Day 4 of 1060 minShell WalkerHands-on lab

Day 4: Image Triage Case: Review a Synthetic Identity

Taught by Mr Elite Updated

By the end of today you can

  • Build a short triage checklist that combines hashing, metadata and artefact checks
  • Run the checklist on a suspect image and record each result
  • Review a synthetic identity application and weigh evidence by strength
  • Write a one-page triage note another analyst can repeat

Warm-up

Answer from memory first, then tap to check. Recalling yesterday is what makes it stick.

On Day 3 we rated every image signal. Name the four ratings and say why a detector score alone is not enough.

Strong, moderate, weak or none. A detector score is a classifier guess, so it must be recorded and backed by other evidence.

Why can Error Level Analysis (ELA) raise a false alarm?

Compression, resaving and sharp edges create uneven error levels even in untouched images.

A bank receives a new account application. The photo ID looks clean, the automatic checks passed, and the applicant is waiting. Someone has to decide whether this goes through, gets escalated or gets stopped. Gut feeling is not an answer you can defend.

Today you stop running Day 1 to 3 tools one at a time and join them into a single workflow. The point is not a cleverer tool. It is a routine you can repeat on any suspect image, and write down so a colleague gets the same result.

Review: three methods, three different questions

Each method we have learnt answers a different question. Mixing them up is how analysts overclaim. Let me restate them plainly before we combine them.

Hash
A fixed fingerprint of the file's bytes. It proves the file has not changed since you recorded it. It says nothing about whether the picture is real.
Metadata
Data stored inside the file, such as camera model, software and dates. Easy to edit or strip, so it is a lead, not proof.
Artefact analysis
Looking at pixels and structure for generation or editing traces: odd hands, text, ELA, noise, trailing data.

Think of them as three witnesses who each saw something different. The hash witness can only say whether the file is the same one you started with. The metadata witness can tell you what the file claims about itself, and a file can lie. The artefact witness describes what the picture itself looks like, but can be fooled by compression and lighting.

No single witness settles the case. A generated image can have a perfect hash, believable metadata that someone typed in, and clean pixels. A real photo can have no metadata, a odd-looking ELA map and a strange shadow. That is why today is about combining them and weighing each one honestly.

Quick check

A file has no camera metadata at all. What does that tell you?

Build the triage checklist

A checklist keeps you honest. The order matters: preserve first, then observe, then judge. Hashing comes first because every later step should be tied to a file you can prove you did not alter.

For every signal, ask two questions. How strong is it, and what innocent explanation could produce it? A blurry earring might be a generator mistake, or it might be a low-resolution upload. If you can easily think of an innocent cause, the rating drops. If the signal is hard to explain innocently and another independent signal agrees, it rises.

Watch me do it: triage one image

I will run the checklist on a file called suspect.jpg. Use any image you own or one from your Day 3 set. Everything here runs on your own machine, and nothing is sent anywhere. Open a terminal in the folder that holds the image before you start.

bash
mkdir -p case04 && cp suspect.jpg case04/ && cd case04
sha256sum suspect.jpg | tee hash.txt
exiftool suspect.jpg > exif.txt
file suspect.jpg
strings -n 8 suspect.jpg | tail -5
Preserve, hash, read metadata, confirm the type, and peek for readable text near the end of the file. Output is saved to hash.txt and exif.txt for your note.

If sha256sum is not found on macOS, use shasum -a 256 instead. The tee command prints the hash on screen and saves it at the same time. The greater-than sign sends ExifTool output into exif.txt, so the terminal stays quiet for that line. That is expected, not a failure.

text
3f9a...c21e  suspect.jpg
suspect.jpg: JPEG image data, JFIF standard 1.01, 1024x1024
Example of what you might see. Your hash will differ. A square size like 1024x1024 is common for generated images but is also common for cropped profile photos, so note it without over-reading it.

