Summary

Honor both HTTP Retry-After formats on 429/503 responses while preserving the existing 30-second cap. Valid HTTP dates previously fell through to one-second fallback; nonfinite or negative values could also produce invalid sleeps. Fixtures use an injected clock and perform no network or real wait.

Evidence and validation

• Reproduced on unchanged upstream main ccf586d43d2a580319b6535c893668904d909eb9 with the regression tests in this PR.
• Focused native tests: python -m pytest -p no:cacheprovider -q tests/test_http_date_backoff.py tests/test_v2_pipeline.py pass after the patch; pre-patch failures are preserved locally.
• git diff --check passes.
• Unchanged upstream main in this environment: 4 failed, 713 passed, 136 skipped. Two failures import the unavailable optional mlx_vlm dependency; two import the absent test_p3_labeltrain fixture helper.
• Full python -m pytest -p no:cacheprovider -q: Exact PR head 8c182b4e1baf: 4 failed, 721 passed, 136 skipped, 35 warnings in 62.28s (0:01:02). Only those same four upstream/environment failures remain; this patch introduces no additional failures.
• This repository currently reports no GitHub status checks; that is not a claim of passing remote CI.
• Tests use local synthetic fixtures; no released model weights, private data, hosted-provider calls, or model quality claims.

Contract references

• https://www.rfc-editor.org/rfc/rfc9110.html#name-retry-after — Retry-After can be an HTTP-date or delay-seconds; scripts/v2/common.py documents bounded backoff for public dataset collectors.

AI assistance

Prepared with AI assistance; the reproduction, patch and tests were reviewed before submission.