docs(audits): Sprint 52 conformance-gap consolidation

[dp-web4]

Summary

Post-Sprint-52 analytical memo (1 file, 229 lines) cataloguing the 8 conformance
xfails wired by PR #189 (Sprint 52 T1).

What's in the memo

• Catalogue of all 8 Sprint 52 xfails (test path + suite + failure mode).
• Audit-origin mapping: 3 xfails restate Sprint 47 T3/V3 cross-language audit findings (Talent decay CRITICAL, weighted composite HIGH, update formula HIGH). 5 xfails (62.5%) are NEW surface gaps not in any prior code-reading audit.
• Actionability tier classification:
  • Tier A (3 xfails) — CROSS-LANGUAGE-EXTERNAL-TOOLCHAIN: Sprint 47 class, needs Rust web4-trust-core.
  • Tier B (4 xfails) — DESIGN-QUESTION-NEEDS-OPERATOR: r6-val-004, r7-rep-001, role-004, fed-001.
  • Tier C (1 xfail) — NEEDS-SPEC-SCOPING: sub-001 sub-dimension rollup.
• Counter-finding preserved: ATP suite is 11/11 exact pass — Sprint 49's "ATP is best-aligned pair" claim is now operationally confirmed.
• Sprint 53+ candidate buckets: 4 autonomous-pickable (MCP-as-inter-society audit, mcp-protocol internal consistency, §7.7 promotion tracking, vector freshness check process), 7 operator-blocked, 3 external-track-blocked.

Key implication for Sprint 53

No Sprint 52 xfail is purely autonomous-actionable. A Sprint 53 framed as "fix the conformance xfails" would block on inputs the current track cannot provide. The autonomous-pickable bucket lies in MCP-audit class work, not conformance-fix class work.

Scope discipline

• Analysis only. No SDK code, no test changes, no spec edits, no vector edits.
• 1 new file in docs/audits/, consistent with Sprint 47/49 audit precedents.
• Policy-reviewed and approved by subagent reviewer (placeholder for hardbound PolicyEntity).

Test plan

• Memo reads coherently end-to-end.
• All 8 xfail line references resolve (test_conformance.py lines 115, 174, 213, 526, 575, 807, 826, 887).
• Audit-origin claims verifiable against Sprint 47 audit and Sprint 49 audit.
• No file changes outside docs/audits/.

Session: private-context/autonomous-sessions/legion-web4-20260515-000024-session.md

Co-authored by: Claude Opus 4.7 (1M context)

[dp-web4]

APPROVED: 1 new file in docs/audits/, analysis-only (no SDK/test/spec/vector changes), consistent with the established Sprint 47/49 audit-memo precedent in this development-phase repo. Substantive, non-padding decision-support: maps all 8 Sprint 52 conformance xfails to audit origin, establishes that 5/8 (62.5%) are NEW surface gaps no code-reading audit caught, and tiers each by actionability. Key actionable output — 'no Sprint 52 xfail is purely autonomous-actionable' — directly prevents a Sprint 53 category-error (a 'fix the xfails' sprint that would block on operator/Rust inputs) and identifies the genuinely autonomous-pickable bucket (MCP-audit class). The ATP counter-finding (11/11 exact) is the Sprint 49 claim hardened into an executed check. Standalone-doc concern noted but acceptable for audit-class artifacts under precedent; content connects directly to sprint planning. No drift.