gov(sprint-12/wave-G): W-Meta-Opus honest review (grade A−) + CSI-15 rename CamPqWitnessIndex → WitnessIndexHashMap

W-Meta-Opus delivered the honest cross-cutting review (180 lines,
22.4 KB). Highlights:

**Wave G grade: A−** (Wave F was B; integration discipline corrected)
**Sprint-12-so-far: B+**

Per-worker grades:
- W-G1 A−  D-CSV-5b cutover; 86/86 tests pass; 4 pre-existing tests
  updated with discipline
- W-G2 B+  CAM-PQ HashMap wiring (15 tests); naming pre-commits to
  semantics the backing doesn't have (→ CSI-15, fixed in this commit)
- W-G3 A−  Batch i4 API (20 tests, clippy clean); all 5 functions
  panic-on-length-mismatch verified
- W-G4 A   Jirak threshold (20 tests); caught + corrected my math
  error in the spec ("p≥4 collapses linear" was inverted; p=4 produces
  MORE convex spacing, not less)
- W-G5 A   Iron rule promotion clean; CLAUDE.md now has 4 iron rules
- W-G6 A−  Workspace conflict resolution (3-LOC fix); no orphan crates

Three CSIs surfaced (CSI-14..18, continuing Wave F's sequence):

**CSI-15 (P2, FIXED IN THIS COMMIT):** Renamed `CamPqWitnessIndex` →
`WitnessIndexHashMap` (12 occurrences in witness_corpus.rs + 1 in
mod.rs re-export). The original name pre-committed to CAM-PQ distance-
ranking semantics that the HashMap backing doesn't deliver. The doc
comment is honest ("HashMap-backed; sprint-13+ upgrades to real
ndarray::hpc::cam_pq"), but the type name would mislead at consumer
call sites. Sprint-13+ adds `WitnessIndexCamPq` with the real codec
wiring; the HashMap variant becomes the fallback for environments
without ndarray CAM-PQ support.

**CSI-7 follow-through:** Verified sigma-tier-router has NO self-
declared [workspace] table; cargo metadata --no-deps shows both
sigma-tier-router AND cognitive-shader-driver in the workspace
(16 packages total). Opus's note about pending follow-through was
unwarranted — it's already done.

**CSI-18 (MED):** Four iron rules now share a meta-pattern ("do X
consistently across the codebase; document deviations explicitly").
Sprint-13 should land `.claude/knowledge/iron-rules-doctrine.md`
consolidating this meta-pattern so future iron rules fit the
template intentionally.

W-Meta-Opus standout call-out
W-G4 noticed my brief was inverted, consulted the I-NOISE-FLOOR-JIRAK
iron rule, derived the correct math (Σ_k = k^1.5 / 10^1.5 with
Σ_1 ≈ 0.031623 and Σ_10 = 1.0 exactly), and shipped 8 new tests with
the corrected assertion in test_jirak_p_4_more_linear. Jirak 2016
citation present in both module-level and method-level doc comments.

Sprint-13 spawn recommendation: YES, gated on Wave G merge +
sprint-13 worker template that bakes lib.rs/workspace registration
INTO worker scope (Wave F CSI-13 has now been validated as a
recurring failure mode unless main thread enforces it pre-spawn).

Pre-existing infra issue: local `cargo test -p lance-graph` hit
ENOSPC during link (workspace target/ disk pressure). Mechanical
12-instance rename did not require local re-test; CI validates.

https://claude.ai/code/session_01UwJuKqP828qyX1VkLgGJFS

Author: claude