diff --git a/.claude/board/EPIPHANIES.md b/.claude/board/EPIPHANIES.md
index bec729e74..d3bee8fcd 100644
--- a/.claude/board/EPIPHANIES.md
+++ b/.claude/board/EPIPHANIES.md
@@ -1,3 +1,13 @@
+## 2026-07-18 — E-BGZ-TENSOR-LANE-REVIEW-1 — bgz-tensor's adaptive codec reviewed against turbovec/PolarQuant/helix; all four axes DEMARCATE/KEEP, not retire; PROBE-WH-MAG-2 confirms the WH-escape-centroid pairing
+
+**Status:** ENGINEERING-REVIEW FINDING (Opus reviewer, source-receipted; PROBE-WH-MAG-2 re-run this session; no refactor, no canon changed — TD-BGZ-TENSOR-PRE-LANE-REVIEW paid in part).
+**Frame:** bgz-tensor predates turbovec/PolarQuant/helix; the review asked per axis consume-the-lane / demarcate / retire. Result: the adaptive codec occupies a lane the three newer primitives do NOT cover — **materialized place+residual RECONSTRUCTION of unbounded heavy-tailed WEIGHT ROWS for GEMM** — so every axis rules DEMARCATE or KEEP.
+**(a) i4+i2 cascade vs turbovec Lloyd-Max/NativeLut — DEMARCATE (different workloads).** turbovec is a data-oblivious ANN SEARCH INDEX that decodes to ids and "trades the matmul away" (lance-graph-turbovec/lib.rs:49-53); the cascade reconstructs f32 rows (adaptive_codec.rs:311-341, cosine>0.9). Placements already split: turbovec = in-row 6×4-bit nibble; ladder = out-of-row (le-contract:216). Only overlap: the uniform i4/i2 leaf (ndarray quantized.rs:671,725) vs a Lloyd-Max leaf — a leaf swap (WI-1), not a lane consume.
+**(b) hadamard_rotate vs PolarQuant rotation/QJL — DEMARCATE; mechanism CONFIRMED.** Re-ran PROBE-WH-MAG-2: bare WH NEUTRAL (0.929/1.317/1.869) but WH+escape+centroid — the SHIPPED pairing (encode() step 3 + Passthrough) — wins on structure (gradient+spike cenB/A 0.209, heavy-tailed escB/A 0.998, noise flat). WH survives because bgz-tensor never applies it bare. PolarQuant's QJL error-correction is displaced for the bounded/cosine lane by the analytic Fisher-z canon (E-FISHERZ); the two PolarQuant probes are unrun scaffolds (WI-2).
+**(c) residual/location vs helix ResidueEdge/Signed360/CurveRuler — DEMARCATE (orthogonal place).** helix place = analytic/address-derived, regenerated from endpoints via the stride-4-over-17 walk, ZERO location storage (residue.rs:19-27, curve_ruler.rs:50-56, the Fisher-2z rung). bgz-tensor place = learned k-means centroid index, residual stored explicitly. helix cannot derive a k-means mean; bgz-tensor stores no regenerable phase. Genuine open overlap: the continuous-field geo exit (256 elevation terraces) is exactly helix's analytic-quantizer regime — WI-3 decides ownership.
+**(d) Passthrough top-10% — KEEP (refuse-the-mold virtue, empirically load-bearing).** Stores exact f32 for high-LFD outliers that no fixed-width register fits; routing them to Signed360 (lossy) or any Lloyd-Max lane re-introduces the failure the tier escapes. E-CHESS needle discipline. Re-run proof: heavy-tailed 1.32→0.998 via escape alone. Only sub-question: escape predicate LFD vs L-inf (WI-4).
+**Cross-ref:** TD-BGZ-TENSOR-PRE-LANE-REVIEW (paid-in-part), E-PALETTE-RESIDUAL-LADDER-1 (the honoured ladder), E-FISHERZ-CANONICAL-COSINE-REPLACEMENT-1 (QJL→Fisher-z displacement), E-H268-PROBE-WAVE-1-RESULTS (PROBE-WH-MAG bare-tile NEUTRAL), le-contract §3 honourable-mention + analytic-canon subsections, adaptive_codec.rs mod probe_wh_mag::probe_wh_mag_2. Open work-items WI-1..WI-4 in TECH_DEBT (none gating).
+
 ## 2026-07-16 — E-X265-PROBE-WAVE-2-RESULTS — two x265-replay conjectures measured: the sprite P-frame IS one helix code (Signed360 exact for helix-manifold motion, ResidueEdge-24bit insufficient); WH-magnitude TRANSFERS on structured tiles only, with the shipped escape+centroid pairing
 
