Skip to content

Retain complete native evidence in benchmark runs - #101

Closed
gauravtiwari wants to merge 1 commit into
mainfrom
codex/product-unit-evidence-20261116
Closed

gauravtiwari wants to merge 1 commit into
mainfrom
codex/product-unit-evidence-20261116

Conversation

@gauravtiwari

Copy link
Copy Markdown
Contributor

Wrapped BoringCache lanes now retain the CLI's common native-evidence export, including complete records. Result metadata keeps primary compiler evidence and additional actions separate; missing or unreadable data remains an explicit gap. Export writing stays inside the timed command. This adds no native log parser or build policy.

The retained LLVM and Ninja measurements are separate from provider rankings. The additive cohort contains 16 successful invocations, 236 action records and 143 primary ccache records. Outcome parity is checked by two all-hit read-only pairs. Live storage of the additional file kind remains unqualified, and one pending lookup file is retained as a gap. Scheduled plans keep their released CLI pins.

Validation: 223 tests, 882 assertions, no failures and three existing skips; catalog validation covers 15 tools and 45 cases; the LLVM matrix retains two lanes. Measurement checksums and public-content checks pass.

@gauravtiwari
gauravtiwari deleted the codex/product-unit-evidence-20261116 branch November 16, 2026 09:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant