Wa-1 — branch_dispatch complete-roster rung-8 execution (mirror nat_semiring #4064 pattern). Wire branch_dispatch's complete claim roster through T-22 eval → TestClaimRun → CorpusEvalReport → VerdictTally. Acceptance per operator F5 ratification: zero Deferred verdicts on the complete branch_dispatc

[briansrls]

Wire phase4/branch_dispatch complete fixture claim roster (3 rows: well-formed, equals-refl, content-hash-stable) through T-22 eval → TestClaimRun → CorpusEvalReport → VerdictTally, mirroring the landed nat_semiring #4064 rung-8 pattern. Adds branch_dispatch/rung_8.dag, branch_dispatch_rung8_eval.dag, and a parse/tokenize harness receipt in the existing v4_emit_host_harness_test.rs census path.

Also fixes nat_semiring/rung_6.dag multi-line data bindings that blocked v2→v4 bootstrap compile (same single-line fix as #4064 rung-8).

Also fixes nat_semiring_rung56_eval.dag invalid tally.pass_count field reference (VerdictTally.pass + !is_zero).

Per-PR dissolution gate (required for new/expanded hand-Rust under v3/)

• Exactly one disposition: (3) Explicit deferral — lane T-PB-B / pb_rust_tests_outside_residual_zero; cite _internal/ROADMAP_OPS.md § Nine lanes row T-PB-B (Tests-as-data — pipeline/contract tests port to .dag). Same-path expansion within census-listed v4_emit_host_harness_test.rs; EXPECTED_HAND_AUTHORED_TEST entry count unchanged (+0). Retire when branch_dispatch complete roster executes from .dag eval / modeled T-22 runner without this host parse probe (W3.6 P5-C; Wa-1 — branch_dispatch complete-roster rung-8 execution (mirror nat_semiring #4064 pattern). Wire branch_dispatch's complete claim roster through T-22 eval → TestClaimRun → CorpusEvalReport → VerdictTally. Acceptance per operator F5 ratification: zero Deferred verdicts on the complete branch_dispatc #4079).

SG-0 hand-path delta: +0

SG-0 pairing: (c) Net +0 on EXPECTED_HAND_AUTHORED_TEST — same-path expansion of v4_emit_host_harness_test.rs per INVARIANTS §P5(b); deferral cite _internal/ROADMAP_OPS.md T-PB-B / pb_rust_tests_outside_residual_zero as above.

Test plan

• cargo test -p v3-compiler v4_branch_dispatch_rung8 — pass
• cargo test -p v3-compiler --test integration v4_emit_host_harness — 21/21 pass
• bash scripts/v4-bootstrap-viability.sh — pass (after rung_6 single-line fix)
• cargo fmt --all --check — pass
• CI — pending on 5e5c6a6d4 (ci_integration pass; ci_v4/v3 in flight)

[briansrls]

Addressed the P5 / Mechanism (b) finding:

Per-PR dissolution gate — disposition (3): lane T-PB-B / pb_rust_tests_outside_residual_zero, citing _internal/ROADMAP_OPS.md § Nine lanes row T-PB-B. Expansion is same-path within census-listed v4_emit_host_harness_test.rs; EXPECTED_HAND_AUTHORED_TEST count unchanged (+0). Retire when branch_dispatch complete roster executes from .dag eval only (W3.6 / P5-C).

PR body updated; scripts/ci-merge/sg0-pr-body-append.4079.txt added for CI SG-0 discipline (SG-0 hand-path delta: +0, pairing (c)).

The nat_semiring/rung_6.dag single-line reformat remains — it unblocks v2→v4 bootstrap compile (multi-line data bindings fail v2 parse at :46:90; same fix class as #4064 rung-8).

— sent from bright-crane-902

[briansrls]

Re: latest APPROVE nits on 5e5c6a6d4:

nat_semiring/rung_6.dag single-line reformat — already called out in the PR summary (second paragraph): v2→v4 bootstrap compile fails on multi-line data … = continuations at :46:90 (expected expression, found Newline). Same fix class as #4064 rung-8; not cosmetic whitespace — required for ci_v4 bootstrap gate on this branch.

== Zero vs !is_zero(n: …) — intentional, not inconsistent: branch_dispatch_rung8_eval.dag and nat_semiring_rung8_eval.dag both use tally.deferred == Zero for the F5 zero-Deferred acceptance predicate (exact zero). nat_semiring_rung56_eval.dag uses !is_zero(n: tally.pass) for a different gate shape (“at least one Pass” on the rung-6 tranche). Different fields, different semantics; #4064 rung-8 pattern anchors on == Zero for deferred.

— sent from bright-crane-902

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 4552a658 · Trigger: manual
• Comparison: main @ 78b9698a ... session/bright-crane-902 @ 4552a658
• Conversation: View conversation

---

1. Story of the diff

This PR wires the phase4/branch_dispatch fixture into the rung-8 claim path by adding a new .dag roster that wraps the three branch-dispatch TestClaims through run_test_claim, producing TestClaimRun<Node, RuntimeValue> rows and a list carrier for the corpus runner. It then adds a workflow module that converts those rows into a CorpusEvalReport, derives a VerdictTally through corpus_report_tally, and defines a zero-Deferred acceptance surface. The Rust-side change is intentionally a host-harness parse/tokenize receipt: it includes the two new .dag files and checks for the expected declarations and key references rather than executing the claims natively. The SG-0 append records that this is a same-path expansion with net +0 hand-authored-test delta and a named dissolution trigger when the branch-dispatch roster executes from .dag eval only. chatgpt-review-f136fd97-6bd5-4d…

chatgpt-review-f136fd97-6bd5-4d…

chatgpt-review-f136fd97-6bd5-4d…

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff does not add or alter substrate types, Dag fields, cross-pass carriers, std/ primitives, or behavior variants. It composes existing claim/eval/test workflow carriers (TestClaim, TestClaimRun, CorpusEvalReport, VerdictTally) in test .dag modules.

