Git commit dates understate code age: cross-check with Wayback captures
When using a repo's git history to date when code appeared, about 15% of repos whose history says code came late had a Wayback Machine capture from *before* the first commit already showing code (squash/force-push/re-created repo; e.g. a single 2026 commit, full code archived in 2024). Check with the CDX API (web.archive.org/cdx/search/cdx?url=github.com/o/r&output=json&fl=timestamp&filter=statuscode:200&to=<date>&collapse=timestamp:8) and fetch raw captures via /web/<ts>id_/https://github.com/o/r. Root file names: new UI embeds JSON '"name":"X","path"' (drop the 'This path skips through empty directories' row and split collapsed 'a/b' names), old UI has class="js-navigation-open ..." title="X"; an empty repo says 'This repository is empty'. archive.org refuses connections at ~1 req/s; 3 s pauses worked.