolet radiation, in practice at 365 or 254 nanometers. Wavelengths below 280 nanometers denature deoxyribonucleic acid rapidly during exposure, which reorders your evidence handling.

Variable frame rate. A recording whose frames are not evenly spaced in time. Surveillance recorders shift rates on motion triggers, so counting frames and dividing by a nominal rate can be wrong by an enormous factor.

White balance. Three gain numbers applied to the red, green, and blue channels to make a scene look neutral. Three numbers cannot invert a spectrum, so white balance neutralizes a cast and does not restore missing reflectance.

Working distance. The clear space between the front of the lens and the subject, which is not the minimum focus distance printed on the specification sheet. It decides whether you can get a light in beside the subject.

Write blocker. A hardware device that physically prevents any modifying operation from reaching a protected storage device. Unlike a memory card's sliding tab, it does not depend on anyone's good behavior.

Write-once media. Storage that cannot be rewritten, whether a recordable optical disc or a cloud object under a compliance-mode lock. It turns "I did not alter the file" from testimony into a property of the system.

Appendix G. Standards, Rules, and Sources

Everything listed here appears in the chapters of this book. Nothing has been added from any other source. Where a chapter gave a citation in a particular form, that form is reproduced.

Federal Rules of Evidence Cited

• Federal Rule of Evidence 107, illustrative aids, effective December 1, 2024, subsections (a),

(b), (c), and (d), with the 2024 Committee Note.

• Federal Rule of Evidence 403, excluding relevant evidence for prejudice, confusion, waste of

time, or other reasons, enacted 1975 and restyled 2011.

• Federal Rule of Evidence 611(a), the source of the pedagogical device practice described in

United States v. Bray.

• Federal Rule of Evidence 702, testimony by expert witnesses, as amended effective December 1,

2023, subsections (c) and (d), with the 2023 Advisory Committee Note.

• Federal Rule of Evidence 901(a), the general authentication requirement.
• Federal Rule of Evidence 901(b)(1), testimony of a witness with knowledge.
• Federal Rule of Evidence 901(b)(4), distinctive characteristics and the like.
• Federal Rule of Evidence 901(b)(9), evidence about a process or system, with its Advisory

Committee Note.

• Federal Rule of Evidence 902(11) and 902(12), the certification requirements that 902(13) and

902(14) incorporate.

• Federal Rule of Evidence 902(13), records generated by an electronic process or system,

effective December 1, 2017.

• Federal Rule of Evidence 902(14), data copied from an electronic device, storage medium, or

file, with its Advisory Committee Note and that Note's limiting sentence.

• Federal Rule of Evidence 1001(c), defining a photograph as a photographic image or its

equivalent stored in any form.

• Federal Rule of Evidence 1001(e), defining a duplicate.
• Federal Rule of Evidence 1003, admissibility of duplicates.
• Federal Rule of Evidence 1006, summaries to prove content.

Federal Rules of Civil Procedure Cited

• Federal Rule of Civil Procedure 26(a)(2)(B), the written report requirement, including

subparagraphs (i), (ii), and (iii), with the 2010 Advisory Committee Note.

• Federal Rule of Civil Procedure 26(b)(1), relevance and proportionality.
• Federal Rule of Civil Procedure 26(b)(3)(A) and (B), work product, as incorporated by

26(b)(4)(B).

• Federal Rule of Civil Procedure 26(b)(4)(B), draft report protection.
• Federal Rule of Civil Procedure 26(b)(4)(C), attorney and expert communications, with its three

carve-outs.

• Federal Rule of Civil Procedure 26(b)(4)(D), the non-testifying consultant.
• Federal Rule of Civil Procedure 26(b)(4)(E), fees for time spent responding to discovery.
• Federal Rule of Civil Procedure 26(c)(1) and 26(c)(1)(E), protective orders and designating who

may be present.

• Federal Rule of Civil Procedure 34(a)(2), entry onto land for inspection, which expressly

enumerates photographing.

• Federal Rule of Civil Procedure 34(c), the bridge to non-party inspection.
• Federal Rule of Civil Procedure 37(e), failure to preserve electronically stored information,

subsections (1) and (2), with the 2015 Advisory Committee Note.

• Federal Rule of Civil Procedure 45, non-party subpoenas, including 45(a)(1)(A)(iii) and

45(d)(2)(A), with the power dating to the 1991 amendments.