 **Status:** FINDING (two probes run against shipped primitives, deterministic seeds, Opus-adjudicated; plans `x265-sprite-replay-probe-v1.md` + the PROBE-WH-MAG-2 home named in `h268-probe-wave-v1.md`). No canon changed. E-SPRITE-IPB-HELIX-1 graded PASS-AT-SIGNED360 (scoped); E-WH-TWO-SIDES-SIG-CHECKSUM-1 leg 1 UPGRADED from bare NOT-TRANSFERRING to pairing-TRANSFERS-on-structured-tiles.
diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md
index 076b03662..8c60d231e 100644
--- a/.claude/board/LATEST_STATE.md
+++ b/.claude/board/LATEST_STATE.md
@@ -1,3 +1,17 @@
+## 2026-07-18 — branch `claude/x265-x266-plans-review-h9osnl` (post-#733) — x265 probe wave-2 landed + bgz-tensor lane review paid
+
+#733 MERGED: the x265 wave-2 probes — PROBE-SPRITE-REPLAY (PASS-AT-SIGNED360
+scoped; ResidueEdge-24bit insufficient → Signed360-only motion primitive;
+arbitrary-motion [H]) + PROBE-WH-MAG-2 (upgrades the WH-magnitude leg to
+transfers-with-escape+centroid on structured tiles). EPIPHANIES
+`E-X265-PROBE-WAVE-2-RESULTS`; ndarray #248 companion merged. This follow-up:
+TD-BGZ-TENSOR-PRE-LANE-REVIEW **paid in part** — Opus review ruled all four
+axes DEMARCATE/KEEP (nothing retired; bgz-tensor's codec occupies a lane
+turbovec/PolarQuant/helix don't cover), `E-BGZ-TENSOR-LANE-REVIEW-1` + the
+TD payment addendum, 4 non-gating work-items WI-1..WI-4. Remaining x265
+queue: the arbitrary-motion sprite follow-up probe, PROBE-GPU-LUT (wgpu
+harness). This entry + the #733 PR_ARC entry = post-merge hygiene.
+
 ## 2026-07-18 — branch `claude/review-claude-board-files-nhqgx1` (PR #729) — P4 ancestry pipeline on the SoA/ractor carrier: triangle read seam + owned columns + the 226-atom FROZEN palette256 codebook
 
 ### Current Contract Inventory — new entry
diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md
index 3760803de..16e796958 100644
--- a/.claude/board/PR_ARC_INVENTORY.md
+++ b/.claude/board/PR_ARC_INVENTORY.md
@@ -33,6 +33,25 @@
 > - **Docs** — knowledge files produced (immutable)
 > - **Confidence (YYYY-MM-DD):** — the ONLY mutable field
 
+## #733 lance-graph: x265 probe wave-2 — sprite-replay + WH-MAG-2 (probe code + verdicts)
+
+**Status:** MERGED 2026-07-18 (merge commit `be68c2f`), branch `claude/x265-x266-plans-review-h9osnl`, doc + test-only probe code.
+
+**Added**
+- `crates/helix/src/sprite_replay.rs` (`probe_sprite_replay_core`, test-only) — x265 I/P/B replay via helix motion codes; **PASS-AT-SIGNED360 (scoped) / ResidueEdge-24bit INSUFFICIENT**, KILL did not fire (one Signed360 code per sprite reconstructs motion+B-frames exactly, for helix-manifold motion; arbitrary-motion generality stays [H]).
+- `crates/bgz-tensor/src/adaptive_codec.rs` `mod probe_wh_mag::probe_wh_mag_2` — WH + escape + centroid; **upgrades the WH-magnitude leg from bare NOT-TRANSFERRING to TRANSFERS-on-structured-tiles** (gradient+spike esc 0.815 / cen 0.209).
+- EPIPHANIES `E-X265-PROBE-WAVE-2-RESULTS`.
+
+**Locked** — the sprite motion primitive is Signed360-only (ResidueEdge-24bit is hemisphere-blind + rank-adjacency-hazarded); WH transfers ONLY as the shipped escape+centroid pairing on structured tiles (compound-column caveat: the win is a codec-pairing property, WH's isolated margin unmeasured).
+
+**Deferred** — the arbitrary-independent-ground-truth sprite motion probe (promotes helix-manifold [proven] → arbitrary-motion [H]); the wgpu/wasm decode tiers (shared PROBE-GPU-LUT harness).
+
+**Docs** — ndarray #248 companion (MERGED same day: plan Results + §5 wave-2 table + §10 RESULT-2).
+
+**Review** — CodeRabbit reviewed; clippy clean on both probe crates (no dead-code warnings despite the preview hint); Bugbot usage-limited.
+
+**Confidence (2026-07-18):** probes deterministic + green (helix 75+4+7, bgz-tensor 203); Opus-adjudicated LAND-AS-IS, all findings P2.
+
 ## #717 lance-graph: D-TRI-1 value-tenant half — the autopoiesis triangle (3 SoA lanes)
 
