feat(mrl): reconcile benchmark flags and sealed canary fixture

[TheHalfMoon]

Summary

Reconcile MRL-0604 benchmark-derived-generation flags and MRL-0606 R2-compatible sealed temporal-canary fixture workflow onto canonical main after MRL-0601/0602/0603/0605 construction-bound identity hardening became canonical.

This replacement is rebuilt directly from canonical 4b29a761e82e9fbe95278bcacd644dff40d1315e; it carries only the four intended feature/test files from superseded PR #295 and does not reuse its history or qualification evidence.

MRL-0604 reconciliation

• validate the original construction-bound TrainingExampleLineageContract instead of rebuilding from mutable live example fields
• validate the original construction-bound ContaminationEvidenceReport and TrainingTransformationBinding instead of copying live nested fields into new objects
• bind flags to one coherent lineage/source/contamination/transformation snapshot
• reject malformed and syntactically valid post-construction lineage, contamination, and transformation identity drift
• preserve closed benchmark classification semantics and no benchmark-access/generation/training/promotion authority

MRL-0606 reconciliation

• validate the original construction-bound TemporalCanaryManifest instead of rebuilding a fresh manifest from mutable live fields
• preserve coherent fixture surface/evaluator/parameter snapshots
• require manifest evaluator identity to match the exact fixture evaluator
• require the executed candidate identity to equal the sealed canary artifact identity
• preserve aggregate-only receipt semantics and sealed=True, fixture_only=True, can_enter_training=False, can_enter_search=False, can_authorize=False
• reject malformed and syntactically valid post-construction manifest identity drift

Exact qualified candidate

• canonical base: 4b29a761e82e9fbe95278bcacd644dff40d1315e
• exact head: 30a5d8441684d1f474565b1f9080d13ab603819e
• live compare: behind_by=0
• exact four-file scope:
  • src/medscale/mesc/_mrl_benchmark_derived_generation_v1.py
  • src/medscale/mesc/_mrl_temporal_canary_fixture_workflow_v1.py
  • tests/test_mesc_mrl_benchmark_derived_generation_v1.py
  • tests/test_mesc_mrl_temporal_canary_fixture_workflow_v1.py

Exact-head qualification

CI run 33360892949: PASS.

Both Python 3.11 and Python 3.12 lanes passed:

• locked dependency sync
• Ruff lint
• Ruff format
• strict mypy
• full pytest/coverage
• MRL machine-state drift/manual-edit gate
• medscale check

CodeQL run 33360892966: PASS. The Python analysis job completed successfully.

Final exact-head semantic review found no unresolved material defect. GitHub reports no submitted review findings and no inline review threads. Qodo remains billing-blocked and CodeRabbit automatic review is disabled because the repository is below its automatic-review threshold; neither provider spend nor human review was requested.

Boundary

This work records metadata and runs only the canonical deterministic in-memory fixture path. It performs no benchmark/corpus read, benchmark-derived generation, real model/provider/network/GPU work, training, promotion, deployment, release, or clinical action. MRL-0607 remains separately dependent on canonical MRL-0606.

Guarded merge is authorized only while canonical main remains 4b29a761e82e9fbe95278bcacd644dff40d1315e and the PR head remains exactly 30a5d8441684d1f474565b1f9080d13ab603819e. No force-push, rebase, destructive history rewrite, CI/CodeQL weakening, governance bypass, provider spend, or real-asset access was used.

[qodo-code-review]

ⓘ Qodo reviews are paused because your trial has ended. Ask your workspace admin to add credits to resume reviews. Manage billing

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b35fd747-3133-4f99-9908-66354110b241

---

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.}