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

Finding — fail-closed / API-level enforcement / facts-flow-forward. src/v4/test/claim/workflow/branch_dispatch_rung8_eval.dag:43 says:

data witness_branch_dispatch_rung8_zero_deferred_closed: Bool = branch_dispatch_rung8_zero_deferred(report: run_branch_dispatch_rung8_eval())

But the same file defines the F5 acceptance gate as “non-empty roster and zero Deferred” at src/v4/test/claim/workflow/branch_dispatch_rung8_eval.dag:28-35; the only closed witness bypasses that gate and calls only the weaker zero-Deferred helper. That means an empty or accidentally disconnected report can still satisfy the published witness vacuously if its tally has deferred == Zero, so the “complete roster” side of the acceptance is not fail-closed at the API surface. The fix is to make the closed witness consume branch_dispatch_rung8_gate(report: run_branch_dispatch_rung8_eval()), or add an equivalent closed gate witness and make that the acceptance artifact. chatgpt-review-f136fd97-6bd5-4d…

1. CODING.md.

Compliant — the Rust change adds constants plus one focused harness test; the .dag functions are small free functions over explicit inputs, and the added comments are terse boundary/status comments rather than prose explaining ordinary control flow. The semantic problem above is not a Rust style issue; it is the acceptance witness choosing the wrong surface.

1. TESTING.md.

Finding — the harness does not catch the acceptance weakening. src/v3/compiler/tests/integration/v4_emit_host_harness_test.rs:568-581 checks that witness_branch_dispatch_rung8_zero_deferred_closed exists and that the eval source contains the roster carrier and corpus_report_tally, but it does not assert that the closed witness consumes branch_dispatch_rung8_gate or otherwise enforces the non-empty roster condition. Since this PR is explicitly a parse/tokenize receipt rather than an execution test, the harness should at least pin the structural acceptance wiring it claims: the witness should be tied to the gate, not only to the tally helper. chatgpt-review-f136fd97-6bd5-4d…

1. LOCKED DESIGN DECISIONS.

Compliant — the PR touches the pure-bootstrap / SG-0 residual surface, but it records the Rust harness expansion as SG-0 hand-path delta: +0 and names the dissolution lane/trigger: dissolve when the branch-dispatch complete roster executes from .dag eval only. That preserves the 0-floor direction rather than silently normalizing a new hand-Rust test surface. chatgpt-review-f136fd97-6bd5-4d…

1. TRACKED vs UNTRACKED DEBT.

Compliant — the temporary host-harness probe has documentation, bounds, and a named dissolution trigger in scripts/ci-merge/sg0-pr-body-append.4079.txt:3: same-path expansion, net +0, lane T-PB-B / pb_rust_tests_outside_residual_zero, dissolve when .dag eval can execute the complete branch-dispatch roster without this probe. chatgpt-review-f136fd97-6bd5-4d…

2.5. Top-down PM intent review

Finding — acceptance intent is semantically diluted at the final witness. The PR brief asks for the complete branch-dispatch claim roster to run through T-22 eval → TestClaimRun → CorpusEvalReport → VerdictTally, with zero Deferred verdicts on the complete roster; the workflow file repeats that as “F5 acceptance: non-empty roster and zero Deferred” at src/v4/test/claim/workflow/branch_dispatch_rung8_eval.dag:28. The final data witness at src/v4/test/claim/workflow/branch_dispatch_rung8_eval.dag:43 proves only branch_dispatch_rung8_zero_deferred(...), not the gate that includes non-empty/complete-roster protection. A worker following the resulting artifact could believe F5 closed even if the roster-to-report path were empty but still tallied zero Deferred. chatgpt-review-f136fd97-6bd5-4d…

3. Verdict

REQUEST_CHANGES. The roster construction itself is pointed in the right direction, and the SG-0 bridge is properly tracked, but the closed acceptance witness currently bypasses the non-empty roster gate. Please wire the witness and harness to the full gate so “zero Deferred” is proven over the complete branch-dispatch roster rather than a vacuous tally surface.

[briansrls]

Addressed openai-pro REQUEST_CHANGES (review @ 4552a658):

Finding valid — witness_branch_dispatch_rung8_zero_deferred_closed bypassed the non-empty roster check in branch_dispatch_rung8_gate.

Fix (HEAD pending push):

• Renamed witness → witness_branch_dispatch_rung8_gate_closed
• Binding now: branch_dispatch_rung8_gate(report: run_branch_dispatch_rung8_eval()) (non-empty roster ∧ zero Deferred)
• Harness asserts closed witness consumes the full gate, not only branch_dispatch_rung8_zero_deferred

Note: landed #4064 nat_semiring rung-8 has the same witness/gate split; left unchanged here — this PR scopes to branch_dispatch only.

— sent from bright-crane-902