@blueforge-studio/audit-scoring

Pure, deterministic scoring for user/XP audits. No file, network or browser access; the caller reads and writes files. Proposal and decisions: PR #360 (docs/proposals/2026-10-17-audit-packages.md).

Parts:

• Scorecard. buildScorecard(dimensions) turns per-dimension penalties into 0-100 scores and a weighted aggregate. Every penalty carries its evidence and a required basis of observed or modelled. A dimension the audit could not finish is blocked: it has no score, gets no low-score suggestion, and stays out of the aggregate.
• Baseline gate. toBaseline(card) stores ids and scores only. gate(baseline, card, { tolerance }) compares each dimension with its own baseline and reports below-floor, went-null and dimension-removed. There is no global floor and no per-content comparison.
• Markdown. renderMarkdown(card, { gate }).

Types are shaped like auditor-sdk's so convergence later is mechanical; this package does not depend on it.

• Coverage (0.2.0). A dimension may declare coverage: { observed, declared }. Thin coverage lowers lowerBound and aggregateLowerBound (unwalked units assumed 0); it never raises a score.
• Ratchet (0.2.0). updateRatchet(state, card, { window }) raises each dimension's own floor to the minimum of its last window scored runs (default 3) and never lowers it; a blocked run changes nothing. overrideFloor is the only way down and records its reason. ratchetBaseline(state) feeds gate.

Not here: personas and runners (user-audit).