docs: H.268 addendum §7-§10 — comma closure, 96-bit facet carving, kernel-shape rule, replayable-tile synergies#243
Conversation
…rnel-shape rule, replayable-tile synergies Extends pr-x12-h268-morton-wgpu-synergies.md with four sections per the new plan file (.claude/plans/H268-comma-96bit-replayable-addendum-v1.md): - §7 comma closure: Pythagorean-comma/X-Trans anti-moire framing; the coprime-integer surrogate (helix CurveRuler stride-4-over-17); D-QUANTGATE rationale restated to its three real legs (libm cliff, WGSL floats, bijective closure) with the "floats round differently" leg explicitly withdrawn (std::f64::consts receipts, five-backend bit-identical contract). - §8 96-bit facet carving: 48 CAM-PQ + 24 helix ResidueEdge + 24 turbovec = the V3 12-byte content-blind payload; ClassView as carving/LUT selector; Signed360 as out-of-row variant; 6x256^2 CAM-PQ vs bgz17 table-family clarification. - §9 kernel-shape rule: VNNI/AMX for matmul-shaped, LUT/texture for lookup-shaped (turbovec NativeLut 11.4x measured receipt); ITU claim scoped to compute kernels only. - §10 replayable-tile synergies: the 4x4 Morton tile serving H.268 mode-decision AND cognitive-shader dispatch — nine consequences, all probe-gated; no kill condition weakened. Old §6 Cross-references renumbered to §11; blackboard entry appended; knowledge-doc test suite green (117/117). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
|
Warning Review limit reached
Next review available in: 33 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (3)
📝 WalkthroughWalkthroughThe PR adds an H.268 addendum plan and records its comma-closure, 96-bit facet-carving, kernel-shape, and replayable-tile sections in related knowledge and blackboard documents. ChangesH.268 Addendum
Estimated code review effort: 2 (Simple) | ~10 minutes Suggested reviewers: Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_12216f05-e457-47a8-8b39-e0e62b0809e6) |
Review findings applied: - P1: the "6x256^2 clarification" paragraph in §8 was wrong on both arithmetic and attribution — a 256x256 u16 table is 128 KB (not 64 KB), and the 388 KB SpoDistanceMatrices benchmark is the PALETTE structure (3 S/P/O planes x 128 KB = 384 KB, palette_distance.rs:145-158), not CAM-PQ. CAM-PQ has no fixed 256^2 distance-table footprint; its ADC tables are per-query 6x256 f32 = 6 KB (cam_pq.rs:76-84). Corrected in the knowledge doc, the blackboard entry, and the plan file's B2 bullet (with the original wrong claim preserved as a stricken note). - P2: the non-existent std-SIMD const path is documented in helix constants.rs:17-23 as `const::simd::*`, not `std::simd::const::*` — spelling aligned with the source. - P2: added the le-contract.md §3 cross-ref for the ClassView carving granularities. Knowledge-doc suite still green (117/117). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 69d4aa3b20
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| **6×256² clarification (do not conflate the two table families):** the "six | ||
| 256×256 tables" belong to **CAM-PQ** — 6 subspaces × one 256×256 u16 | ||
| distance table each = 6 × 64 KB = **384 KB**, closely matching the measured | ||
| 388 KB `SpoDistanceMatrices` benchmark footprint (§1 above). **bgz17's |
There was a problem hiding this comment.
When readers use this addendum to plan the GPU-LUT path, this reverses the table families: the shipped CAM-PQ path precomputes only 6 × 256 per-query distances (src/hpc/cam_pq.rs:76-84, 6 KB as f32), while the 256×256 u16 shape belongs to the palette/SPO distance-matrix family (src/hpc/palette_distance.rs:145-154). Describing CAM-PQ as six 256×256 tables creates a nonexistent 384 KB CAM-PQ asset and contradicts the earlier row that says the CAM-PQ path is six lookups, so downstream probe/memory planning will be routed to the wrong structure.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Confirmed and fixed in 33b8807 (an internal pre-review caught the same reversal concurrently). The paragraph now attributes the 388 KB benchmark to the palette SpoDistanceMatrices (3 S/P/O planes × 256² u16 @ 128 KB = 384 KB, palette_distance.rs:145-158), states that CAM-PQ has no fixed 256² footprint (per-query ADC tables are 6×256 f32 = 6 KB, cam_pq.rs:76-84), and keeps bgz17's one-table-per-palette shape separate. The same correction is propagated to the blackboard entry and the plan file's B2 bullet.
