Changelog

Initial publication. The four-axis protocol — Specification, Performance diagnosis, Explanation capability, Configuration sensitivity — with v1 calibrations anchored against the exemplar literature and the 2024–2026 retainer-audit distribution.

Artifacts: consolidated paper (§1–§5), Appendix A diagnostic question library (68 questions across the four axes), Appendix B scoring sheet and replication pseudocode for the four load-bearing procedures (entity-blinding, WRC/SPA-lite gate, κ-on-judge, harness-component swap), session-14 verified bibliography.

Calibration commitments held by v1:

• §2.1 Specification — entity-blinding 10pp rank-stability threshold; residualization t > 2 + 60% retention conjunction; purpose-masking ≤ noise-floor.
• §2.2 Performance — WRC/SPA-lite 5% adjusted-significance gate over the AI-inclusive enumerated variant universe; sub-period ±50% relative stability band; categorical complexity-premium sign-consistency criterion.
• §2.3 Explanation — variance-disclosure operational reconciliation; held-out judge κ > 0.6 across base-model-family-diverse panel; audit-trail re-execution recovery ≥ 95% on 100-decision sample; surrogate-model top-four-feature share ≤ 60%.
• §2.4 Configuration — base-model-swap Sharpe-delta ±25% relative band; harness-component-swap conjunctive (10pp absolute / 50% relative fab-rule + 10pp absolute trade-flip + 50% relative turnover); adversarial-misinformation trade-flip ≤ 10% at 0.5% feed-frequency.

v1 deferred to the v2–vN cycle: systemic-risk lens at the foundation-model-family resolution; multi-agent emergent-behaviour surface; novel-architecture calibration extensions (biological/quantum/fully-agentic); systemic-to-per-strategy calibration linkage.