docs: x-FuSa spec v1.14.0 — detection heuristics, attestation, traceability MUSTs, coverage metrics

[SoundMatt]

Summary

Follows directly from v1.13.0 (evidence-artifact schemas + content-quality baseline). That release said what good content looks like but left it unenforceable — no detection algorithm, no accountability, no required traceability, no completeness metric. This closes those four gaps.

1. §1.6.1 Detection heuristics — concrete, checkable rules instead of prose:

• Rule A (placeholder text) — deny-list match, always an ERROR (FUSA-STUB001), disposition-suppressible only.
• Rule B (blanket qualitative fallback) — distinct-value-ratio < 0.1 across ≥10 entries, a WARNING (FUSA-STUB002) by default, deliberately advisory since a similarity score can misfire on genuinely repetitive-but-real content.

2. §1.6.2 Attestation — the false-flag fix. Rather than trying to make the heuristic perfectly precise, an artifact can carry an optional attestation object (status, implementationAuthor, independentReviewer, reviewedAt, contentHash) reusing the exact independence model already in FuSaOps' own vv package (ISO 26262-2:2018 §6.4). A non-stale (hash-pinned), genuinely-independent "reviewed" attestation suppresses Rule B. --strict/--require-attestation escalates an unsuppressed Rule B finding to exit 1. This is the DCO pattern applied to evidence artifacts: nothing blocks by default, a project that wants a hard gate turns it on, and the gate is satisfied by an honest human assertion, not by gaming the similarity score.

3. Traceability MUST — .fusa-hara.json's safetyGoals[].fssrRef promoted from SHOULD (singular) to fssrRefs MUST, ≥1 entry — a safety goal with no decomposing requirement is exactly the gap ISO 26262-8 Clause 6 traceability exists to prevent.

4. Completeness metrics — fmea/tara gain summary.coveragePct + --min-coverage N, mirroring trace --func-coverage, so an artifact can't quietly analyze 5 convenient cases while looking thorough by entry count alone.

SpecVersion bumped 1.13.0 → 1.14.0. Scope note (unchanged from v1.13.0): spec-design only, no tool implements any of this yet, FuSaOps doesn't consume it.

Test plan

• go build ./... / go vet ./... — clean
• go test ./... — all packages pass
• Manual diff review — doc-only change plus the two version constants