Generated by Claude Code
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md:
- Around line 123-125: Update the description of the X-Trans pattern in the
surrounding passage to characterize its 6×6 arrangement as
“less-regular/quasi-random” rather than “non-repeating,” while preserving the
comparison to Bayer and the anti-aliasing discussion.
In @.claude/plans/H268-comma-96bit-replayable-addendum-v1.md:
- Around line 80-82: Qualify the “bit-exact replay” claims in
.claude/plans/H268-comma-96bit-replayable-addendum-v1.md lines 80-82 and
.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md lines 290-293 to apply
only to the proven CPU/wasm integer execution path, unless a deterministic
floating-point/backend probe is added; align both statements with the existing
integer-only GPU caveat.
- Around line 83-86: Grade the period-17 weak-dependence claim as [H] and mark
it probe-gated, adding the required statistical test for validating dependence
or correlation against the relevant baseline. Update the claim at
.claude/plans/H268-comma-96bit-replayable-addendum-v1.md:83-86 and mirror the
same qualification and probe requirement at
.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md:294-298; do not treat the
permutation self-test as sufficient evidence.
- Around line 101-108: Update the ndarray landing-map entry to match the
delivered synthesis diff: either include the promised §5 probe-queue changes in
the landed scope, or remove “probe-queue additions” from the Files description.
Ensure the execution record accurately lists only the sections and artifacts
changed by the supplied diff.
- Around line 67-75: Qualify the address-derived phase claim so it describes
phase-side seekability, not CABAC random access, unless independently
framed/reset decoder regions are available. Update the anti-CABAC wording in
.claude/plans/H268-comma-96bit-replayable-addendum-v1.md lines 67-75, add the
required A8/context-reset precondition in
.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md lines 267-271, and apply
the same qualified wording in .claude/blackboard.md lines 758-763.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 1e9aed8c-01b9-40dd-b948-365c9f99bb42
📒 Files selected for processing (3)
.claude/blackboard.md.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md.claude/plans/H268-comma-96bit-replayable-addendum-v1.md
… A8-gated seek, integer-path replay, [H] dependence, landing-map scope) Operator refinement of the §8 table-family paragraph — three flavours of 256: 1. CAM-PQ = 6x256^2 compressed to 6x256: latent per-subspace 256^2 centroid-distance families, shipped as per-query 6x256 f32 ADC rows (6 KB, cam_pq.rs:76-84), never materialized. 2. bgz17 = the explicit 256^2: one materialized 256^2 u16 table + kxk u8 compose per palette; the 388 KB SpoDistanceMatrices benchmark is this flavour on 3 S/P/O planes (3x128 KB = 384 KB). 3. V3 facet = explicit 6x256^2 as ADDRESS, codec-agnostic: 6x(u8:u8) rails = six coordinate pairs into 256^2 spaces = 96 bits; classid -> ClassView switches which codec's 256^2 family each rail indexes. Plus five external-review qualifications across knowledge doc / plan / blackboard: - X-Trans corrected to "repeating but less-regular 6x6 tile" (was "non-repeating"). - (a) reworded to phase-side seekability — CABAC's serial context chain remains; bitstream-level seek requires A8's framed/context-reset regions. - (f) bit-exact trajectory replay scoped to the proven CPU/wasm integer path; float/GPU stages outside the claim. - (g) period-17 "friendlier to Jirak" graded [H, probe-gated] — the permutation self-test proves bijectivity, not decorrelation; needs a correlation-spectrum probe vs a PRNG baseline. - Landing map aligned with the delivered diff (no §5 probe-queue changes were needed). Knowledge-doc suite green (117/117). