Lane A: R1 invert interpreter dispatch authority (roster becomes the authority)

[briansrls]

Summary

Lane A R1: invert interpreter primitive dispatch authority so the .dag roster is canonical and Rust is derived, not the other way around.

v1_interpreter_authored_roster_arms()  [.dag canonical]
  → v1_interpreter_dispatch_emit        [derives enum + lookup]
  → v1_interpreter_dispatch_generated.rs [drift-gated committed artifact]
  → v1_interpreter.rs                   [lookup_* then exhaustive handler match]

Head: a74dede10a on session/merry-bat-341-r1-dispatch-authority (merge main #7733; no content edits).

---

Main merge (wind-down, 2026-08-03)

Merged origin/main at 19cc776d4e (#7733 main repair — restores v1_generic_params_needing_clone_bound 9-arg call site lost to stale regen clobber). Merge commit a74dede10a; pushed; CI rerun triggered. No PR content changed — approvals preserved.

Generated-file oracle (git diff origin/main...HEAD -- <file>)

File	Three-dot diff	Verdict
src/v1/stage0/src/v1_compiler_emit_rust.rs	empty	OK — branch carries no delta vs main; #7733 already landed the 9-arg v1_generic_params_needing_clone_bound + emit_bare_type_params fixes; merge did not revert
src/v1/stage0/src/std_occurrence_binding.rs	empty	OK — impl<N: Clone> OccurrenceBindingFoldState<N> intact; no spurious removal
src/v1/stage0/src/v1_interpreter_dispatch_generated.rs	new file (+826)	OK — intentional branch addition only
src/v1/stage0/src/bootstrap_stage0_crate_layout_generated.rs	+1 line (v1_interpreter_dispatch_generated.rs in HAND_MAINTAINED_STAGE0_FILES)	OK — additive enrollment only
src/v1/stage0/src/gunbc_stage0_crate_layout_generated.rs	+v1_interpreter_dispatch_generated in pub_mod projections	OK — additive enrollment only

Local cargo check -p v1-compiler-tests --tests pass after merge.

---

Done (green by execution where verified)

• R1 inversion landed: v1_interpreter_authored_roster_arms() — 187 rows, enumeration: AuthoredInRoster; v1_interpreter_dispatch_emit generates lookup + enums; six handler-macro sites consult lookup_* before matching generated enum variants.
• Deleted on purpose: interpreter_dispatch_arm_rows builtin bridge, per-site *_roster macro expansions, stringly match dispatch at six sites.
• Drift gate enrolled: V1InterpreterDispatchGeneratedRsArtifact in generated_artifact.dag / generated_artifact_emit.dag / generated_artifact_drift_test.dag; .gitattributes merge driver row added (auto-heal ec50f5c4f8).
• Regen fixed-point: v1_interpreter_dispatch_generated enrolled in frontier.dag + stage0_crate_layout_generated projections; regen_stage0 --verify passed.
• Accidental deletion restored: type_ref_hit_ne_bind_measure_active removed alongside intentional roster macro deletion — restored in 04_method.dag / v1_compiler_infer_method.rs.
• Review fixes landed: stale authority prose + handler macro comments (review 47740); generated-artifact enrollment (review 47731); fail-closed unknown site keys (review 47777); invalid OccurrenceBindingFoldState impl emit repaired (f4e1b865c7, review 47813).
• CI heal fix: ci.yml heal staging + skew-guard exclude for v1_interpreter_dispatch_generated.rs (5418fa4a72).
• Local verification (this host): cargo build -p v1-compiler pass; cargo test -p v1-compiler --test interpreter_dispatch_authority — 2/2 pass.
• Reviews: 2 APPROVE, zero stale providers (per operator wind-down message).

---

Blocked / not done (operator still-bat-561 corrections — KEEP REWORKING)

PR is not draft but not merge-ready until operator merges: four substantive corrections still open post-R1:

1. Dead spelling copy — handler macros still carry { "spelling" } literals dispatch ignores; route on arm identity + body only.
2. Model dispatch site — add InterpreterDispatchSite coproduct per row; stop routing generation via roster_site_key prefix conventions (v4_bridge., native_intercept., etc.).
3. Collision refusal — pre-generation fold refusing duplicate arm_identity_to_rust_variant mappings and per-site spelling→identity collisions.
4. Vocabulary cleanup — rename v1_interpreter_derived_arms() (now returns authored rows); retire uninhabited DerivedFromDispatch once witness RED controls allow.

Also open: explicit bidirectional compile-failure construction witnesses per operator (roster-without-handler → non-exhaustive match; handler-without-roster → macro miss).

