Which photo metadata survives a download: the dataset

What the sample is

Two frames of 300 accepted files each. The first is Wikimedia Commons, taken from the file namespace with the platform's own random generator. The second is the public timeline of mstdn.social, restricted to media posts.

Each frame is fetched several ways. The Commons files come back as the original file, as the copy the platform displays, and at two requested thumbnail widths of 200 and 1200 pixels on a 30-file subsample, with one older archived version as well. The Mastodon files come back as the stored file, as the thumbnail, and as the remote copy. Every response is then read three ways: from a head window, from a tail window, and, for 60 of them, in full.

That gives 600 file rows and 2540 unit rows, of which 1557 are head readings, 923 tail readings and 60 full readings. Seven requests failed and 52 were throttled, and both are recorded rather than dropped.

The data carries no image content, no coordinate values, no serial numbers and no account names. Every measured field is a presence flag. The identifiers for the Mastodon files are salted hashes and the salt is not published, so a row cannot be traced back to an account.

What is here

• DATASET.csv The measurements: one row per fetched file, one column per measured field, every field a yes or no.
• DATASET.json The run: where each sample came from, how the reading program was checked against two other readers, and the counts per group.
• PREREGISTRATION.md The question, the predicted share and the counting rules, all fixed before the first file was fetched.
• LICENSE.txt CC0 1.0 public domain dedication, so the data can be copied, changed and reused without asking.

The programs that produced the table are published beside it:

• MEASURE.mjs Requests each accepted address and reads the metadata block out of the bytes that came back.
• ANALYSE.mjs Turns the measured rows into the shares and intervals the article reports, with nothing typed in by hand.
• REPLICATE.mjs An independent second driver written against the pre-registration.
• REPLICATION.md The report of where the method and the run disagree.

How to cite

Viallo, Which Photo Metadata Survives a Download?, measured 10 September 2026. Viallo. https://www.viallo.app/studies/downloaded-photo-location-metadata