Three formats and who each is for
| Format | For | Contains |
|---|---|---|
| PDF report | A client, a manager, a claim file | Score, band, scale, region map, signals, limitations |
| JSON export | A case system, a spreadsheet, an archive | Every tile score, file details, provenance state, timestamp |
| Share link | A colleague, a counterparty, a conversation | The analysis only. No image travels |
| Copy summary | A message or a ticket note | Verdict, score, band, decision line, in plain text |
The common mistake is sending a screenshot of the result screen. It carries none of the tile values, cannot be searched, and looks like an opinion rather than a record.
The PDF report
Built for somebody who was not there when the check ran. It opens with the file name and the time, then the score and its band, then the scale showing where that score sits, then the region map, then the signals behind the verdict.
It closes with the limitations, and that section is not optional decoration. A report that travels to a client or a counterparty needs to carry its own caveats, because the person reading it will not visit a website to find them.
The image and its heatmap are flattened into a single picture before printing. Browsers default to leaving background graphics out of a print, which would silently drop a CSS overlay and produce a report showing the photograph with no map on it.
The share link, and why it carries no image
A shareable result normally means uploading something. That would undo the point of a tool that never sends your file anywhere, so the link works differently.
The whole analysis is compressed into the URL fragment, the part after the hash. Browsers never transmit a fragment to a server, so a shared link discloses nothing to us, to a CDN or to any access log. Opening it rebuilds the report in the recipient's browser from the link itself.
Carried in the link
- Score, band and verdict
- The decision line
- Every region tile value
- The file name and check time
- Content Credentials state
Never carried
- The image itself
- A thumbnail of it
- Anything identifying you
- Anything stored on a server
A full report with an eight-tile region map comes to roughly four hundred characters, which pastes into any chat window. The recipient sees the reasoning without ever receiving the picture, which is often exactly the right split.
What to record in a case file
-
Save the original file, untouched
Not a screenshot and not a re-export. The file you checked is the evidence; the report describes it.
-
Export the JSON
Every tile score, the grid dimensions, the provenance state and a timestamp. This is what lets somebody reconstruct the reading later without the original tool.
-
Export the PDF too
For anybody who will read rather than parse. Two formats cost nothing and serve different readers.
-
Note what you did, not what you concluded
Which file, when, what the result was, and what you did next. A conclusion recorded without its inputs is not reviewable.
-
Keep it under the same retention as the case
Do not build a separate register of scores against named people, which turns a review step into profiling.
Sharing with somebody who will disagree
When a result is going to a claimant, a supplier or a student, the framing matters as much as the file. Send the report rather than the verdict, and say what you are asking for rather than what you concluded.
A message saying our check flagged this image and we would like the original file from the device is answerable. One saying our system determined this image is AI-generated invites a fight over a number that cannot support the weight being put on it.
What not to put in a report
The exports are deliberately narrow, and a few things people ask for are absent on purpose rather than because nobody got to them.
There is no verdict field saying fake or genuine. A band and a score communicate a likelihood; a single word communicates certainty that the underlying measurement cannot support, and once a word like that is in a document it travels further than the caveats around it.
There is no confidence percentage separate from the score. Two numbers invite a reader to multiply them, and there is no meaningful second figure to give. The score is the confidence.
There is no signature or certification mark on the PDF. A report that looks certified would be doing exactly the thing the limitations section says not to do, and a document that contradicts itself will be read selectively by whoever benefits from the half they prefer.