[helly25]

Reconcile the performance candidates without losing rejected implementations or treating a completed subset as the whole plan. Retain the exact language-list reservation trial as an unselected experiment; production behavior stays unchanged.

AG;DR

• Reconcile original O01-O22 and later O23-O32 decisions against main 77aa9ef9 and the separately published follow-ups. Document actual remaining pipeline, archive, catalog, representation and integration requirements; keep the full goal active.
• Preserve the one-line language StringList reserve prototype privately. Six focused release suites pass on each host; no runtime, BUILD rule or research executable enters this PR.
• Retain v1 Google Benchmark's complete Linux and partial Mac evidence. The direct-clock v2 repeats all 16 cases in three 9/7/copy sessions, with 1,296 timed observations and 144 warmups per host, fresh-process cold-load semantics, full list-content oracles and Linux affinity.
• Report all 16 case medians and costs. Cold linked snapshot/catalog changes are small; they do not establish a useful practical command gain against the complete controls. Do not add an arbitrary empty-list threshold or claim an allocation/RSS/CLI improvement.
• Verify every raw v2 value, iteration count, fastest-seven mean, candidate/copy ratio and final median. Capsule: 13,361 records / 1,316,116 compressed bytes; SHA732d305e29e2c4eb148c303520de8350c21b13d08a8ce91fc7b5dd598051ba24. All 8,682 Linux transport hashes verified. Earlier successful samples are preserved and are never relabeled.
• Keep existing public heads unchanged. Append tracking addenda, register the report/audit/capsule with the current release-site layout and add only the exact evidence asset exception. Actual current-base documentation/site hooks and required public CI remain to be recorded after integration.

No final runtime/full-suite/sanitizer/local clang-tidy/production binary/ThinLTO/startup or whole-command speedup claim is made by this rejection screen. The new audit itself does not independently rerun every historical benchmark or prove completion of the active goal.

Dependency and public integration validation

This is a documentation-only child of #1140 (ac4e69b2f434126ca17fb500322c181565f92c41), preserving every earlier public report, map and exact evidence exception. The frozen reservation report and capsule hashes remain exact; the audit links are corrected as detailed below. Integrated release-site tests passed 32/32; all normal signed commit hooks and Trunk passed for the seven changed files. Local clang-tidy was explicitly skipped. This pushed PR must pass its own required CI; earlier experimental/private or parent checks are not readiness. The audit records an explicit main-77aa snapshot; GitHub and the ignored orchestration ledger govern the moving graph. No runtime, BUILD rule, temporary driver or allocation/RSS/startup claim is added.

Actual CI correction

The initial pushed head failed the full release-site build because two relative audit links referenced reports from the other unmerged chain (#1131/#1132). Signed correction 1dfd581c2ffc1ba25d61cb8eb22e6b8ab0158f14 pins only those links to the already-published immutable report commits; table spacing follows the required formatter. All 22 audit links now pass the actual release-site Links.rewrite implementation locally, and the pinned report Git objects were verified. The reservation report and evidence capsule are unchanged. Normal correction commit/push hooks and Trunk passed; no local clang-tidy or unpublished Markdown API preview ran. Fresh own-head CI remains required; initial failed CI is retained, not relabeled.

Synchronization after #1140 and #1144

#1140 merged as 50fe2426d690156e5c388eb879b6295c4e123801; #1144's native-case symlink fix is also on main. Merge commit 2a0e828c19be5655546971b4b34d884f3fcde14d resolves the six stacked integration conflicts while preserving all current-main documentation and mappings and this PR's seven-file audit/evidence payload. Both original language experiment reports and all 13,361 individual capsule records remain byte-identical and their length/SHA-256 hashes were checked. Normal merge commit and push hooks passed. Fresh current-main Test CI and repeat measurements are pending; the historical experiment remains labeled with its original source and images.

Fresh current-main remeasurement

Commit ec2f900572912046942b036229d8a471c6bf2cf8 retains a separate complete repeat of the original language-list reservation trial on main 50fe2426d690156e5c388eb879b6295c4e123801. Three sessions per host cover all 16 cases, 1,296 timed observations and 144 warmups per host; every raw observation, fastest-seven mean and ratio was independently recomputed. All 4,345 transferred Linux files and all 8,680 capsule records passed length/SHA-256 verification. Baseline, candidate, harness and driver are identical across hosts; binary hashes and exact-image controls are retained.

The full linked cold snapshot improves 0.909% macOS / 0.420% Linux; cold catalog changes -0.483% / +0.296%. Linux 8-language lists with zero or one value cost +1.694%/+1.892%; the 128-language empty case costs +1.683% (copy -0.827%). The lean macOS cold snapshot costs +4.718% (copy +2.226%). Complete case and control results are in docs/benchmarks/language-list-reserve-requalification-20261108.md. No practical command-level gain is established, so keep the reservation prototype unselected; this PR retains the audit and evidence without production runtime changes.

The 542,618-byte lossless capsule SHA-256 is 7ecdc251f181d979cc1ca839f725dc25343027588bca740923d895e4b82205fc. All 32 release-site tests passed locally, and normal commit hooks passed. Final pushed-head CI remains required.