Federal Rule of Criminal Procedure Cited

• Federal Rule of Criminal Procedure 53, cited in Chapter 35 only to correct its misuse: it is a

criminal rule barring photographing in the courtroom during judicial proceedings, and it does not by its own force govern civil proceedings, hallways, lobbies, or the building exterior.

Cases Cited, Alphabetically

• Bannister v. Town of Noble, Oklahoma, 812 F.2d 1265 (10th Cir. 1987)
• Barnes v. General Motors Corp., 547 F.2d 275 (5th Cir. 1977)
• Baugh ex rel. Baugh v. Cuprum S.A. de C.V., No. 12-2019 (7th Cir. Sept. 13, 2013)
• Becker v. Computer Sciences Corp., 541 F. Supp. 694 (S.D. Tex. 1982)
• Belcher v. Bassett Furniture Industries, Inc., 588 F.2d 904 (4th Cir. 1978)
• Bergner v. State, 397 N.E.2d 1012 (Ind. Ct. App. 1979)
• Bolstridge v. Central Maine Power Co., 621 F. Supp. 1202 (D. Me. 1985)
• Branzburg v. Hayes, 408 U.S. 665 (1972)
• Bray v. Bi-State Development Corp., 949 S.W.2d 93 (Mo. App. E.D. 1997)
• California v. Ciraolo, 476 U.S. 207 (1986)
• Chandler v. Florida, 449 U.S. 560 (1981)
• Chapman ex rel. Estate of Chapman v. Bernard's, Inc., 167 F. Supp. 2d 406 (D. Mass. 2001)
• Commonwealth v. Serge, No. 150 MAP 2004, decided April 26, 2006, affirming the Superior Court

decision reported at 837 A.2d 1255 (Pa. Super. 2003)

• Datskow v. Teledyne Continental Motors, 826 F. Supp. 677 (W.D.N.Y. 1993)
• Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993)
• Dow Chemical Co. v. United States, 476 U.S. 227 (1986)
• Florida v. Riley, 488 U.S. 445 (1989)
• Food Lion, Inc. v. Capital Cities/ABC, Inc., 194 F.3d 505 (4th Cir. 1999)
• Foster v. Crawford Shipping Co., 496 F.2d 788 (3d Cir. 1974)
• Four Corners Helicopters, Inc. v. Turbomeca, S.A., 979 F.2d 1434 (10th Cir. 1992)
• Fusco v. General Motors Corp., 11 F.3d 259 (1st Cir. 1993)
• General Electric Co. v. Joiner, 522 U.S. 136 (1997)
• Gladhill v. General Motors Corp., 743 F.2d 1049 (4th Cir. 1984)
• Griffin v. State, decided by Maryland's high court on April 28, 2011
• Grimes v. Employers Mutual Liability Insurance Co. of Wisconsin, 73 F.R.D. 607 (D. Alaska 1977)
• Haley v. Byers Transportation Co., 414 S.W.2d 777 (Mo. 1967)
• Hall v. General Motors Corp., 647 F.2d 175 (D.C. Cir. 1980)
• Hinkle v. City of Clarksburg, 81 F.3d 416 (4th Cir. 1996)
• Houchins v. KQED, Inc., 438 U.S. 1 (1978)
• Jackson v. Fletcher, 647 F.2d 1020 (10th Cir. 1981)
• Jenkins v. Associated Transport, Inc., 330 F.2d 706 (6th Cir. 1964)
• Kearney v. Salomon Smith Barney, Inc., 39 Cal. 4th 95 (2006)
• Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999)
• Kyllo v. United States, 533 U.S. 27 (2001)
• Long Lake Township v. Maxon, decided by the Michigan Supreme Court on May 3, 2024
• Lorraine v. Markel American Insurance Co., 241 F.R.D. 534 (D. Md. 2007)
• National Press Photographers Association v. McCraw, 90 F.4th 770 (5th Cir. 2024)
• New York State Association for Retarded Children, Inc. v. Carey, 706 F.2d 956 (2d Cir. 1983)
• Pell v. Procunier, 417 U.S. 817 (1974)
• People v. Cauley, 32 P.3d 602 (Colo. App. 2001)
• Pierce v. Ramsey Winch Co., 753 F.2d 416 (5th Cir. 1985)
• Project Veritas v. Schmidt, No. 22-35271 (9th Cir. Jan. 7, 2025) (en banc)
• Randall v. Warnaco, Inc., 677 F.2d 1226 (8th Cir. 1982)
• Residential Funding Corp. v. DeGeorge Financial Corp., 306 F.3d 99 (2d Cir. 2002), named in the