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…, Hambly-Lyons [S] anchor Appends the operator-directed honourable mention to the three-flavours block: bgz-tensor's AdaptiveRow (palette centroid index + Hadamard-rotated residual, i8 outlier / i4+i2 cascade) as the index-PLACES/residual-CORRECTS continuous-field exit; out-of-row like Signed360, in-row refinement stays the turbovec nibble lane. Formal anchor graded [S]: Hambly-Lyons 2010 signature uniqueness, analogy-grade until a ladder->signature probe exists. Canonical text lives on the lance-graph V3 substrate (le-contract.md §3 + E-PALETTE-RESIDUAL-LADDER-1). Knowledge-doc suite green (117/117). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…llar 11; only the ladder->signature mapping stays [S] Follow-up to merged #243 (branch restarted from master). The §8 fourth-mode sentence said the Hambly-Lyons anchor was analogy-grade "until a ladder->signature probe exists" without noting the theorem side already ships in both repos: - THIS repo: src/hpc/pillar/signature.rs (Pillar-11 B7 — signature transform + Hambly-Lyons sig-kernel, Gram-matrix PSD check, certification probe over 1,000 Levy paths). - lance-graph: crates/jc/src/hambly_lyons.rs (Pillar 11, feature hambly-lyons -> sigker; forward/converse probes on signature_truncated). Amended so only the ladder-levels -> signature-levels MAPPING for the residual cascade stays [S], with its probe building on the existing Pillar-11 harnesses. Blackboard note appended. Mirrors lance-graph ec58954 (PR #700). Knowledge-doc suite green (117/117). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
2 participants
Summary
Doc-only addendum extending
.claude/knowledge/pr-x12-h268-morton-wgpu-synergies.md(merged in #242) with four new sections, per the new plan file.claude/plans/H268-comma-96bit-replayable-addendum-v1.md:CurveRulerstride-4-over-17, full permutation, banned Fibonacci-mod-17 pathology). D-QUANTGATE rationale restated to its three real legs (libm non-portability, WGSL floats not IEEE-pinned, bijective closure) — the "float constants round differently" leg is explicitly withdrawn with receipts (std::f64::consts::{GOLDEN_RATIO, EULER_GAMMA}compile bit-exact on the pinned toolchain, φ=0x3FF9E3779B97F4A8 / γ=0x3FE2788CFC6FB619; no std-SIMD const-constants path exists — the assumed spelling wasconst::simd::*, per helixconstants.rs:17-23; the five-backend unfused bit-identical GEMM contract + wasm parity CI). φ-PLACES / walk-QUANTIZES / γ-CORRECTS division of labor. x264 vs AV1 vs address-derived contrast stays [H] behind the J2 fence.ResidueEdge+ 24 turbovec (6×4-bit) = 96 bit = exactly the V3 12-byte content-blind payload;ClassViewas the carving/LUT selector (cross-ref le-contract.md §3);Signed360as the out-of-row variant. Table families disambiguated (corrected post-review, commit 33b8807): the 388 KBSpoDistanceMatricesbenchmark = the palette structure (3 S/P/O planes × 256² u16 @ 128 KB = 384 KB,palette_distance.rs:145-158); CAM-PQ has no fixed 256² footprint — its ADC tables are per-query 6×256 f32 = 6 KB (cam_pq.rs:76-84); bgz17 = one 256² u16 + k×k u8 compose per palette.Old §6 Cross-references renumbered to §11. Blackboard entry appended. Companion lance-graph PR: AdaWorldAPI/lance-graph#698 (E-H268-REPLAYABLE-TILE-1 + #697 post-merge PR_ARC entry).
Review
An internal filigree review (overclaim/receipt/grade audit) ran before ready-for-review; its P1 (the §8 table-family paragraph was wrong on arithmetic and attribution) and both P2s (const-path spelling, le-contract cross-ref) are fixed in commit
33b8807. All other receipts verified against source, including an independent recomputation of the φ/γ bit patterns and the 11.4× turbovec measurement (lib.rs:39-47).Verification
python3 -m unittest tests.knowledge.test_pr_x12_knowledge_docs).tropical_spmv/scalar_sparseusages).constants.rs/residue.rs,cam_pq.rs,palette_distance.rs, turboveclib.rs).🤖 Generated with Claude Code
https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki