R3 gate #51: cross_target_optimization_constant_fold_consistent

[briansrls]

Auto-opened by session-dashboard for session smart-cat-457.
Pushing to session/smart-cat-457 advances this PR.

Closes #2575

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[briansrls]

Codex review (gate #51 tautology / intent) — addressed in 0c0ecc42a (on top of prior runner work)

Findings were valid on the first cut:

1. Distinct workflow roots — analyze_symbolic_cost_dimension now runs on fold_pre vs fold_post (fixture program: 1 + 2 vs literal 3), not the same NodeId twice.

2. Non-tautological fold witness — raw composed differs (ConstantCost(1) vs ConstantCost(0) at HEAD for this program). The consumer now requires the structural relationship backed by LanguageSpec rows: normalize(sequential(composed_post, Constant(delta))) == normalize(composed_pre) where delta is the OperatorRealization cost for Int addition, and delta is asserted identical across rust_int_add / python_int_add / go_int_add (fail if cross-target drift). That ties the gate to the folded-away surface cost, not only identical Int type rows.

3. Cross-target overlay — unchanged second phase: normalized algebra bound at fold_pre is composed with each target’s Type(Int) realization row; all three normalized overlays must agree.

Fixture TestClaim.source comment + docs/r3-program-plan.md §1.8 row #51 text were updated so the ledger matches what the runner proves.

Please re-review when CI is green.

[briansrls]

Dashboard triage (item: briansrls comment @ 07:54Z — Codex resolution summary)

Verified against current session/smart-cat-457 / PR head: the three bullets in that comment still match the implementation (eval_r3_gate51_cross_target_constant_fold_binary_dimension_report_equals → analyze_symbolic_cost_dimension on separate fold_pre / fold_post roots; gate51_constant_fold_pre_post_reports_align with rust_int_add / python_int_add / go_int_add Operator rows for delta; second-phase Type(Int) overlay). No further commit is required for that thread — it is author documentation of an already-landed fix, not a new review finding.

Merge readiness (policy snapshot): fmt + ci jobs green; v3 check still pending on this workflow run. GitHub PR reviews list is empty (no api-review APPROVE / REQUEST_CHANGES visible here), so the bar you gave (≥2 distinct approving api-reviews, no REQUEST_CHANGES, CI fully green, mergeable clean) is not satisfied yet — no squash-merge from this agent until those land.

— sent from smart-cat-457

[briansrls]

Codex (8912) — ledger scope: fixed in 7fc3a4d36

Agreed: docs/r3-program-plan.md §1.8 row #51 must not read PASSING while docs/r3-structure.md still phrases gate #51 as certification-corpus-wide. The row is now CONSUMER_LANDED (partial) with an explicit deferral of corpus-wide PASSING until that evidence exists. Implementation + representative witness + integration ratchet on PR #2582 are unchanged.

Issue #2575 body was updated on GitHub (stale NYI line, acceptance note for partial vs corpus).

Merge readiness: still need ≥2 api-review APPROVE + no REQUEST_CHANGES; re-check CI on the new head after the push (prior v3 snapshot may be stale).

— sent from smart-cat-457

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 0c0ecc42 · Trigger: schedule
• Thinking: 325s wall

---

✅ Mixed code/docs/model fixture PR; no blocking concerns found.

[briansrls]

Triage: scheduled Codex api-review (08:09Z, commit 0c0ecc42)

• Code change: none — “no blocking concerns” stands; current HEAD (7fc3a4d36) only adds the §1.8 ledger honesty downgrade after the earlier Codex scope thread (corpus vs representative witness).
• CI on latest head: fmt / ci / v3 all SUCCESS on workflow run for 7fc3a4d36 (self_host_ratchet skipped as configured).

Merge readiness: GitHub still shows one reviews[] entry (this api-review), state: COMMENTED, reviewDecision empty. Policy here is ≥2 distinct approving api-review providers (e.g. codex + openai-pro) with no REQUEST_CHANGES — not met yet, so no squash-merge from this agent. Please land a second api-review on current HEAD (or dismiss/re-review if your process treats COMMENTED “no blocking concerns” as non-counting).

— sent from smart-cat-457

[briansrls]

Item: composer-2 tree visibility (8921)

The artifact describes a completed tree visibility check for docs/r3-program-plan.md against origin/main (blob e0a2a0a4…). That is an informational / hygiene signal (“check ran”), not a finding that the PR branch is wrong or that r3-program-plan.md on this PR fails the check. No code or doc fix commit is indicated by that line alone.

If a later artifact ever reports a diff vs expected tree for this PR’s docs/r3-program-plan.md edit, paste that delta and we can act on it.

Merge readiness (now): REST pulls/2582/reviews still shows one review (Codex api-review, COMMENTED). v3 is pending on the current workflow run. mergeStateStatus was BLOCKED when queried. Policy (≥2 distinct api-review approvals / no REQUEST_CHANGES / full green) is not satisfied — no squash-merge from here.

— sent from smart-cat-457

[briansrls]

Superseded by #2577 (merged for gate #51 r3-gate-51-d1628b69). Closing duplicate. — wise-bear-525 (R3 V Mgr)

[briansrls]

Verification (dashboard item: duplicate / supersession @ 2026-05-10T08:28:41Z)

Checked current GitHub + origin/main:

• R3 gate #51: make constant-fold cost witness executable #2577 is MERGED; main tip includes c03d0cd7c (“R3 gate Lane a p12 to p6 #51: make constant-fold cost witness executable (R3 gate #51: make constant-fold cost witness executable #2577)”).
• R3 gate #51: cross_target_optimization_constant_fold_consistent #2582 is CLOSED with the “Superseded by R3 gate #51: make constant-fold cost witness executable #2577 … duplicate” note — valid disposition: this PR was not the merge vehicle; gate Lane a p12 to p6 #51 execution landed via R3 gate #51: make constant-fold cost witness executable #2577 (session snappy-eagle-65 lineage in that PR’s commits), so there is no missing fix commit on R3 gate #51: cross_target_optimization_constant_fold_consistent #2582 for this thread.

Merge readiness (#2582): N/A — PR is closed; it was CONFLICTING at close, so the “squash-merge now if policy met” bar does not apply here. Work that was blocking downstream is unblocked by #2577 on main.

Note for session/smart-cat-457: this worktree’s branch head is not an ancestor of current main (local branch predates / diverges from the shipped merge); any follow-up should branch from origin/main @ #2577 rather than reviving #2582.

— sent from smart-cat-457