Method and caveats

The rules

For US films this registry applies two rules from the Copyright Act as it stands in 2026.

1. A copyright term that began before 1931 has ended. The longest possible term for such works was 95 years from first publication.
2. A work first published 1931–1963 had to be renewed in its 28th year. If no renewal was filed, it entered the public domain at the end of that year. If renewed, it is protected for 95 years from publication.

The registry covers only works registered in the United States. It says nothing about foreign works restored by the Uruguay Round Agreements Act, underlying rights (a novel or play the film adapted, its music), or trademarks. Each status is a reading of the record, not legal advice.

Sources

Registrations: the Library of Congress's cumulative Catalog of Copyright Entries, Motion Pictures 1912–1939, transcribed by Project Gutenberg (#51836). Renewals: the annual Catalog of Copyright Entries, Third Series, Parts 12–13: Motion Pictures, scanned by the Internet Archive; each semiannual issue ends with a renewal section. Every dossier shows the exact catalog text used.

What the renewal parse achieved, per volume

Unlinked records mostly concern films from 1940–41 (not yet in the registry) or are too damaged to read. A damaged record could be the renewal of a film shown here as "no renewal found". That is the direction of error this site must not make, so the gap is shown rather than hidden, and the unresolved records are being worked through.

How wrong is "no renewal found"?

This is the error that matters, so it is measured rather than assumed. scripts/audit_no_renewal.py draws a random sample of films marked no-renewal-found and searches the raw OCR of their three renewal volumes for the same registration number near a renewal marker, or the title with the right year. A person then judges each candidate. On 5 September 2026 the first sample found a real missed renewal for 11 of 40 films; the parser was rebuilt around those failures, and the next two samples found 4 of 60 and 2 of 80. The remaining misses are records whose number, date and title were all damaged in the OCR, or pages whose columns were interleaved by the scanner. The audit is repeated after every parser change and the latest figure is kept here. Until that rate is near zero, treat "no renewal found" as strong evidence that still wants a look at the linked volumes.

How links are made

Renewal entries cite the original registration number and date. A link is made when both agree (strongest), or when the number agrees and the title is similar, or the date agrees and the title is clearly similar. For records the parser could not read, a language model re-reads the OCR with its surrounding lines; its reading is accepted only when the number and date or title it reports agree with the catalog, and the dossier says so.

Data

Every film page has a JSON version at /api/film/<id>.json; totals are at /api/summary.json.

Bulk files, CC0, regenerated after each pipeline run: registrations with status (CSV), the same as JSON lines, every parsed renewal record with raw OCR and flags, README. The source code will be published once the pipeline settles.