R3 gate #1: tier3 termination mirror dissolved

[briansrls]

Auto-opened by session-dashboard for session royal-tern-883.
Pushing to session/royal-tern-883 advances this PR.

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]

Verified the dashboard-only review against current HEAD c95b54e. The six termination lattice helper bodies are absent from src/v3/compiler/src/dag.rs, and the remaining declarations resolve through src/v3/std/termination.dag. The noted include_str!/pub fn tripwire limitation was explicitly non-blocking; I also checked for pub(...) fn reintroductions locally and found none, so there is no code change to make for this review. — sent from royal-tern-883

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 101a1d02 · Trigger: schedule
• Thinking: 210s wall

---

BLOCKING (1)

Root Cause

• docs/briefs/r3-pb-tier3-perf-budget-worker.md Phase-1 baseline capture is specified as a prerequisite but no baseline exists in the base tree or PR diff → land the frozen termination baseline first or keep the termination bench until capture completes.

Non-blocking — Strengths

• src/v3/compiler/src/dag.rs The lattice operation mirror deletion correctly moves operation authority back to std.termination and reduces duplicate hand-Rust state under P5.

ROADMAP — Incomplete

• Tier-3 perf-budget Phase-1 hand-Rust mirror benches: The termination bench retirement is in the right dissolution lane, but the required frozen baseline data is not present before the bench disappears.

⚠️ One ordering issue blocks this: capture or commit the termination perf baseline before deleting the only hand-Rust bench that can produce it.

[briansrls]

Violations (could not place on specific lines):

• src/v3/compiler/benches/tier3_fixtures/README.md:5 BLOCKING: Retiring the termination Phase 1 bench before a committed tier3_baseline.json row violates the R3 perf-budget ordering in docs/briefs/r3-pb-tier3-perf-budget-worker.md and makes the hand-Rust termination baseline unrecoverable after the mirror is deleted.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 0406d174 · Trigger: manual
• Comparison: main @ 5065653a ... session/royal-tern-883 @ 0406d174
• Conversation: View conversation

---

1. Story of the diff

This PR removes the remaining hand-Rust executable mirror for the termination lattice from dag.rs and narrows dag.rs back to carrier-only responsibility. The load-bearing change is the comment shift at src/v3/compiler/src/dag.rs:890-893: .dag remains the carrier and lattice-operation authority, while Rust keeps only carrier shapes needed by current consumers, and executable operations like merge_evidence / join_evidence must stay in std.termination. The supporting changes retire the corresponding Criterion benchmark from tier3_mirror_perf.rs and update the tier-3 fixture README to say the termination bench retired with the mirror (src/v3/compiler/benches/tier3_fixtures/README.md:5-6). The test previously asserting Rust mirror behavior is converted into a dissolution guard: it checks the Rust functions do not reappear in dag.rs and that the named declarations still resolve to src/v3/std/termination.dag (src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:495-527).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — this touches substrate-adjacent Rust carriers, but the decision is to reduce implementation authority: src/v3/compiler/src/dag.rs:890-893 explicitly assigns carrier/lattice authority to .dag and keeps Rust to carrier shapes only.

1. INVARIANTS.md + modeling-discipline.md.

Compliant — single authority / progress is dissolution / coprod mirror dissolution are honored by deleting the executable Rust mirror and adding a reintroduction guard. The test forbids the Rust mirror functions from reappearing at src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:498-508, then checks those operation names resolve to std.termination at src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:512-526.

1. CODING.md.

Compliant — the diff removes free-function implementation code from the broad dag.rs substrate module instead of adding methods, hidden state, or a new helper object. The remaining module comment at src/v3/compiler/src/dag.rs:891-893 defines a narrow interface boundary: Rust carrier shapes only; executable lattice operations stay in the .dag authority.

1. TESTING.md.

Compliant — the test is behaviorally aligned with the change’s purpose: the contract is no longer “Rust mirror returns the same lattice values,” but “Rust mirror is dissolved and std.termination remains the source.” The first half asserts non-reintroduction via include_str! at src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:496-508; the second half asserts source authority via Dag::new() and declaration spans at src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:512-526.

1. LOCKED DESIGN DECISIONS.

Compliant — this does not alter a locked substrate shape or introduce a new target, connective, behavior, or bootstrap exception. It moves toward the locked zero-floor / generated-authority direction by retiring a hand-Rust mirror rather than adding one, as reflected by src/v3/compiler/src/dag.rs:890-893.

1. TRACKED vs UNTRACKED DEBT.

Compliant — no new scaffold, TODO, compatibility bridge, or temporary representation is introduced. The prior scaffold language is replaced with a steady-state boundary statement at src/v3/compiler/src/dag.rs:888-893, and the test names the dissolution guard directly at src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:495-508.

2.5. Top-down PM intent review

Compliant — the highest-level intent here is dissolution of a hand-Rust termination mirror, not preservation of its benchmark or duplicate executable behavior. The diff preserves that intent directly: dag.rs says executable lattice operations must stay in std.termination (src/v3/compiler/src/dag.rs:890-893), the bench no longer imports or runs merge_evidence (src/v3/compiler/benches/tier3_mirror_perf.rs:15-16), and the test asserts the removed functions must not reappear in dag.rs (src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:498-508). I do not see semantic dilution from “dissolve the mirror” into permanent scaffolding.

3. Verdict

APPROVE — the PR removes duplicate Rust execution authority for termination lattice operations, updates the perf fixture story accordingly, and adds a focused regression guard against reintroducing the mirror. I found no diff-cited invariant violation.

[briansrls]

Verified against current head 8aad1c7: this ordering finding is valid. There is no committed src/v3/compiler/benches/tier3_baseline.json, and this PR removes the only live termination Phase-1 bench row. I cannot safely fix this inside the dissolution PR by committing a placeholder/local baseline because docs/audit/c1-tier3-baseline-capture-procedure.md §6 explicitly forbids non-canonical or placeholder tier3_baseline.json data. I also cannot keep the termination bench without reintroducing the hand-Rust termination lattice mirror that this gate is assigned to dissolve. This is the documented C1/R3 sequencing STOP condition, so I escalated to the PB manager for a sequencing/waiver decision rather than papering over it in code. — sent from royal-tern-883

[briansrls]

Closing this PR as blocked by the C1 Phase-1 baseline prerequisite. PB Manager confirmed R-7 should be a hard prerequisite: src/v3/compiler/benches/tier3_baseline.json must land on main before any T-Tier3 mirror dissolution PR deletes the corresponding Phase-1 hand-Rust bench. Keeping this PR open as mergeable would permanently lose the termination baseline capture path, while committing placeholder/local baseline data is forbidden by docs/audit/c1-tier3-baseline-capture-procedure.md §6. This work should be respawned after C1 Phase-1 baseline capture lands. — sent from royal-tern-883

[briansrls]

Closed as blocked on C1 Phase-1 baseline capture prerequisite. See prior comment for sequencing details.