release: coordinate v0.5.3 benchmark efficiency and shared-tooling adoption #326

Summary

Coordinate the v0.5.3 benchmark workflow workstream: keep routine CI useful, review scientific timing combinations, integrate shared progress/recovery support, and remove duplicate release acquisition only under a verified evidence contract.

This tracker covers the recommendations requested on 2026-10-29. It does not alter v0.5.2, stop its running measurement, authorize code/Git changes, or redefine unrelated release scope.

Current implementation boundary — 2026-11-08

#321 is complete. #320, #322, #323, #324 and #325 closed through PR #335, merged as 7c3a3625220773369c089e64e568a979abc0bb7f. Its Linux, macOS, Windows, Cargo packaging and static checks passed. All six native prerequisite issues of this tracker are closed; PR CI success does not establish final signed main-push release evidence.

PR #339 completes the remaining #330 common-command/guide, canonical dispatch and automatic-draft work, and #338's independent validation lanes and managed Cargo compilation cache wiring. It uses authorized research-repo-tools development pin cd8594f658b5bf3ade063a685834565cc62e0cb3. Both serial and parallel local full CI passed with 438 Python and 1,143 Rust tests, and Astra's source review found no remaining trust-boundary blocker.

This tracker owns the implementation-readiness handoff and closes when #339 merges, alongside #330 and #338. Required PR checks and review still gate merge. #333 owns final native release validation, authenticated hosted retained-evidence handoff, matched cold/warm timing evidence, exact published supplier availability, normal manifest/lock/isolated invocation adoption and actual release readiness. Those outstanding release obligations do not keep this implementation tracker open.

Historical context

The v0.5.2 local plan contains 2,673 measurements and expanded public API coverage. Removing broad peer/diagnostic timing did not establish an end-to-end speedup. At the earlier read-only observation, baseline acquisition was complete and current acquisition was still running; no final duration is claimed here.

Ordinary CI compiles benchmarks without sampling, but a separate PR/main workflow still runs unfiltered exact timings. The hosted release path collects 1,403 current-version cases again under a different artifact contract. Missing old APIs are already tracked in #320 and research-repo-tools#166.

Proposed Changes, Order and PR Groups

Use independently reviewable PRs for CI policy and scientific selection. Shared availability, progress, resume and reuse are distinct contract boundaries; combine implementation only with a demonstrated review benefit. Under #333, coordinate final published dependency pin/lock updates across the release to avoid churn, while keeping recovery and publication-trust changes independently reviewable. These are PR group candidates, not a requirement for one PR per issue.

Benefits

Makes the next release's cost reductions, scientific guarantees, upstream availability and Astra decisions explicit instead of treating fewer commands or merged upstream code as completion.

Implementation Notes and Acceptance

• Mandatory workstream scope: complete Make routine CI validate benchmarks without automatic full timing runs #321 and Review release benchmark combinations while preserving public scientific coverage #322; reconcile the implementations in Adopt shared benchmark workload availability across API revisions #320, Adopt one verified benchmark acquisition for release reports and artifacts #323, Adopt shared benchmark budgets, progress, and remaining-time reporting #324 and Adopt validated benchmark checkpoint and resume without mixing incompatible evidence #325 with merged PR/CI and native/scientific/documentation evidence. Final supplying-publication and normal-package adoption checks belong to release: plan, publish and verify la-stack v0.5.3 #333.
• Supporting work: generic implementation/installed-package validation belongs upstream (research-repo-tools#166–research-repo-tools#169), and verified publication belongs to research-repo-tools#170. la-stack owns workload semantics, native targets, consumer wiring, release ordering and approval safeguards.
• Astra checkpoints: Review release benchmark combinations while preserving public scientific coverage #322 scientific coverage; upstream research-repo-tools#166 availability; upstream research-repo-tools#168 safe recovery; upstream research-repo-tools#169 and local Adopt one verified benchmark acquisition for release reports and artifacts #323 evidence/release ordering. Make routine CI validate benchmarks without automatic full timing runs #321/Adopt shared benchmark budgets, progress, and remaining-time reporting #324 need no routine Astra escalation unless their contract expands.
• Deferred scope: blanket sampling reductions, a new benchmarking engine, retrospective changes to v0.5.2 and unrelated Rust/toolchain/API work are outside this recommendation set. No requested capability is silently deferred.
• If an architecture checkpoint becomes larger than a bounded release, record the exact issue, remaining acceptance and choices: retain with revised scope or obtain an explicit maintainer decision to defer while preserving safe existing behavior. Do not mark fallback-only behavior as completed adoption.
• Distinguish implemented, merged, published, pinned and consumer-verified states. Development rehearsals may use the exact maintainer-authorized pin. release: plan, publish and verify la-stack v0.5.3 #333 owns release adoption through a normally resolved published package, without sibling/editable/Git/path/private-API bridges; these release gates do not block implementation closure.
• Preserve comprehensive correctness tests, native features/platforms, unchanged historical evidence, scientific comparison eligibility and protected publication checks.
• Record comparable future timings only when acquisition is separately authorized. Separate aggregate work saved from critical-path elapsed time; estimates are not measured speedups.
• Readiness cutoff: once the retained implementation, required PR CI/merge and remaining Align release guide and commands with research-repo-tools evidence-reuse flow #330 guide/workflow gates are verified, hand off to release: plan, publish and verify la-stack v0.5.3 #333 for final published-tooling adoption and v0.5.3 release preparation. This tracker closes for workstream readiness, not a claim that the crate is published; actual registry publication must be recorded by the maintainer release process.

Next executable item: complete required PR checks and review for #339, whose merge closes this implementation tracker, #330 and #338. #333 then owns the remaining release validation, published-tooling adoption and v0.5.3 release preparation. Scientific and trust checkpoints already recorded in #335 and #339 need renewed review only for substantive changes. The six recorded native prerequisites are closed; no reverse dependency on upstream publication is introduced.