---

Findings (not visible in diff alone)

Finding	Window / denominator	Outcome
Enrolling V1InterpreterDispatchGeneratedRsArtifact without regenerating ci.yml	CI run 30781222013	generated_artifact_drift_gate_passes false; heal HealAuthorCommitRequired on ci.yml
cargo test -p v1-compiler interpreter_dispatch_authority name filter	1 invocation, exit 0	False green: 0 tests ran. Correct: --test interpreter_dispatch_authority → 2/2 pass
Bot action_required trap	PR runs before force-push	Auto-heal bot commits blocked pull_request CI
Roster macro deletion hazard	d0ddcef18b diff review	type_ref_hit_ne_bind_measure_active deleted accidentally; restored
Main E0061 clobber (6 vs 9 args)	main pre-#7733	Textually disjoint merge of regen outputs; repaired on main at 19cc776d4e; branch merge oracle clean
New committed artifact → ci.yml	Registry 15→16 static rows	ci.yml is AuthorCommitRequired — landed in human commit 5418fa4a72

---

Test plan

• regen_stage0 --verify — pass
• cargo build -p v1-compiler — pass
• cargo fmt --all --check — pass
• cargo test -p v1-compiler --test interpreter_dispatch_authority — 2/2 pass
• Merge main Repair main: restore #7708 emitter fixes lost to #7709 stale seed regen #7733 + generated-file oracle — pass (a74dede10a)
• CI full floor on a74dede10a — in progress (post-merge push)

---

Next (if continued)

1. Operator corrections 1–4 (above), in order.
2. Author explicit compile-failure RED fixtures for both roster↔handler directions.
3. Re-run full CI; confirm heal no longer exits 1 on author-commit drift.

Do not merge from this session — operator merges manually.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[gunbai-bot]

Addressed all three findings from review 47740 in 6b1e184365:

v1_interpreter_primitive_surface_authority_note — Replaced the pre-R1 "DERIVED, NOT TRANSCRIBED from macro token lists" paragraph with the inverted model: v1_interpreter_authored_roster_arms() is canonical, v1_interpreter_dispatch_emit generates drift-gated lookup/enums, and handler macros align at compile time via the generated enum match.

derived_versus_declared_note — Updated to describe the current ArmEnumeration population: almost all rows are AuthoredInRoster; the two DeclaredHere frontier rows remain; DerivedFromDispatch is documented as uninhabited legacy kept for witness RED controls.

v1_interpreter.rs handler macro comments — Removed stale "SINGLE AUTHORITY" / v1_builtin_roster / dual-expansion claims across all six handler macros (v1_builtin_arms, v1_algebra_method_arms, bridge/map/intercept/parse-table). Comments now state roster authority lives in .dag and generated lookup_* routes spellings before the exhaustive enum match.

Operator corrections 1–4 from still-bat-561 (dead spelling copy in macros, modeled dispatch site, collision refusal, vocabulary cleanup) remain open and are tracked separately from this doc-only pass.

[gunbai-bot]

Addressed review 47777 in 58000573b4.

Finding confirmed: expected_v1_interpreter_dispatch_generated_rs walked a hardcoded site_keys list while roster_site_key could yield bare "eval_call" for malformed rows — those rows were silently omitted from generated lookup with no refusal.

Fix:

• known_dispatch_emit_site_keys() is the single closed site list; generation uses it exclusively.
• roster_rows_with_unknown_emit_site() collects rows whose roster_site_key is outside that list.
• expected_v1_interpreter_dispatch_generated_rs() now returns V1InterpreterDispatchGenerated | V1InterpreterDispatchGenerationRefused and refuses when count(unknown) > 0.
• generated_artifact_emit propagates refusal into ArtifactGenerationRefused (drift gate goes red).
• Acceptance test adds: every_authored_roster_row_maps_to_known_emit_site, dispatch_emit_succeeds_on_live_roster, and RED control bare_eval_call_site_key_is_not_emittable ("eval_call" ∉ known sites; "eval_call_bridge" ∈).

Live roster (187 rows) still emits successfully; only the failure path is newly fail-closed.

[gunbai-bot]

Thanks for the approvals (reviews 47788, 47799, 47801).

Review 47801 — landed in 5418fa4a72:

• Tightened closing_contract_note / roadmap execution contract: compile-time exhaustiveness is now scoped honestly — single-family sites (eval_builtin_inner, eval_algebra_method_inner, map grounding, parse-table) vs multi-family bridge/native-intercept sites where the per-family wildcard means a missing handler body is unreachable!, not a compile error. Per-family enum split stays follow-up.
• generated_artifact_emit now forwards V1InterpreterDispatchGenerationRefused.reason instead of substituting a fixed string.
• Also committed the pending ci.yml heal staging for v1_interpreter_dispatch_generated.rs (HealAuthorCommitRequired class from the prior heal run).

Reviews 47788 / 47799: no action — nothing to block.

— sent from merry-bat-341

[gunbai-bot]

Review 47803: No action — the two honesty items (multi-family wildcard backstop, hand-maintained roster scaffold) are already named in the diff with dissolution triggers; agree with APPROVE.

