Repository navigation
Conversation
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
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. Comment |
|
Closing this original stacked proposal as selectively incorporated or superseded by merged #56 (generic scientific core) and #57 (generic scoped MCP). Captured CPU runtime and comparison owners/P5a evidence incorporated via qualified source and PR56; later runtime and material evolution supersedes the old assembly. Independent cleanup review checked this exact head The original branch is retained. Its head is not ancestral to main, so this cleanup neither deletes its history nor merges its excluded ancestry. |
Integration disposition — 2026-10-14
Current reconciliation (2026-10-14): qualified generic content from this original stack is superseded and integrated in merged PR56 on main, without unqualified PR43 ancestry. This original draft/branch remains retained as source/evidence and for any producer-v2-specific dependent guards. It is no longer a pending generic-feature merge. Do not merge the deferred legacy stack wholesale; its original actual-data acceptance remains open.
Original source and qualification record follows:
Captured scientific recipes can now run under an explicitly inventoried CPU Python snapshot through the existing approved Runs lifecycle. An optional Rheon adapter opens the original checkpoint and unchanged lab scripts as a real Tuldok project, then runs the full CPU comparison from the saved Pipeline graph.
The actual UI workflow passed: 12 timed trajectories / 1,440 frames, four full accuracy replay cases, 24 native launches, six hash-verified downloads and exact results after restart. The maximum reported velocity/tracer relative error was 9.005e-8. Original lab files and checkpoint bytes are unchanged. Separate real runtime qualification covers computation, weights-only reload and cancelled/drained execution; generic startup without adapters passed.
This is an existing-checkpoint CPU comparison on previously exposed seeds, not fresh training, GPU qualification or historical timing reproduction. Accuracy arrays are computed in memory but not retained by this script; release-consuming fitting and full-field evaluation are the next milestones.
Validation: independent source/numerical review; 22 helper, 12 snapshot, 4 CPU material and 10 adapter checks pass, along with affected syntax/controller/plan checks. Final full Python regression: 807 passed in 1,050.525 seconds; two optional-reader skips are covered by 20 additional passing native-reader tests. All 190 Python source hashes stayed unchanged. Hosted UI 38456049102 passed; lifecycle 38456049347 also passed at exact head
05ea66e4b14e7e23b2d153f98380d44686cb53ed.Evidence: P5a verification, including actual receipts, numerical audit, source hashes and desktop/narrow screenshots.
Stacked on draft PR49. No merge or deployment.