probe: h268-probe-wave-v1 — WH-MAG (NEUTRAL), SIG-CHECKSUM (PASS), WALK-SPECTRUM (KILL of §10(g) decorrelation half)#702
Merged
Conversation
…M specs + board index Probe wave converting three cheap conjectures from the merged H.268 arc into measured verdicts (CPU-only, no new deps): PROBE-WH-MAG (bgz-tensor, WHT16 magnitude preconditioner vs direct i4+i2), PROBE-SIG-CHECKSUM (jc, depth-2 signature as replayable trajectory digest behind hambly-lyons), PROBE-WALK-SPECTRUM (helix, stride-4-mod-17 autocorrelation vs PRNG at palette-lattice periods). Pass/neutral/kill criteria pinned per probe; INTEGRATION_PLANS prepended. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…irect i4+i2 on tile classes Per .claude/plans/h268-probe-wave-v1.md P1: #[cfg(test)] mod probe_wh_mag with a WH round-trip self-check (tolerance 1e-5, following ndarray's own wht_f32 round-trip precedent — plan said 1e-6, loosened with in-code rationale) and the three-class ratio table (256 seeded tiles/class, dim=16, SplitMix64, no new deps). Verdict deliberately unasserted — adjudicated externally per plan. Measured (informational): gradient+spike B/A=0.9290, heavy-tailed B/A=1.4113, uniform-noise B/A=1.8689. Crate suite 202 passed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…vs LCG at lattice periods Per .claude/plans/h268-probe-wave-v1.md P3: test-only module, N=4096, plain circular autocorrelation, seeded LCG baseline, no new deps. Measured (informational, adjudicated externally): period-17 harmonics |R| = 0.998/0.996/0.994 (exact wraparound formula R(17m)=1-8m/N, hand-derived and confirmed); palette-lattice periods walk max |R| = 0.875 (tau=64) vs LCG lattice max 0.0205 — 42.7x, far past the plan's 3x KILL threshold. Derivation note recorded in-module: gcd(17,tau)=1 does NOT imply low autocorrelation for an exactly-periodic parity waveform — the period-17 pattern has strong sidelobes at every lag's phase class. The "known period-17 structure" half measures true; the "decorrelated from the lattice by construction" half measures false for the raw residue-parity sign mapping. Full helix suite green (74 lib + 4 integration + 7 doctests). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…ctory digest Per .claude/plans/h268-probe-wave-v1.md P2: new sig_checksum module behind the hambly-lyons feature (probe, NOT a 12th pillar; prove() registry untouched). Thresholds taken verbatim from hambly_lyons.rs (FORWARD_TOLERANCE=1e-9, CONVERSE_THRESHOLD=0.05). Results (informational, adjudicated externally): replay_identity = exact bit-equality; tree_like_edit_invisible = 0.0 distance; non_tree_edit_caught = 11.3137 (level-2 signed area is the signal; level-1 delta = 0). Math finding recorded in-module: at DEPTH=2 a displacement PARALLEL to the neighbor chord is exactly signature-invisible (bilinearity identity) — the depth-2 digest's null space exceeds tree-like equivalence; the probe scans for a non-degenerate index and uses the perpendicular offset. jc suite green with feature (64) and without (60, zero-dep default). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
…cts adjudicated WH-MAG NEUTRAL (bare-tile leg CLOSED not-transferring; PROBE-WH-MAG-2 named, deferred); SIG-CHECKSUM PASS with the depth-2 parallel-chord blind-spot bound; WALK-SPECTRUM KILL of the §10(g) "decorrelated by construction" half (period-17 structure half confirmed; D-QUANTGATE unaffected). Adjudicator independently verified R(17m)=1-8m/N, C(13)=-15 -> R(64) ~ -0.875, and the depth-2 null-space identity. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
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_124bd658-1a32-4669-a448-bbc16dd26e2d) |
|
Warning Review limit reached
Next review available in: 47 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 (5)
📝 WalkthroughWalkthroughThis PR adds three deterministic, CPU-only Rust probes: Hadamard-wrapped tile reconstruction, depth-2 signature checksums, and stride-4 mod-17 walk autocorrelation. Planning and board documents record their measured NEUTRAL, PASS, and KILL outcomes. ChangesH.268 Probe Wave
Estimated code review effort: 3 (Moderate) | ~25 minutes Possibly related PRs
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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 8
🧹 Nitpick comments (1)
.claude/plans/h268-probe-wave-v1.md (1)
24-25: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winRecord the deterministic generator parameters.
The plan requires fixed seeds but does not specify the actual seeds, walk start value, or LCG parameters. Add those constants—or reference their exact definitions—so the reported measurements can be independently reproduced from the plan.
Also applies to: 58-62, 84-86
🤖 Prompt for 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. In @.claude/plans/h268-probe-wave-v1.md around lines 24 - 25, Update the deterministic synthetic tile-field sections and their corresponding measurement sections to record or reference the exact generator constants: per-class seeds, walk start value, multiplier, increment, and any output transformation. Ensure the plan’s ≥256-tile generation procedure is independently reproducible without relying on unspecified defaults.
