Skip to content

Conversation

@Mythmaker28
Copy link
Owner

This PR introduces a single, authoritative status file and points docs to it.

  • Adds deliverables/status.json with:
    • version: v2.2.2 (balanced)
    • dataset counts (221/30/22.6% Ca)
    • modeling decision: NO-GO with strict criteria (R² ≥ 0.20; MAE < 7.81)
    • evidence block linking to lab CV metrics path.

Goal: de-confuse GO vs NO-GO and establish one source of truth that future README/Pages/scripts can read from.

Next steps (separate PRs):

  • Wire README + Pages to read deliverables/status.json
  • Add requirements.txt and minimal repro scripts (build table, train RF, emit metrics)
  • Publish lab ZIP (shortlist, plates, protocol, SHA256) as release assets

Copy link
Owner Author

Added deliverables/lab_v2_2_2/cv_metrics.json with the exact CV results (r2=-0.1803; MAE=8.49; ΔMAE=24.23%; coverage90=0.9005; ECE≈0.00045). This anchors the NO-GO call to a concrete artifact. Next, we can: (1) drop cv_predictions_uq.csv as an asset, (2) publish lab ZIP as a Release, (3) wire README/Pages to deliverables/status.json.

If you prefer a GO-on-UQ (calibration-led) framing, we can reflect that explicitly as a lab-validation GO while keeping modeling NO-GO for predictive claims. Up to you, but the criteria gate remains unmet on R²/MAE.

Copy link
Owner Author

Added minimal repro guard:

  • requirements.txt (pinned)
  • scripts/eval_cv_metrics.py
  • CI workflow .github/workflows/repro-minimal.yml

This keeps master untouched but guarantees that the NO‑GO(pred)/GO(lab) verdict is backed by a reproducible, machine-checked artifact. Next, I can prepare the draft Release v2.2.2-lab with your ZIP + SHA256, and wire README/Pages to deliverables/status.json.

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.

2 participants