After these steps you should have two saved files, hash.txt and exif.txt, and a short list of observations. Open exif.txt with cat exif.txt or any text editor and read it slowly. That is already a repeatable record, and the next section shows how to write it down properly.

Quick check

Why do we save the hash before running any other tool?

Recording findings so someone else can repeat them

A finding another analyst cannot repeat is only an opinion. For every check, write the command or action, the tool version if it matters, what you saw, and the rating. Keep wording factual: say what you observed, not what you assume.

text
Check: ExifTool metadata
Command: exiftool suspect.jpg
Seen: no Make or Model; Software absent
Rating: weak (often stripped)
Check: Visual, hands/ears
Seen: left earring missing on right ear
Rating: moderate
A simple log format. One block per check, with an honest rating.

Notice the difference between the two wordings. Seen: no Make or Model is an observation anyone can confirm. This image was stripped to hide its origin is an assumption, and it belongs, if anywhere, in your reasoning, clearly labelled as such.

At the end of the log, add three short lines: overall verdict, confidence (low, medium or high) and recommended action, such as release, escalate for manual review or hold. Then add one line on what further evidence would change your mind. That last line shows the reader you have considered being wrong.

Lab: review a synthetic identity application

Now we apply the same thinking to a realistic case. The lab gives you eight observations on file for one bank application. Some point to a real person, some to a fabricated one, and some are noise. Your job is to sort them by strength, as you would in a real review.

Your turn

Real or Generated?

This lab is the Day 1 to 3 workflow in miniature. You are not hunting for one trick, you are weighing eight observations and deciding which ones deserve trust. It rewards the habit of rating signals instead of reacting to the loudest one.

Open the lab in a new tab

What to try

What to notice

  • Which observations are easy to fake and which are hard
  • Whether any detail only looks suspicious because of compression or stripped metadata
  • How your confidence changes when two independent signals agree

If it did not work

Treating missing metadata as proof of generation

Rate it weak. Many platforms strip metadata. Look for independent signals before concluding anything.

Running tools on the original file

Work on a copy and re-check the hash if you suspect the file changed.

Writing a verdict without listing the checks behind it

Log every check with its rating first, then derive the verdict from the log.

Trusting a detector score as the answer

Record it as one weak or moderate signal and support it with other evidence.

Recap

  • Hashes prove integrity, metadata gives leads, artefacts give visual and statistical clues
  • Preserve first, observe second, judge last
  • Rate each signal separately and state overall confidence
  • Always consider an innocent explanation for each signal
  • A good note lets another analyst repeat your steps and reach the same result
  • Agreement between independent signals matters more than any single oddity

Cheat sheet

sha256sum file
Record the file fingerprint (shasum -a 256 on macOS)
exiftool file
Show metadata fields
file name
Identify the real file type
strings -n 8 file
Show readable text embedded in the file
Ratings
Strong, moderate, weak, none
Triage order
Preserve, metadata, structure, pixels, rate and conclude

Before the next day

Mission: One-page triage note

Complete the lab, then write a one-page triage note on the application photo. The aim is a document a colleague could pick up and repeat without asking you anything.

About 35 min

You are done when: A one-page triage-note.md with checks, findings, ratings, verdict, confidence and recommended action.

Want more? Swap notes with another learner or re-read yours after a break and try to repeat every check from the note alone. Fix anything unclear.

Quiz

4 questions. Your score is saved with your progress.

1. Which order best describes a defensible triage workflow?

2. Two weak signals agree and one strong signal points the other way. What is the sound approach?

3. Why is a hash NOT evidence that an image is genuine?

4. What makes a triage note useful to another analyst?

Finish the day

Mark the day complete to save your progress.

Go to Day 5

Next: Day 5 moves to video. We break a clip into frames and streams with ffmpeg and ffprobe, and the same preserve, observe, rate habit carries straight over.

Tell someone you are doing this:WhatsAppXLinkedIn

Your notes

Private to this device. Write what you want to remember from today.