docs(planning): rung-3 acceptance spec — phase4/branch_dispatch

[briansrls]

Summary

Additive rung-3 acceptance predicate spec for phase4/branch_dispatch (Phase 4 widening, #4018). Cites gate vocabulary from v4-ladder-rung-specs-2026-05-30.md §2 and v4-ladder-rung-spec-rung3-2026-05-30.md (#4003) — no §2 duplication. Records fixture-particular subject (Branch + two Transform arms), claims, baseline wedge (no RoundTripClaim row yet), and matrix render.

Test plan

• Docs-only; no substrate change.
• CI discipline/fmt (expected pass).

Companion: loop_linear_bound rung-3 spec lands in a separate PR per wise-seal-69 sequencing.

Made with Cursor

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 345aee2b · Trigger: manual
• Comparison: main @ f7f629a3 ... session/royal-wolf-898-rung3-branch-dispatch @ 345aee2b
• Conversation: View conversation

---

1. Story of the diff

This PR adds a new planning spec for rung 3 of the v4 correctness ladder, scoped specifically to phase4/branch_dispatch. The document binds that fixture to the existing rung-3 companion vocabulary instead of redefining predicates, keeps W1 as .dag-only round-trip readiness, and explicitly preserves the current FAIL/SKIP baseline because the fixture has no authored RoundTripClaim yet and W1b fidelity remains unlanded under T-36. It also separates the real ladder row from the phase1-aux/dag_round_trip_mvp1 diagnostic row, so workers should not mistake the aux PASS for closing phase4/branch_dispatch rung 3. chatgpt-review-1c9fdd72-59b0-49…

chatgpt-review-1c9fdd72-59b0-49…

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff is a Markdown planning artifact only; it references existing substrate concepts such as ComputationNode { behavior: Branch } but does not add substrate types, Dag fields, walkers, predicates, or new variants.

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

Finding — docs/planning/v4-ladder-rung-spec-rung3-phase4-branch-dispatch-2026-05-30.md:7, Boundary Discipline / Documentation Describes Live State. The scope header links the rung 0–2 roster as [src/v4/test/claim/branch_dispatch/rung_0_to_2_three_targets.dag](../src/v4/test/claim/branch_dispatch/rung_0_to_2_three_targets.dag). Because this new file lives under docs/planning/, ../src/... resolves to docs/src/..., not repo-root src/...; the correct relative link should be ../../src/v4/test/claim/branch_dispatch/rung_0_to_2_three_targets.dag. This is a load-bearing acceptance-spec pointer, so the boundary does not actually carry a usable reference for the worker following the document. chatgpt-review-1c9fdd72-59b0-49…

1. CODING.md.

Compliant — no Rust implementation is added; the Markdown uses bounded tables and brief rationale, and the doc explicitly avoids duplicating companion predicate definitions.

1. TESTING.md.

N/A — docs-only acceptance-spec PR; it does not change executable behavior. The document’s purpose is to define the future test/claim surface, and it honestly keeps the current rung 3 result at FAIL (dag=SKIP) until a RoundTripClaim is authored. chatgpt-review-1c9fdd72-59b0-49…

1. LOCKED DESIGN DECISIONS.

Compliant — the doc does not dilute the rung split or overclaim target-language round-trip support: W1b remains explicitly unlanded, row aggregate remains FAIL, and rust/python/go Branch round-trips are kept out of scope until rung 5. chatgpt-review-1c9fdd72-59b0-49…

1. TRACKED vs UNTRACKED DEBT.

Compliant — the TODO/follow-up surfaces are bounded to this fixture/rung and have named closure conditions: the RoundTripClaim row waits on module-loader/claim authoring, the host-gate extension is called out as a Phase 4 follow-up, and the baseline is superseded by the first PR that binds the RoundTripClaim and extends a phase4-aware host gate script. chatgpt-review-1c9fdd72-59b0-49…

2.5. Top-down PM intent review

Compliant — the PR preserves the high-level intent rather than watering it down. It keeps tests as .dag claim data, does not add hand-written implementation, distinguishes aux diagnostics from ladder closure, and does not convert W1b or target-language round-trip fidelity into optional/implicit success. The only issue I found is a broken relative link in the scope header, not a semantic change to the plan.

3. Verdict

REQUEST_CHANGES

The planning shape is otherwise sound and appropriately fail-closed, but the new authoritative spec contains a broken load-bearing link to the rung 0–2 roster. Fixing ../src/... to ../../src/... should be enough for me to approve.

[briansrls]

Proactive fix in 6f5316929: same roster link correction as #4035 (../../src/v4/... from docs/planning/).

— sent from royal-wolf-898

[briansrls]

openai-pro RC on 345aee2b (broken ../src/v4/… scope link) is already addressed in 6f5316929 — link is now ../../src/v4/test/claim/branch_dispatch/rung_0_to_2_three_targets.dag, which resolves correctly from docs/planning/ to repo-root src/v4/. No further doc change needed on this PR.

— sent from royal-wolf-898