🤖 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/board/EPIPHANIES.md:
- Line 7: Update the probe references in EPIPHANIES.md to use
repository-root-relative paths by adding the crates/ prefix to
bgz-tensor/src/adaptive_codec.rs, jc/src/sig_checksum.rs, and
helix/src/walk_spectrum.rs. Preserve the referenced module and probe names
unchanged.
In @.claude/plans/h268-probe-wave-v1.md:
- Around line 9-12: Remove model identifiers from the documented workflow: in
.claude/plans/h268-probe-wave-v1.md lines 9-12, replace “Sonnet” and “Opus” with
neutral role names; in .claude/board/EPIPHANIES.md lines 1-3, replace
“Opus-adjudicated”; and in .claude/board/INTEGRATION_PLANS.md lines 12-13,
replace “Sonnet drafters + Opus adjudication,” preserving the stated
responsibilities.
- Around line 35-39: Update the PASS/NEUTRAL/KILL classification in the probe
plan to define mixed P1 outcomes explicitly: classify as PASS when any eligible
class has B/A < 0.9, KILL only when all classes have B/A > 1.1, and NEUTRAL for
every remaining combination. State the precedence clearly and preserve the
existing class-based thresholds.
In `@crates/bgz-tensor/src/adaptive_codec.rs`:
- Around line 502-518: Update gen_heavy_tailed so its 2–3 outlier assignments
always target distinct cell indices; track already selected indices or otherwise
resample collisions while preserving the existing outlier count and value
generation.
- Around line 634-659: The sanity validation around the ratios collection
currently permits NaN results, allowing unusable probes to pass. Update the
ratio assertion in the test’s structural validation loop to require each ratio
to be finite, while preserving the existing length check and diagnostic context
identifying the tile class.
In `@crates/helix/src/walk_spectrum.rs`:
- Around line 4-11: The walk-spectrum analysis and its palette-lattice
conclusion must not use a single seeded LCG sample as a principled noise floor
or significance threshold. Update the relevant measurement/reporting logic
around the spectrum calculation to derive and report the prescribed Jirak 2016
weak-dependence rate, or remove all noise-floor, threshold, and significance
claims and label the LCG result descriptive only.
- Around line 64-82: Update walk_residue to use the canonical CurveRuler by
constructing it with CurveRuler::from_place(start) and returning index(k as u32)
instead of reimplementing the stride/modulus calculation. Preserve walk_signs’
existing parity-based behavior and remove the now-unnecessary local arithmetic.
In `@crates/jc/src/sig_checksum.rs`:
- Around line 311-334: Add an explicit assertion after computing level1_delta in
the non_tree_edit_caught probe that it remains approximately zero, using an
appropriate floating-point tolerance. Keep the existing total-distance
assertion, and make the new failure message identify an unexpected level-1
change so the probe verifies discrimination comes from level 2.
---
Nitpick comments:
In @.claude/plans/h268-probe-wave-v1.md:
- Around line 24-25: Update the deterministic synthetic tile-field sections and
their corresponding measurement sections to record or reference the exact
generator constants: per-class seeds, walk start value, multiplier, increment,
and any output transformation. Ensure the plan’s ≥256-tile generation procedure
is independently reproducible without relying on unspecified defaults.
🪄 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: 577dbe99-780b-417d-9a16-8414825e8d29
📒 Files selected for processing (8)
.claude/board/EPIPHANIES.md.claude/board/INTEGRATION_PLANS.md.claude/plans/h268-probe-wave-v1.mdcrates/bgz-tensor/src/adaptive_codec.rscrates/helix/src/lib.rscrates/helix/src/walk_spectrum.rscrates/jc/src/lib.rscrates/jc/src/sig_checksum.rs
Comment on lines
+9
to
+12
| > Execution model per operator directive: Sonnet drafters (grindwork, | ||
| > edit-only + targeted own-crate `cargo test`), Opus filigree reviewer | ||
| > (adjudication vs pass/kill), main thread = spec, central gates, | ||
| > commits, PR, autonomous resolve + merge after clean CI/reviews. |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win
Remove model identifiers from all wave documentation.
The shared documentation contract prohibits model identifiers in file content, but all three new entries embed them:
.claude/plans/h268-probe-wave-v1.md#L9-L12: replace “Sonnet” and “Opus” with neutral role names..claude/board/EPIPHANIES.md#L1-L3: replace “Opus-adjudicated”..claude/board/INTEGRATION_PLANS.md#L12-L13: replace “Sonnet drafters + Opus adjudication”.
📍 Affects 3 files
.claude/plans/h268-probe-wave-v1.md#L9-L12(this comment).claude/board/EPIPHANIES.md#L1-L3.claude/board/INTEGRATION_PLANS.md#L12-L13
🤖 Prompt for 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.
