[zackees]

Summary

Closes #615. Separate benchmark PR requested by the owner; CPU work remains in #616 / #614.

The reported failure was TCMalloc's generated module-lock mismatch. Merged #608 already supplied the matching lock patch, pinned Bazel 9.1.0 and fail-closed dependency resolution. This PR preserves the original failure and proves restored full current-main publication; it does not duplicate that repair or manufacture a new source change.

Verified production path

• Full run https://github.com/zackees/mimalloc-pprof/actions/runs/39815683542 on exact main 227eef06ff03f92966f38508441d6c2f60a75135, 15 blocks, normal execution.
• All six workflow jobs passed, including allocator build, complete measurements, raw/site validation, branch publication, Pages deployment, and independent publication audit.
• 2,250 original samples, 30 cells, minimum 15 blocks per cell, all five allocators, no validation errors. Sealed envelope raw samples equal the original artifact.
• Published revision 0dde10e817210be049a5f4dae460f526f64f49b2; independent local artifact, branch and live Pages audits all passed.
• Existing allocator-builder/workflow tests: 30 passed. The lock-corruption and tool-version rejection tests remain intact.

Evidence

.github/assets/benchmark-615/ contains a reviewed README, archive digest and a 43-member archive with per-file hashes: original failure, complete raw/provenance/validation/site artifacts, successful workflow metadata/logs, test output and local audit logs. No generated replacement samples, removed allocators, relaxed checksums or validation thresholds.

This main-branch dispatch exercises the schedule's publication path; it does not claim that a future scheduled run has already succeeded. Carried-forward scaling data retains its own provenance, so #607 is not closed by this verification. CPU/RSS/p99 acceptance for #614 remains pending its separate original campaign.