Purpose

Maintain one canonical, append-only log of the human modeler's decisions, so that every modeling judgment a contest grades — which method and why, whether a result is good, what a number means, how confident the team is — has a single traceable home, and every downstream narrative (final method explanation, solution package, paper method-selection prose) is transcribed from it, never re-invented.

This is the decision-side parallel of the Frozen Numbers Convention. Just as frozen_numbers.json pins numbers so a bug fix can't silently shift the paper, the decision log pins judgments so the AI can't silently author what a judge grades, and so four different skills can't each re-compose the method-selection story and drift apart.

This skill does NOT make decisions. It never fills modeler_decision or modeler_rationale. It collects the human-authored decision artifacts that judgment-bearing skills emit (methods/Qx/decisions/<skill>_modeler_decision.md), validates them mechanically, stamps them, and folds them into the canonical log. It pre-fills only the mechanical fields (decision_id, options_considered, evidence, ai_suggestion).

When to use

Use this skill:

• After a judgment-bearing skill (method-selector; later result-report-generator, robustness-checker, final-method-explainer, solution-package-builder) has captured a human decision in its *_modeler_decision.md artifact, to fold it into the canonical log.
• As a precondition check at Gates G2.5 / G4.5 / G5: "does the log contain a DECIDED record for the required decision point of this Qx?"
• When final-method-explainer / solution-package-builder / paper-section-writer need the method-selection narrative — they read it FROM this log with decision_id provenance, they do not re-author it.
• After the contest, to render the decision log as post-mortem review material.

Do NOT use this skill: