qm_4d Track B — Architecture Decision Records

Design records for the chess-spectral v1.5.0 QM extension Track B (full move-as-unitary dynamics). Track A (kinematic-only qm_4d) shipped in python/chess_spectral/qm_4d.py on branch chess-spectral/qm-4d-kinematic. Track B implementation begins after these ADRs are accepted.

ADR	Title	Status (post-Phase-3.5)	Phase 4 milestone
001	Phase convention for unitary moves	✅ Accepted (Phase 3.5 PASS)	B[1]
002	Time-evolution semantics	✅ Accepted (Zeno; Stinespring deferred to v1.7+)	B[2]
003	Per-channel move-transformation derivation	⚠️ Accepted with two amendments — (1) Phase 3.5: FIB channels fall back to measurement-only re-encode (§3.3); (2) Phase 4 B1: strict-unitary tier restricted to same-orbit non-capture moves (cross-orbit → measurement-only)	B[3]
004	Z_2 superselection structure	⚠️ Accepted with amendment — §3.4 weakened to sector-flip-via-state_to_psi	B[4]
005	Pawn pseudo-Hermitian η-metric	⚠️ Accepted with amendment — decompose M_pawn = M_single + M_double (η-metric on single only)	B[5]

Phase 3.5 empirical validation: PHASE_3_5_PROBE_RESULTS.md is the authoritative validation record. Three of five ADRs got amendments based on probe results; their original bodies are preserved as the design record at the time of decision, but Phase 4 implementation reads ADR + probe-results doc together, with the probe-results doc winning on design-level decisions.

Phase 4 B1 amendment (ADR-003): ADR-003-AMENDMENT-orbit-restriction.md is the second amendment to ADR-003, surfaced by PR #76's projector-sandwich implementation. The strict-unitary tier (A_1, predicted STD4_*) is restricted to same-orbit non-capture moves on Z_8^4's 35 B_4 orbits; cross-orbit moves route through the measurement-only re-encode path established by the Phase 3.5 FIB amendment. v1.7+ orbit-quotient refactor (Option (b)) deferred.

Notebook context: chess_spectral_research_notebook.md §15-§17, chess_spectral_4d_notebook.md qm_4d Pre-flight Findings.