Skip to content

Fix deterministic replay, artifact integrity and resumable research - #41

Merged
docxology merged 4 commits into
mainfrom
work/biofirm-research-text-20261116
Nov 16, 2026
Merged

docxology merged 4 commits into
mainfrom
work/biofirm-research-text-20261116

Conversation

@docxology

Copy link
Copy Markdown
Collaborator

Changes

  • Canonicalize configuration ordering for hash-equivalent replay; validate initial state, bounded noise, reconstructed transitions and posterior entropy.
  • Align action IDs with effective signed control, including zero-strength decisions.
  • Checkpoint complete research artifact pairs before starting another job; preserve sentence-local decimal ranges and evidence offsets.
  • Repair the published configuration example and refresh the generated gate record from actual measurements, preserving its historical baseline.

Verification

  • Independently reviewed source; final review digest: 1ee547a38c75d7a0ba407efcfbd84df4b7d5b2fa5dced2bb58bb0ae0b583baa9.
  • Frozen Python 3.12.16 / PyMDP 1.0.5 dependency profile. Actual thirteen-arm measurement: full suite 160 passed, zero skipped, 90.91% coverage; fast suite 145 passed; original pinned GNN arm 15 passed. The committed measured record contains the actual per-arm results.
  • Final commit eacb659e5de198c864d2d9cca94ee3f3be2ac867: native freshness check and original GNN conformance passed, with unchanged source and no remaining descendants. Receipt SHA-256: d2f67bb5beae2033cdebd7f96be90a5e37157fe1aa7bde1c5982e47364155ccb.
  • Earlier negative receipts were retained and resolved through actual source/resource changes, not waivers.

No scientific calibration, new campaigns, live-provider usage, authoritative dataset changes, richer action-space decision, dependency-pin changes or historical baseline rewrite is included.

- Adopt the provenance-reviewed BF-C, BF-D and BF-F software continuation.
- Preserve native simulation work and scientific/provider acceptance holds.
- Adopt originally authored BF-A, BF-B and BF-E source and regressions.
- Document canonical ordering, effective actions and strict historical artifact checks.
- Keep frozen dependencies, scientific inputs and model definitions unchanged.
- Apply only the seven concrete formatting differences reported by the native lint arm.
- Preserve regression inputs, assertions and source behavior.
- Adopt actual 13-arm measurements at the unchanged frozen dependency lock; preserve historical baseline and negatives.
- Describe parser skips according to export availability, including the observed zero-skip full suite.
- No measurement values, thresholds, pins or checker rules are hand-adjusted.
@docxology
docxology merged commit 2d15cc9 into main Nov 16, 2026
7 checks passed
@docxology
docxology deleted the work/biofirm-research-text-20261116 branch November 16, 2026 22:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant