R2IL wave: the chain carrier wins, the vocabulary survives optimization, and the boundary that qualifies both

[AdaWorldAPI]

Autoattended 6-slice wave: 4 Sonnet probe workers, 1 Sonnet scribe, 1 Opus synthesis worker, 1 Haiku guarded executor, 1 read-only Opus meta-reviewer. Orchestrator compiled centrally, adjudicated every gate, applied every fix, and is the sole writer of all board files.

The named next measurement was blocked. Wider corpora need r2sleigh to lift new binaries and that sibling is absent from this checkout. Re-planned around what the existing corpus can actually answer: it holds two builds of one source at different optimization levels (71 fns / 3040 ops vs 72 / 2300, disjoint address keys) — a real held-out split.

Findings

F-10 — the def-use chain carrier confirms #1014's prescription as code. 1,872 length-3 chains collapse to 27 signatures against the window's 97 (3.6× tighter); top-10 occupancy 0.887 vs 0.505; and 95.9% of the top chain's occurrences skip at least one intervening op (median span 6, max 148) — invisible to a window matcher at any width the data would justify.

F-9 — the idiom vocabulary survives optimization completely. 10/10 top-K transfer; 61 of 64 TRAIN trigram types survive; TEST carries 94, of which 33 are TEST-ONLY. The optimizer added to the vocabulary while cutting ops/function 42.82 → 31.94. The partial-survival pre-registration was refuted and recorded in place.

F-11 — the boundary, and it qualifies the others. The convention's residual (36,747 count units) is of comparable-or-larger magnitude than its classified output (17,557 rows), 88.1% one named reason, conservation exact. Consequence: F-1, F-2 and F-10 are claims about a seven-opcode projection of two binaries, not about x86-64.

F-12 — the Stamp loss curve. 0 at N≤64, 33.3% at 96, 55.2% at 143, 87.5% at 512. Modelled wider registers recover 15/0 dropped at N=143 — modelled only, not a proposal, costs unmeasured. Any width change is the operator's ruling.

The meta-review changed the findings, not just the code

Verdict FIX-THEN-LAND, with 4 P0s — three gates that no input could fail (one of which labelled a true-by-construction identity "the falsifier for this whole section"), plus this arc's own knowledge-doc header stating F-1 as an unqualified claim about "real machine code" that the same document later contradicts. Also caught: an unchecked prov_op_site uniqueness assumption underpinning every span figure, and a gate awarding PASS with no assertion at all (an all-comment TSV would have passed it).

All P0s and the load-bearing P1s fixed; every gate re-run green with strictly harder assertions (top1*2 < total, distinct*10 < total, frac > 0.5, n > 64, and a shared reason_for join now exercised by both the loop and its can-fire).

Two findings were weakened by the review and are recorded weakened: F-10 drops "prescription confirmed" — the chain carrier's over-admission is 0 by construction, so what remains is an uncontrolled concentration comparison with no occurrence-matched control; F-11 gains an explicit units caveat (ore rows vs residual count units, never established equal).

Process notes

• A worker caught an error in its own brief — the slag by_address section is 4 columns, not the 5 I specified — and parsed what the file declares rather than what it was told.
• The Haiku executor stopped at command 1 on a formatting failure, wrote its receipt, attempted no fix, and escalated. Contract behaviour. Supervisor fixed and re-carded: 8/8 GREEN, including the deliberate corpus-absent guard (exit 2, the never-fabricate path proven live).
• Anti-fabrication rule in every brief: workers cannot run anything, so they know no number — every gate had to be a relation with the value printed. All four complied.

Also in this PR

• .claude/knowledge/r2il-behavioral-carrier.md — the consolidated reference (F-1..F-12) so the next session loads it instead of rediscovering the arc.
• PR_ARC gap-fill for Board hygiene owed by #974 and #975 #977..charter: BELIEF-ABI-RESTORATION-1 — only the residue deserves a new tenant #1005 from the scribe's primary-source reconstruction. Withdrawn (repository-boundary violation) #976 is absent from history entirely; Plan: alpha-channel rung overlay — the empty row of the thinking table #978/Handover: alpha-channel plan session (post-#978) #979 landed as direct commits; one figure discrepancy with Two measured corrections: the DisMech oracle population, and the academic carve fill #975 is flagged unadjudicated — it belongs to the session that owns that arc.

---

Generated by Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[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: Pro Plus

Run ID: f0e35f20-187c-4217-b499-288fdcab3bbc

---

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.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot 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_d85861b9-97fc-4db2-9306-5605746094fb)