 **Status:** MERGED 2026-07-17 (merge commit `74d16f92`), branch `claude/review-claude-board-files-nhqgx1`, 3 commits, autonomously merged under the operator's standing merge authority.
diff --git a/.claude/board/TECH_DEBT.md b/.claude/board/TECH_DEBT.md
index 604c7eed6..6116940a9 100644
--- a/.claude/board/TECH_DEBT.md
+++ b/.claude/board/TECH_DEBT.md
@@ -54,6 +54,31 @@ doc-comment ("predates X, kept because Y") / retirement of the redundant
 half. Cross-ref: `E-PALETTE-RESIDUAL-LADDER-1`, le-contract §3
 honourable-mention subsection (carries a pointer to this entry).
 
+**PAID (partial) — 2026-07-18, engineering review (Opus reviewer; no refactor, evidence-grounded).**
+The four axes were reviewed against source + a re-run of PROBE-WH-MAG / PROBE-WH-MAG-2.
+RULED (credit): all four = DEMARCATE/KEEP — bgz-tensor's adaptive codec occupies a lane
+none of turbovec/PolarQuant/helix covers (materialized place+residual reconstruction of
+unbounded heavy-tailed WEIGHT ROWS for GEMM; the three newer primitives are search-index /
+analytic-cosine-rank / analytic-address-location respectively). No consume-the-lane, no retire.
+  (a) DEMARCATE — turbovec is a search index (decodes to ids, trades the matmul away,
+      lib.rs:49-53); this reconstructs f32 rows. Placements already split (turbovec = in-row
+      6×4-bit; ladder = out-of-row, le-contract:216).
+  (b) DEMARCATE — hadamard_rotate VALIDATED by PROBE-WH-MAG-2 (verified this session:
+      gradient+spike cenB/A 0.209, heavy-tailed escB/A 0.998) but ONLY in the shipped
+      escape+centroid pairing; bare WH is NEUTRAL. PolarQuant's QJL is displaced for the
+      bounded cosine lane by Fisher-z (E-FISHERZ).
+  (c) DEMARCATE — helix place is analytic/address-derived (regenerated, zero location
+      storage); bgz-tensor place is a learned k-means centroid (stored). Same decomposition,
+      different place kind.
+  (d) KEEP — Passthrough refuses the mold by design (E-PALETTE-RESIDUAL-LADDER-1 virtue);
+      load-bearing per re-run (heavy-tailed 1.32→0.998 via escape). Never route to a lossy lane.
+OPEN (debit — 4 NEEDS-PROBE work-items, none gating): WI-1 Lloyd-Max leaf swap (low);
+WI-2 run-or-retire the two unrun PolarQuant probes polarquant_hip_probe/turboquant_correction_probe
+(medium); WI-3 helix-vs-ladder on the continuous-field geo exit — the one genuine competition (medium);
+WI-4 escape predicate LFD-vs-L-inf agreement (low). Remaining payment = 4 doc-comment demarcations
++ WI-2/WI-3 (the only two that could still change a lane assignment). Cross-ref:
+`E-BGZ-TENSOR-LANE-REVIEW-1`, `E-FISHERZ-CANONICAL-COSINE-REPLACEMENT-1`, `E-H268-PROBE-WAVE-1-RESULTS`.
+
 ## TD-RUNG2-144-VOCAB-SPLIT — the two 144s of rung 2 are divergent vocabularies (2026-07-15)
 
 **Open.** `E-RUNG-CONTENT-LADDER-1`'s rung-2 row cites `sigma_rosetta.rs`