In @.claude/plans/h268-probe-wave-v1.md around lines 9 - 12, Remove model
identifiers from the documented workflow: in .claude/plans/h268-probe-wave-v1.md
lines 9-12, replace “Sonnet” and “Opus” with neutral role names; in
.claude/board/EPIPHANIES.md lines 1-3, replace “Opus-adjudicated”; and in
.claude/board/INTEGRATION_PLANS.md lines 12-13, replace “Sonnet drafters + Opus
adjudication,” preserving the stated responsibilities.
…gate, canonical CurveRuler, level-1 assert, band precedence, descriptive LCG note Six of eight CodeRabbit findings on the probe wave applied (one is a false positive, one a nitpick folded in): - gen_heavy_tailed now resamples outlier-index collisions (2-3 DISTINCT cells guaranteed); only the heavy-tailed ratio moved: B/A 1.4113 -> 1.3169; verdict unchanged (NEUTRAL, bare-tile leg not-transferring). Reported numbers updated in the plan Results + EPIPHANIES entry. - probe ratio sanity now REJECTS non-finite values (NaN acceptance removed; invalid MSE also asserted at source). - walk_spectrum consumes the canonical CurveRuler::from_place().index() instead of reimplementing the stride arithmetic (consume-the-canon rule); measured table bit-identical. - walk_spectrum module doc relabels the LCG baseline DESCRIPTIVE ONLY — the verdict rests on the closed-form C(tau mod 17) structure, no significance claim made; future significance phrasing must use Jirak 2016 rates per I-NOISE-FLOOR-JIRAK. Plan carries the same scope note. - sig_checksum now ASSERTS level1_delta < FORWARD_TOLERANCE so the level-2-discrimination conclusion is verified, not just printed. - plan P1 verdict bands rewritten with explicit precedence (PASS if any of (a)/(b) < 0.9; else KILL if ALL > 1.1; else NEUTRAL incl. mixed) — the original bands left mixed outcomes uncategorized. - plan gains a reproducibility note pinning where every generator constant lives (the probe code is normative). - EPIPHANIES probe paths now repo-root-relative (crates/ prefix). Skipped with reason: the "remove Sonnet/Opus model identifiers" finding — lance-graph has no such rule (its own CLAUDE.md Model Policy uses the tier names); the no-model-identifier rule elsewhere in the workspace covers the assistant's model ID, which appears nowhere. All suites green after fixes: bgz-tensor 202, jc 64 (+60 featureless), helix 74+4+7. rustfmt clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
AdaWorldAPI
pushed a commit
that referenced
this pull request
Jul 16, 2026
…E v5 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
AdaWorldAPI
added a commit
that referenced
this pull request
Jul 16, 2026
…w-h9osnl docs(board): post-merge hygiene for #702 — PR_ARC entry + LATEST_STATE v5
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The h268-probe-wave: three deterministic, CPU-only probes converting conjectures from the merged H.268 arc into measured verdicts (plan:
.claude/plans/h268-probe-wave-v1.md; adjudication: internal Opus review, math independently verified). No new dependencies, no unsafe; graded probes assert only structural sanity so verdicts can't red the suite.Probes (code):
bgz-tensor/src/adaptive_codec.rs—mod probe_wh_mag: WHT₁₆ magnitude preconditioner vs direct i4+i2 on three seeded tile classes (256 tiles/class). NEUTRAL; bare-tile leg CLOSED NOT-TRANSFERRING (B/A = 0.929 / 1.411 / 1.869): WHT₁₆ spreads outlier energy tile-wide, inflating the per-cell quantization floor; the row-level win needs the Passthrough escape + centroid residual the probe deliberately isolates away. Shipped row codec untouched. Follow-up PROBE-WH-MAG-2 named, deferred.jc/src/sig_checksum.rs(behindhambly-lyons; NOT a 12th pillar) — depth-2 signature as replayable trajectory digest. PASS (replay bit-exact; tree-like excursion 0.0 < 1e-9; non-tree displacement 11.31 > 0.05) with a documented exact-identity bound: a parallel-chord interior displacement is signature-invisible at depth 2 — the digest's null space exceeds tree-like equivalence.helix/src/walk_spectrum.rs(test-only) — stride-4-mod-17 parity-sign autocorrelation vs seeded LCG. KILL of the "decorrelated from the palette lattice by construction" half of §10(g) (lattice |R| max 0.875 at τ=64 vs PRNG 0.0205 = 42.7× ≫ the 3× kill bar; hand-verified C(13)=C(4)=−15 → −0.882); the "known period-17 structure" half CONFIRMED (R(17m)=1−8m/N). D-QUANTGATE / the walk's bijectivity-quantization role unaffected.Board/docs: EPIPHANIES
E-H268-PROBE-WAVE-1-RESULTSprepended; plan file created with pinned PASS/NEUTRAL/KILL bands + Results filled; INTEGRATION_PLANS prepended. Companion ndarray PR carries the matrix-doc §10(g) correction + §5 results sub-table.Verification
--features hambly-lyonsand 60 without (zero-dep default intact); helix 74 lib + 4 integration + 7 doctests. rustfmt clean on touched files.🤖 Generated with Claude Code
https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki
Generated by Claude Code
Summary by CodeRabbit
Tests
Documentation