[MrScripty]

The lab reports substantial non-training workflow time, diagnostic publication failures after completed training, and custom HTML needed for scientific comparison. This documentation-only plan defines six incremental implementation milestones while preserving existing runtime, Run, Study and Pipeline ownership.

Plan: docs/plans/lab-workflow-reliability/plan.md.

1. Exact submission-bound preflight, bounded diagnostic evidence and same-Run publication recovery.
2. RES-1 coordinator health and safe custody reconciliation before recovery/readiness.
3. Live preparation/queue/execution/validation/publication timing with honest overlap and wait reasons.
4. Measured verified preparation reuse and bounded artifact batching, retaining per-item integrity/partial outcomes.
5. A small supported Python workflow SDK and matching existing CLI for submit/monitor/cancel/collect/compare.
6. Scientific baseline/candidate comparison in existing UI and lineage, with compatible data/units, per-case criteria and diagnostic/config/time/memory links.

The contract/acceptance report supplies exact invariants, security/resource limits, cancellation/retry/recovery semantics, consumer dispositions, compatibility/migration/rollback and future T1–T8 procedures. Publication recovery explicitly uses a versioned attached result Ref for new readers; legacy failed Run projections remain unchanged, and candidate/package promotion from recovered legacy failed Runs is unsupported initially.

Validation: unchanged adopted Coding Standards structural validator PASS; all 23 local links/anchors resolve; new-file/staged whitespace and eight-file docs-only scope PASS. Independent GPT-6.1 Sol High review is CLEAR after resolving one blocking consumer-contract finding and two recommendations; exact reviewed hashes and evidence are included. Standards Engine application routing is unavailable in this environment; no full Engine certification is claimed.

Baseline main 300f12d. PR89 (33069a8) and PR90 (429d0c4) remain open dependencies at preparation time, not merged facts. D1 is completed prior work and is not repeated. Historical timings remain observations, not a controlled whole-workflow speedup.

No product code, workflow, catalog or runtime changed. No application/CUDA/scientific tests ran; all product acceptance remains pending. Implementation requires a subsequent explicit instruction. This PR is for plan review only; do not merge or start implementation as part of this documentation task. Lead retains serial integration of PR89/90 and future implementation.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[MrScripty]

Disposition (2026-10-21): closing this original documentation proposal as superseded by the retained and evolved canonical plan on main, carried into implementation by PR92 (merge 2f333f960b80ca7fe23c90269b32f647b2c827a8) and subsequent integration through PR118.

Independent GPT-6.1 Sol High review compared original head b446d9a19dbd553b80bf9f3af574b8b34ddbd5cb with main 133d53c0e2985a6559ba9cebd7d4a9f530051f06: all eight files remain; four reports are byte-identical; ledger/issues are additive; plan/contracts changes record implementation progress and the explicit current-version support-policy revision at 9d699cc. No unique unretained technical requirement was found. Exact-head CI39290115808 passed; main's five push workflows also passed.

This closes an obsolete proposal without merging its stale plan state. The branch, original review/evidence and commits remain preserved. PUB-1's original failed-run evidence and SCI-1's owned criterion/compatible actual lab pair remain unavailable; plan closure does not grant scientific or hardware acceptance.