Archive audits
For archives, agencies, publishers and licensed-data marketplaces preparing an image catalogue for licensing, including to AI developers: an account of what the catalogue holds, which files are the same picture, and what each file says about who made it and on what terms.
What an audit does
- Groups duplicates and variants. Every file that is the same picture is put in one group: identical files under other names, and copies that were resized, re-encoded, cropped, rotated, mirrored, or given a border or a caption. Very small crops and collages can be missed. It looks for copies, not look-alikes, and two photographs of the same scene usually stay separate, but it can occasionally join different photographs of the same scene by the same photographer, even ones taken months apart. Every group lists the evidence that joined it, so each can be reviewed, and a link that may be two photographs of one scene is labelled "same scene, check".
- Attributes images to creators. It reads the creator, credit and rights fields embedded in each file (EXIF, IPTC and XMP, including PLUS). The diagnostic reads embedded metadata only; a full audit also compares it with the contributor records you share with us.
- Flags unclear rights. Images with no creator named, no copyright or rights statement, different creators or rights holders named for the same picture, or attribution present on some copies and stripped from others.
- Produces a timestamped, verifiable record. Each image's record becomes a leaf of a Merkle tree, and the tree's root can be timestamped through OpenTimestamps. Anyone holding the report can later check that no record in it has changed since that date.
What you receive
Three files, and a fourth when the record is timestamped. No portal and no account.
| File | What it holds |
|---|---|
| report.html | A readable report: summary counts, each group of the same picture with the evidence that joined it, attribution coverage, every flagged image, and the method and its limits in plain language. |
| report.csv | One row per image: path, group, relation, creators, rights holders, usage terms, credit, flags, width, height, format, ISCC, pHash, leaf hash and any read error. Cells that a spreadsheet would run as a formula are quoted. |
| report.json | Every record with its digital fingerprints and attribution fields, the groups and their evidence, the thresholds used, and each record's Merkle proof against the root. |
| anchor.json | The OpenTimestamps filing of the root, when the audit is timestamped. |
| signals/ | If you state your terms: drafts of them as TDMRep, AI-preference and RSL files, ready to publish. Every image is reserved; a permission you state is signalled only for images whose metadata names you as the rights holder. These are your statements, not a finding of who holds what. |
Read a sample report, made by
registry audit run from public-domain photographs and planted
copies, with fictional names in the metadata, or download its
CSV,
JSON and
filing.
What never happens
- Your images never leave your systems. The audit software runs where the images are. What leaves is digital fingerprints and metadata text, never a picture, a thumbnail or a copy.
- We never rule on ownership. An audit reports what the files say and which files are the same picture. Metadata can be wrong, and the report says what was checked and what was not.
- Findings are not legal advice. They are evidence for you and your counsel to weigh.
- Your data stays yours. Audit findings are confidential and are never published or shared without your consent.
We decline audits whose purpose is to license work the client does not hold rights to.
A free diagnostic
We will run the diagnostic on a sample of about 1,000 images from your catalogue, free. You receive the same report and files as a full audit, for that sample, so you can judge the findings before deciding anything.
The diagnostic is run with the operator, on a machine you control. You can also try the public preview on a few images of your own; they are not stored.
Contact
Write to outreach@bakau.org. Say roughly how large the catalogue is and where the files are kept.