[AdaWorldAPI]

Operator-authorized lab probe for D-RU-6 (d-relalg-unified-v1.md §8, §10, §14 U6). Lab only. No production path changes, and this PR is not an ADMISSION B demonstration. D-RU-6 stays Queued/Lab. Please don't merge on green CI alone.

What it is

crates/crepe-plan-probe/examples/walk_tile.rs is a separate example. It evaluates crepe's graph_walking program (Walk(x,z,len) for len in 0..=50, NoWalk, set semantics, least fixpoint) over 256-slot RMAP foci and 256×256 Boolean tiles. walk_mask (#1637, #1642) runs beside it as the measured baseline.

Reuse, not a new engine:

• mork-pathmap-probe::rmap::{Focus, Witness, Refusal, tile_addr};
• a mask-risc Ternlog program for the tile product;
• tile_gemm::BitMat::compose as an independent kernel oracle;
• walk_mask's oracles, fixtures and executor. walk_mask.rs is compiled in as a module, and its only change is pub(crate) on the items reused; it has no behaviour change and its 8 tests still pass.

How it runs:

• Hop: a predicated Boolean tile product. For each coordinate y, C |= broadcast(row y of A) & spread(bit y of each row of B). Every address is a loop coordinate.
• Paging: more than 256 nodes are paged across foci, and cross-page edges are counted.
• Refusals: too many foci, a scope that misses reached ids, a stale witness (focus generation, facts or seal), and any program containing MaskOp::Gather are all refused.

Tests: 23 (8 of them are the baseline's)

• Tuple sets equal ascent, crepe and walk_mask on: mixed lengths {1,2,3,5,7}, cycles, alternative path lengths, zero-length and over-length edges, ids that are not nodes, and 255/256/257/300 nodes across foci. Also the benchmark generator over two foci, and a sparse graph over three.
• Recursion depth: a chain that needs ≥ 200 iterations (230 measured), and a zero-length 200-ring that terminates.
• Kernel: the tile product against BitMat::compose.
• Negative tests: a stale witness, an incomplete focus (cap and scope), a wrong path length (counts equal, the tuple oracle catches it), and an unauthorized gather.

CI gains cargo test --example walk_tile. The benchmark itself is not run in CI.

Disable runs (all red, each reverted)

Disable	Red
admit lets Gather through	the gather test
witness check removed	the stale test
escape refusal removed	the scope test
page cap removed	the cap test
length bound exclusive	6 tests
cross-page products dropped	2 tests
spread reads y ^ 1	≥ 8 tests
decode shifts a length	≥ 8 tests

Measured (release, 4 vCPU, load < 1.3, medians of 7)

fixture	foci	mask count-only ms	tile count-only ms	tile/mask	mask + decode ms	tile + decode ms	tile/mask	oracle
edges(128)	1	0.43	0.72	1.7×	4.14	4.54	1.1×	equal
edges(256)	1	1.28	1.32	1.0×	15.8	18.2	1.2×	equal
edges(512)	2	3.07	8.95	2.9×	68.3	74.7	1.1×	equal
edges(1024)	4	12.7	72.0	5.7×	277	384	1.4×	UNCHECKED (tile = mask)
chain(230)	1	7.43	27.6	3.7×	16.8	33.6	2.0×	equal
mixed(256)	1	8.65	80.0	9.3×	298	568	1.9×	equal

• Time split for the tile path. Tile count-only covers every phase except decoding. At edges(1024) the phases are, in ms:

  attest	validate	bind	project	recurse	count	decode
  0.84	< 0.001	0.80	1.74	67.6	0.16	318

  Binding costs about 4.7% of the count-only time; recursion costs 94%.

• Memory. Resident growth was measured as VmHWM in a fresh process per cell. In-process heap counting is impossible here because PathMap enables jemalloc as the global allocator. At edges(1024), count-only resident growth is 14.2 MiB for walk_mask and 2.2 MiB for tiles. Only non-empty tiles exist: 1.28 MiB, against 19.1 MiB of dense planes. Decoding erases the difference: 50.7 vs 44.9 MiB.

• Value-derived reads at edges(1024): walk_mask execution makes 419,230. The tile path makes 0 in execution, 83,846 codebook lookups in preparation, and 4.04M codebook decodes in decoding.

Verdict. Co-addressed execution removes the value-derived reads but does not keep the mask approach's speed. It is 1.0–9.3× slower count-only and 1.1–2.0× slower end to end. End to end looks closer only because tuple decoding dominates both engines. The loss is reported as measured: no measured cost was optimized away.

ADMISSION B classification

Step	Classification
attestation, codebook build	same-table sequential scans, counted
slot resolution	one frozen-codebook lookup per edge endpoint, keyed by the edge's own value. This is the shape the ruling admits, but the witness is the probe-local rmap::Witness digest, not the attested sorted key the ruling names. Not demonstrated.
edge tiles	projection of the same table
recursion	co-addressed. Demonstrated (refusal guard, disable-verified).
decoding	frozen-codebook decoding, counted

Board

• New entry D-RU-6-LAB-1 (entries/2026-10-21-d-ru-6-tile-walk-probe.md).
• The D-RU-6 row stays Queued/Lab and now carries the benchmark requirement (refs D-RMAP-2 §5b, crepe-plan-probe: walk_mask tuple oracle, UNCHECKED above oracle size, CI fixtures #1642). Plan §8 carries the same requirement.
• The entries index is regenerated; the supersession index is unchanged.
• No new D-id.
• Not touched: D-CMR-1, relalg normalize, its rewrite catalogue, mork-pathmap-probe sources. File ownership is claimed on EXO-50.

Open

• A contract-level attestation for the codebook, before any ADMISSION B claim.
• Whether a block-sparse co-addressed product can close the count-only gap without a value-addressed read.
• Identity-scale paging (65,536 nodes) is not measured.

---

Generated by Claude Code

[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: 6619723c-8052-46e8-804c-8f75c57702cc

• 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 2026-10-21T01:05:38.841444Z	ce4997a	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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce4997a6dd

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".