Catalog vs Raw Files

Short version: nothing anyone contributed has been deleted. Every one of the 11,794 files handed to this archive is still on disk, byte for byte as it arrived, and every one of them is in the catalogue with a recorded reason for how it is being used. What differs between collections is how a file is published, and whether it is published yet.

This page exists because contributors reasonably ask why the site shows fewer pictures than they handed over. The honest answer has three parts: some files are the same picture twice, some are outside what this archive covers, and some are still waiting on a person to look at them. None of those means a file was thrown away.

To check your own files, use Where did my file go? Drop in the files you sent and it reports, for each one, whether the archive has it and what was done with it, with a link to the page wherever there is one. Nothing is uploaded — your browser computes a fingerprint from each file's contents and sends only that, so a folder of video can be checked without re-sending it. The rest of this page explains the same thing in aggregate.

First: the disk and the catalogue agree

If you count the raw folders yourself you may get a slightly lower number than 11,794. That is a counting artefact, not missing material. Scripts that count by listing extensions usually list .jpg .jpeg .png .mp4 .mov and stop there, which silently skips the 43 files stored as .webp (33), .heic (8), .gif (2). The catalogue counts by reading every file, so it sees them.

Sidecar files — the .json exports that came alongside the media — are metadata, not media. They are read for dates and captions and are counted as neither photographs nor video.

Every file is in exactly one of five states

So 7,044 files are published and visible on the site — 4,765 as dated records and 2,279 on memorial pages.

Duplicates: why one copy and not four

2,268 files — 1,707 images and 561 videos — are copies of something the archive already holds. During the protests the same photograph was forwarded through many channels, so the same frame arrives repeatedly, often re-compressed, cropped or watermarked along the way.

They are found by perceptual hashing, which compares what a picture looks like rather than its bytes, so a re-saved or slightly cropped copy is still recognised. Where several copies exist the archive keeps the best one — highest resolution, least compression, most intact metadata — and that is the copy that gets the public page. The others stay in the catalogue and are recorded as duplicates of it. If you contributed a copy that lost this comparison, your file is still held; a better copy of the same moment is what readers see.

The memorial material is published differently, not withheld

2,279 files come from two public memorial registers. Each is a photograph of a named person who was killed — a portrait, an ID card, a funeral, a page from a family album. Filing those as loose records on a daily timeline would have stripped the names off and left a wall of anonymous faces.

So they are published by person instead. Each name gets one page carrying their photographs, their dates, and their biography where the register recorded one. That is why this collection shows a large file count and a smaller page count: many files, one person.

The memorial is here.

What is still held, and why

2,225 files are catalogued but not yet public. Every one is waiting on a human decision, and the reason is recorded against the file:

• 690 — in-window (medium) but no topical text signal
• 683 — unverified provenance (Pinterest) — needs a human
• 253 — no resolvable date — forwarded from an archive channel that posted it
• 186 — no resolvable date
• 143 — new collection (duniyakapanojuly.org) — arrived without a policy decis
• 102 — carries a signal that the image may not be from Bangladesh at all (for example, it was sourced from an Indian news domain). Held until checked, because publishing another country's protest as this one would be a serious error.

Holding is not rejection. These move to the site as they are checked.

Outside the period

The revolution this archive documents runs from 1 July to 5 August 2024. The window the archive keeps runs further, to 30 August, because what followed the fall belongs to the same record.

257 files carry a date outside that window. Most are later material from the same sources — anniversary coverage, tribunal reporting, photographs published months afterwards. They are real and they are kept, but stretching the boundary further would make the timeline mean less. If you believe a date was read wrongly, say so — dates inferred from a filename or a file's modification time are the least reliable thing in the catalogue.

A dating mistake that hid three channels

Every message in the Telegram delivery is a forward, made in 2026, so the message's own date says nothing about when a photograph was taken. The archive therefore dates each item from the date the originating channel first posted it. For a channel that was posting as the uprising happened, that is a good proxy for the day itself.

For three channels it is not. They are retrospective archives: they re-publish 2024 material long after the fact, so the posting date lands one to two years after the events shown.

• জুলাই আগস্টে গণহত্যা — posted 2025-01 to 2025-02, 183 files
• The Red July Official — posted 2026-01 to 2026-07, 51 files
• জুলাই গনহত্যা ডকুমেন্টস — posted 2024-10 to 2025-04, 19 files

Because the date fell outside the publication window, 253 files were automatically marked as outside what this archive covers — and were never offered for review. That was wrong. The content is squarely in scope; what fell outside the window was the channel's posting date, not the photograph.

The rule has been corrected, and all 253 files are now held for dating rather than rejected. They cannot be dated automatically: the message text names a day in only a handful of cases, and a video's embedded timestamp is the moment Telegram received the upload, which repeats the same wrong date. So they are going back to the contributor who supplied them, with a worksheet showing each file, to be placed by someone who knows the material. A guessed date is worse than a gap, and anything nobody can place will stay held rather than be invented.

When a contributor sends the same channel twice

This is the commonest reason a large delivery adds a small number of records, and it is worth setting out plainly, because from the outside it looks like most of the files were thrown away.

A Telegram export renames its media every time it runs. The same photograph comes out as photo_1@13-05-2026…jpg in one export and photo_2@14-05-2026…jpg in the next. Nothing about the filename says they are the same picture, so a second export of a channel already sent looks like an entirely new set of files. The archive compares the contents instead: two files with the same checksum are the same file, whatever they are called.

Nothing in the duplicate rows was discarded. Every file is on disk and in the catalogue; the archive simply shows one copy of each photograph rather than four.

An export is not the same thing as the files

A Telegram export is two parts: a JSON file describing every message, and a folder of the media those messages contain. They can be sent separately, and when only the JSON arrives the archive can read every caption, sender and date and still has no photographs to show.

Every export listed above now has its media. BJTY Archive was for a time in the position this section describes — the archive held its messages and could date them, with no photographs to show. Its media folder has since been delivered and the files are catalogued. What became of them is set out on the accounting for that contributor.

Collection by collection

What the archive will not do with your files

• Nothing is deleted. The raw folders are never written to. Every state above is a note in a database beside the file, and every one of them can be reversed.
• Nothing is re-licensed. A large part of this material is agency preview imagery carrying watermarks. The archive does not grant anyone the right to republish what it does not own.
• Credit is not invented. Where the chain of custody is a forwarding channel rather than the photographer, the page says so instead of guessing at a name.

If something here is wrong

Tell us, and be specific — the collection, the filename, and what you expected. The commonest real errors are a misread date and a wrongly suppressed duplicate where two similar photographs are actually two different moments. Both are fixable, and both are easier to fix with a filename than with a description.

Figures read from the catalogue on 04 August 2026.