Pushed regen fixed-point fix separately (emit_bare_type_params in seed) — CI was failing regen_verify_gate_passes because the post-#7709 emitter used bounded type params twice on enum accessor impls.

— sent from merry-bat-341

[gunbai-bot]

Review 47813 — fixed in f4e1b865c7 (HEAD).

Verified against current tree:

• std_occurrence_binding.rs:94 is back to valid Rust: impl<N: Clone> OccurrenceBindingFoldState<N> { (not <N: Clone> in the applied position). cargo check -p v1-compiler-tests --tests compiles; regen_stage0 --verify reports regen_divergence_count=0.

• The emit_bare_type_params change in emit_enum_shared_accessors is the correct direction, not the regression: the pre-Fix leaf-keyed qualified lookup collision in Rust emit #7709 seed duplicated type_params (<N: Clone>) for both the impl header and the type application, which is exactly how you get OccurrenceBindingFoldState<N: Clone>. The fix keeps bounded params on the impl (impl<N: Clone>) and bare names on the application (OccurrenceBindingFoldState<N> via emit_bare_type_params).

• What landed badly was f5c828d87d (WIP auto-heal/regen): it committed the broken OccurrenceBindingFoldState<N: Clone> bytes while the seed fix was still mid-flight. f4e1b865c7 restores the fixed-point bytes the emitter now produces; no further emitter edit needed for this blocker.

Please re-review on f4e1b865c7 — build was red on f5c828d only.

— sent from merry-bat-341

[gunbai-bot]

Review 47816: No action — agrees with the honest exhaustiveness scoping already in the closing contract (5418fa4a72 / f4e1b865c7). Shared-enum unreachable! backstop and per-family enum split as follow-up are documented; no new edits warranted by this review.

— sent from merry-bat-341

[gunbai-bot]

Review 47818: No action — APPROVE aligns with current tree: roster authority, drift-gated emit, deleted dual-authority bridge, and honestly documented residuals (shared-enum unreachable!, handler-body scaffold) are already in the closing contract and scaffold rows. Nothing new to land from this review.

— sent from merry-bat-341

[gunbai-bot]

Review 47831: No action — APPROVE matches current tree; nothing new to land from this review.

CI fix pushed (452a8154de): f1a810b broke build because module-path macro invocation does not resolve for non-macro_export rules, and $crate:: on macro_export rules trips macro_expanded_macro_exports_accessed_by_absolute_paths from nested handler macros. Fix: #[macro_use] on v1_interpreter_dispatch_generated (declared before v1_interpreter) + bare arm-macro names in handler expansions. Also aligned roadmap node v1-interpreter-primitive-dispatch-authority with site-scoped exhaustiveness (review 47830).

— sent from merry-bat-341

[gunbai-bot]

Review 47834: No action — APPROVE matches current tree; nothing new to land from this review.

CI fix pushed: regen was failing because lib.rs #[macro_use] did not survive regen_stage0. Fixed in emit (05_emit_rust.dag) + synced gunbc_stage0_crate_layout_generated.rs; regen_stage0 --verify passes locally.

— sent from merry-bat-341

[gunbai-bot]

Review 47840 — fixed: narrowed v1_interpreter_primitive_surface_authority_note to the site-scoped exhaustiveness bar (single-family sites → compile-time inner-match refusal; eval_call_bridge / eval_call_native_intercept → runtime unreachable! until per-family enum split). Matches roadmap_authority.dag and the acceptance contract.

— sent from merry-bat-341

[gunbai-bot]

Review 47843 — addressed in HEAD:

1. Variant collision (fixed): expected_v1_interpreter_dispatch_generated_rs now refuses when distinct_identities ≠ distinct_rust_variants within any emit site. Witness: live_roster_has_no_rust_variant_collisions + rust_variant_collision_synthetic_red_control (synthetic.one_two / synthetic.one.two).

2. Dispatch-site routing scaffold (fixed): dispatch_emit_site_routing_scaffold_note on v1_interpreter_dispatch_emit names the interim prefix-convention routing and a Dissolve-on: trigger for modeled dispatch_emit_site per row; enrolled in acceptance contract.

3. Closing-contract overclaim (fixed): closing_contract_note no longer says a discriminating RED lives in interpreter_dispatch_authority.rs — that file is lookup positive control only; REDs are bare_eval_call_site_key_is_not_emittable_red_control, rust_variant_collision_synthetic_red_control, and cargo build on single-family sites.

4. Bridge unreachable! (no code change): intentionally narrowed bar already in roadmap_authority.dag, acceptance contract, and handler macros — runtime backstop until per-family enum split, not claimed as compile-time refusal.

— sent from merry-bat-341

[gunbai-bot]

Review 47850: No action — APPROVE matches current tree; scaffolds and dissolution triggers already enrolled as described.

— sent from merry-bat-341