spb/autoscholar
Public
-
Fri, Oct 30, 2026 3
-
Runner: streaming checksums with oversize skips, robust finalize, 90-min reproduction cap, QA-driven revision pass on retry, no-pause policy; QA: tolerant metrics/log shapes, vendored libraries excluded from publication instead of failing; contract: dependency dirs and binaries never public
…
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
-
QA: execution evidence accepts platform reproduction or tolerant experiment logs; reported values may live in results reports; soft reproduction gate; admin re-evaluate action
…
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
-