[AdaWorldAPI]

What

D-RFT-1, PROBE (crates/dir-rmap-probe/src/compose.rs). It composes user → group membership with the role carriers over the DR256 register's shared 256-slot coordinate, under one Witness.

• Path mask: per (user i, role r), member_i ∧ valid_i ∧ carriers_r.
• Bag is the path count and Distinct is its existence. They are two readings of one computation.
• Stratum comes from the measured cardinality witness: the left fan (present keys per user) and the right fan (carrier slots per role). The caller does not choose it.
• NULL key: payload set, presence clear. It matches nothing, the same model as Quack's validity plane.
• Refused: XOR (an even number of parallel paths reads 0), seal mismatch, stale binding, and bag overflow (checked add).

r2il polyfill (lance-graph-r2il):

• IntXor (R2IL ordinal 22) lowers to MaskOp::Xor, or to BoolExpr::Xor when fused.
• ANY and ALL lower to Terminal::Any/All and push 1 or 0. They were Unimplemented before.
• New procedures: exists, all, differ (IntXor then ANY, mask equality as a fold), intersect, union, keep, count_and.

How it executes. Every count, existence and mask equality in the composition is one of those procedures. Each is compiled once by FoldDialect (Composer::new, a prepared statement) and replayed through mask-risc execute_into over zero-copy row slices. The probe has no counting loop of its own.

Support first. A user's roles are the union of its present keys' role words. Only coordinates in that union are counted; the others have no path by construction.

Evidence

• Tests (tests/compose.rs):
  • IAM diamond: Bag 2, Distinct 1.
  • Refusals: XOR, NULL key, seal mismatch, stale binding, overflow.
  • Functional stratum.
  • A random fixture against an independent oracle: a DFS over the Prepared CSRs, not Qr's closure. Compared lane to lane, with supports compared by differ; anti-vacuity on counts 1 and ≥ 2.
• Disables: 7 on compose and 5 on r2il, all red.
• PostgreSQL 16.14 (examples/rft1_pg_parity.rs, WITH RECURSIVE):
  • the per-coordinate lane is equal over 600 users × 64 roles;
  • bag 1421 = count(*);
  • distinct 1374 = count(DISTINCT (u, role));
  • NULL-group membership rows match nothing in either engine.
• Bench (examples/rft1_bench.rs, release, median of 101 runs, 600 users):

step / arm	time	allocations per run
bind	46 µs
pages	39 µs
program prep	1.4 µs
A — composition (r2il → mask-risc)	310 µs	39
C — every coordinate, direct ndarray::simd	131 µs	0
C+ — A's plan, direct ndarray::simd	14 µs	0
B — tuple baseline, lab only, not a candidate	134–146 µs	10

The first wiring ran one fold per (user, role) coordinate and took 17 ms with 78 210 allocations. Program replay plus support-first brought it to 310 µs.

Open

• Per-execution cost. A − C+ is about 115 ns of fixed cost per mask-risc execute_into on 4-word planes (about 2 600 executions). Not profiled.
• No count-product terminal. mask-risc has no terminal that produces many coordinates' counts in one execution (a bit-matrix count product or a masked positional popcount), and ndarray has no positional popcount.
• Presence plane is a copy. Members::of copies member into an owned valid plane (8 KiB per page). A real source would supply its own presence plane.
• Fixture scale. Spec::iam(8000) exceeds 256 security slots, so the bench runs Spec::small().

Manifests

• dir-rmap-probe adds lance-graph-r2il and moves lance-graph-mask-risc from a dev-dependency to a dependency.
• OGAR patches are added for ogar-loco and ogar-r2il, plus a lance-graph patch so the contract resolves once.
• The CI workflow paths now include r2il and the contract.

No production ABI changes.

Board: entries/2026-10-21-d-rft-1-typed-composition.md, and the STATUS_BOARD D-RFT-1 row (In PR).

---

Generated by Claude Code

[AdaWorldAPI]

@codex 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: 0aff59e5-3ccf-42b7-b95d-d51672e94dd2

• 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 2026-10-21T20:48:12.587945Z	f67e9c5	Manual request

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: 3a5d41f786

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".

[AdaWorldAPI]

Rebased onto main and fixed the two red checks. Both were mine:

• an entries-index duplicate id: the entry's heading id is now D-RFT-1-COMPOSE;
• citation decay: my r2il insertions moved pub struct FoldDialect, and the metta-datalog plan now cites it by symbol instead of by line number.

New in this push: the PostgreSQL primitive card in lance-graph-r2il. The full table is in the D-RFT-1 entry's update section.

Dialect:

• KEY_RUNS lowers to CountKeyRunsU32, i.e. COUNT(DISTINCT key) on a key-ordered lane. An unordered lane is refused.
• Scalar IntAdd, IntMult and IntSDiv are checked. Overflow and division by zero are refused as FoldError::Overflow and FoldError::DivisionByZero.

Procedures: is_null, is_not_null, and_between_i32, in_u32, count_valid, avg_parts, bool_or, bool_and, not_exists, except, count_distinct_sorted, count_union_all. A kept WHERE mask is the Lance 13 with_row_addr_prefilter input, as in #1633.

Checks: 6 more disables, all red. r2il has 28 tests; clippy -D warnings and fmt are clean; the r2il-mask-abi-probe and the dir-rmap compose suites are green.

@codex review

---

Generated by Claude Code

[AdaWorldAPI]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: f67e9c5a7f

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".