P-037 Phase B / B1: allowlist docs/evidence/p037-b- as a prefix, before R_B lands

Caught before pushing this time, not after: R_B (the Phase-B baseline) is
about to add five new docs/evidence/p037-b-baseline-*.json/.md artifacts.
Every one of them would trip the same allowlist gate d5f9134 just fixed
for scripts/p037_mos_snapshot.py/p037_verdict_snapshot.py -- an exact-
file PHASE_B_GOVERNANCE_FILES entry would need one more addition per
future Phase-B evidence take, forever, the same shape of problem a2d
itself hit and answered with dynamic r_d_files/D_AFTER_EVIDENCE_PINS
discovery from its own epoch record.

Phase B's own evidence surface is simpler than a2d's: every artifact it
will ever write lives under one already-owned, already-fixed prefix. So
PHASE_B_PREFIXES gains "docs/evidence/p037-b-" alongside the existing
"formal/p037-kernel/" entry, covering every present and future Phase-B
evidence file by construction rather than by remembering to add another
exact name each time. Directly simulated against the five filenames R_B
is about to add (p037_a2d_epoch.PHASE_B_PREFIXES, checked via
str.startswith) before committing, not discovered by CI after the fact:
all five covered. The three PHASE_B_GOVERNANCE_FILES entries this prefix
also happens to cover (p037-b-epoch.json and its two ledgers) are left
listed individually -- redundant, not wrong, and not worth shrinking a
tuple a prior commit already reasoned about.

tests/run_tests.py re-run in full (0 failures), ruff, mypy, the a2d
epoch gate, the proof boundary and the B production-diff gate (IDENTICAL
against 5571ba4) all stay green. test_p037_evidence.py's own
fresh-record-valid check showed its usual dirty-tree transient while
this edit was still uncommitted -- expected, gone once committed, not a
regression (same as d5f9134's own commit message already documented for
the identical mechanism).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AN6xHbpovxrmZ4AMjS7WQA

Author: claude

tests/test_p037_a2d_epoch.py

@@ -104,7 +104,21 @@
     "scripts/p037_mos_snapshot.py",
     "scripts/p037_verdict_snapshot.py",
 )
-PHASE_B_PREFIXES = ("formal/p037-kernel/",)
+#
+# R_B (the Phase-B baseline, taken at T_B) adds new docs/evidence/p037-b-
+# baseline-*.json artifacts plus their own manifest -- an exact-file list
+# would need one more entry per take, forever, the same problem a2d itself
+# hit and answered with r_d_files/D_AFTER_EVIDENCE_PINS (dynamic discovery
+# from the a2d epoch record's own named_later fields). Phase B's own
+# evidence surface is simpler -- every one of its artifacts already lives
+# under one exact, already-owned prefix -- so a prefix is the proportionate
+# fix here rather than porting a2d's discovery machinery for a single
+# baseline take. The three PHASE_B_GOVERNANCE_FILES entries this prefix
+# also covers (p037-b-epoch.json and the two ledgers) are left listed
+# individually rather than pruned -- redundant, not wrong, and this test's
+# own history already prefers additive fixes over shrinking a tuple a prior
+# commit reasoned about.
+PHASE_B_PREFIXES = ("formal/p037-kernel/", "docs/evidence/p037-b-")
 
 # 10.6.14b: order step 6 (D after) lands its evidence as exactly these six
 # files under docs/evidence/, and nothing about them is inferred or