[AdaWorldAPI]

This PR covers item D of the post-merge review of #1681 (GPT:gpt-ada-01/20261021-pr1674-1682-consolidated-review-01). I did it on the operator's instruction (EXO-50 /038). It is measurement only: compose.rs, r2il and mask-risc are unchanged.

Files

• crates/dir-rmap-probe/examples/rft1_cost.rs, the instrument (new)
• .claude/board/entries/2026-10-21-d-rft-1-cost-breakdown.md, the findings (new), plus the regenerated entries index

Findings

Fixture: Spec::small(), 600 users on 3 pages (Spec::iam(8000) exceeds 256 slots). Release build, 4 cores, load around 2 during the runs. The ranges below cover three runs.

item	measured
(1) preparation vs execution	Composer::new takes 0.8–1.4 µs; arm A takes 212–270 µs per pass
(2) execute_into calls per pass	2769 = present 600 + union 600 + path 1374 + fan 192 + plane_count 3
(5) support-first pruning	1374 path folds out of 38 400 coordinates (3.6%)
(3) ns per call	path count_and(3) 54–70; present keep(intersect 2) 48–62; support keep(union 4) 101–129; fan count_and(1) 95–121; plane_count compile + scratch 319–400 ns per page
(2)×(3) vs measured	the prediction covers 87% of arm A; path folds are 41% of the prediction
(4) four-word mask cost	54–70 ns per call through mask-risc vs 2.2–2.8 ns for the same work in ndarray::simd, i.e. about 52–67 ns of fixed overhead per call
(6) allocations	39 per pass; 0 per prepared execute_into; 18 per pass from compiling plane_count inside compose
(7) batching upper bound	at most 72–92 µs could be removed (33–34% of arm A)

Reading. Fixed per-call overhead dominates arm A, not the mask work itself, so the lever is fewer invocation boundaries.

Not built. A batched multi-coordinate fold (a count product or a positional popcount) needs its own admission decision and falsifiers. It must also keep the Bag/Distinct, NULL, witness and refusal contracts.

The three comparisons (A, C, C+) keep the definitions from rft1_bench.

Open

• Unexplained: count_and(1) over one plane costs more per call than count_and(3) over three, in every run.
• Unattributed: 13% of arm A.

Gates

Clippy --examples -D warnings and fmt are clean. All results are in ns or µs; nothing is in milliseconds.

I will not merge this PR; it waits for your review.

---

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: 4cf72ea5-a7f3-46fe-8803-b98f9f0d2e65

• 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 2026-10-21T22:27:20.561577Z	30f6418	PR opened

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: 30f6418a8c

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".