2015 Advisory Committee Note as a line of cases the rule rejects

• Robinson v. Missouri Pacific Railroad, 16 F.3d 1083 (10th Cir. 1994)
• Sanchez v. Denver & Rio Grande Western Railroad Co., 538 F.2d 304 (10th Cir. 1976)
• Silvestri v. General Motors Corp., 271 F.3d 583 (4th Cir. 2001)
• State v. Matthews, 479 Md. 278, 277 A.3d 991 (2022)
• Swajian v. General Motors Corp., 916 F.2d 31 (1st Cir. 1990)
• Thomas v. C.G. Tate Construction Co., 465 F. Supp. 566 (D.S.C. 1979)
• United States v. Bray, 139 F.3d 1104 (6th Cir. 1998)
• United States v. Causby, 328 U.S. 256 (1946)
• United States v. Janati, 374 F.3d 263 (4th Cir. 2004)
• United States v. Johnson, 114 F.3d 808 (8th Cir. 1997)
• United States v. Kyler, 429 F. App'x 828 (11th Cir. 2011)
• United States v. Quinn, 18 F.3d 1461 (9th Cir. 1994)
• United States v. Rembert, 863 F.2d 1023 (D.C. Cir. 1988)
• United States v. Stearns, 550 F.2d 1167 (9th Cir. 1977)
• United States v. Taylor, 530 F.2d 639 (5th Cir. 1976)
• United States v. Wood, 943 F.2d 1048 (9th Cir. 1991)

Standards Bodies and Their Documents

Scientific Working Group on Digital Evidence. The live body in this field. It publishes free versioned documents and maintains photography, imaging, and video committees. Documents named in this book:

• Image Processing Guidelines, document 15-M-002
• Guidelines for Low Light Crime Scene Photography
• Lighting Techniques
• Guidelines for the Use of Reflected Ultraviolet Radiation in Forensic Photography, document

19-P-001

• The companion reflected infrared document, 19-P-002
• Guidelines for Digital Imaging of Footwear and Tire Impressions, document 17-P-003, version 2.0
• Image categories in forensic science, 2025
• Best practices for maintaining the integrity of imagery, current as of March 3, 2025
• Best practices for image authentication, version 2.0, March 3, 2025
• Compression and file format guidelines
• Best Practices for the Forensic Use of Photogrammetry, version 1.2
• Technical Overview of Digital Video Files, document 17-V-001
• Best Practices for Data Acquisition from Digital Video Recorders, document 17-V-002
• Considerations for the Use of Time-Based Analysis of Digital Video for Court, document 19-V-003
• Best Practice for Frame Timing Analysis of Video Stored in ISO Base Media File Formats, document

19-V-005, version 1.1

• Core Technical Concepts for Time-Based Analysis of Digital Video Files
• Position on hashing algorithms, September 29, 2019
• Collection best practice, published November 20, 2025
• Committee guidance on technical aspects of digital cameras, photographic equipment

recommendations, macro photography, evidence photography in a controlled setting, and forensic image analysis

Scientific Working Group on Imaging Technology. Defunct. Cite it for foundational principles only, with its date attached, and note that its old web address no longer belongs to it.

• Section 1, overview, version 3.3, dated June 11, 2010
• Section 11, documenting image enhancement, and its four operative principles
• Impression photography guidance, the source of the overall, midrange, and close-up sequence
• Guidance listing cropping alongside brightness and contrast adjustment as image processing

Organization of Scientific Area Committees for Forensic Science. Administered by the National Institute of Standards and Technology, created in 2014, with a Forensic Science Standards Board, a Standards Review Panel, Scientific Area Committees, subcommittees, and cross-cutting resource task groups, more than 800 volunteer members and affiliates across 19 forensic disciplines. It does not publish standards. It drafts, pushes drafts to accredited standards developing organizations, and maintains a registry of selected published and proposed standards.