D-HOLO H6b measured; H6c pre-registered

[AdaWorldAPI]

What

Runs the pre-registered H6b falsifier (plan d-holo-address-indexed-algebra-v1.md) and pre-registers H6c, which tests H6b's three open terms.

The new example crates/lance-graph-mask-risc/examples/holo_tile_forms_probe.rs (test = true) compares four set representations on the 256 × 256 Context × Operation tile. Each is checked cell by cell against a row oracle:

• P, product form: contexts grouped by identical operation mask
• I, Morton interval unions
• D, the dense 8 KiB mask, using ndarray::simd mask ops
• H, hybrid: P while smaller than D, else D

There are 14 patterns, including the doughnut amendment.

H6b results (v4, avx512f, three runs)

Rule	Measured	Verdict
Structured patterns: P or H total ≤ 0.5× D at N = 100	P 0.64–1.28, H 0.71–1.01	not met, undecided
Adversarial patterns: H total ≤ 1.25× D	checkerboard and random 50 % pass; fragmented 1.19–1.33	two pass, fragmented undecided
Component count predicts better than density	components +0.34 to +0.55, density −0.07 to +0.12	components ahead, weak

• Construction dominates the totals. Each build sweeps all 65,536 cells, about 0.3 ms. Per query alone, P is 0.21–0.73× D on the structured patterns.
• The hybrid's byte switch is refuted. It keeps P on the sparse and dense patterns (60 groups), where H costs 1.8–4.8× D.
• P's cost follows its group count, not its bytes (post hoc Spearman +0.81 / +0.88).

H6c (pre-registered here, not run)

• Q1 — crossover in group count. Fixtures with g = 1..64 groups in two layouts, blocked and interleaved. Prediction: P/D per query crosses 1 between g = 4 and g = 16 in both layouts.
• Q2 — group-count switch. A hybrid H' that switches on group count. It counts as a safe default only if it stays at or below 1.25× D everywhere and is never 2× or more above D.
• Q3 — construction from a declaration. Every arm, D included, is built from the same declaration, then H6b's structured rule is rerun.
• Test-first disables. Declared build == swept build (off-by-one must fail); H' switches both ways; each fixture really has g groups.

Tests

• 5 oracle and invariant tests pass.
• Three disable runs, each red.
• clippy -D warnings and fmt are clean.

Board

• New entry 2026-10-14-d-holo-h6b.md.
• STATUS_BOARD rows D-HOLO-H6b (measured) and D-HOLO-H6c (queued).
• AGENT_LOG: the four substrate-mapper runs.
• The owed Model policy: six supervised Haiku cards for harvest, deep mapping and spec edits; record #1551 #1552 entries in PR_ARC_INVENTORY and LATEST_STATE.

🤖 Generated with Claude Code

https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 026c5689-b88f-457c-8786-0b98230c09de

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-14T20:20:12.248045Z	c25573e	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.