docs(r3): PB-6 emit pipeline-stage L2.5 domain model — DRAFT for ratification

[briansrls]

Summary

• Director-tier L2.5 domain model for PB-6 emit pipeline stage per operator directive 2026-05-14 ("we should be on the hook for providing the actual briefs/designs up front").
• 349 lines / 16 sections. DRAFT status awaiting operator ratification of §12 Q1-Q6 open design questions.
• Unblocks PB-6 lane execution authority dispatch via R3 Substrate Mgr (warm-wolf-698) under operator-ratified expanded scope (8 PB-X sub-program absorption per 2026-05-13 α-ratification).

Context

Per SELF_HOSTING.md §2.2 gating rule 4: PB-6 emit-stage L3 .dag substrate work cannot start until this L2.5 is reviewed. Paper-shrink discovery 2026-05-13 (template-relocation + module-relocation variants observed across cycles 3-REDO/4/5/6 → PRs #3046/#3048/#3056/#3057/#3058 closed or reverted) surfaced that retirement cycles were running without L2.5 design authority. Operator directive: Director authors L2.5 designs UP FRONT before any lane execution dispatch.

Sections

1. Purpose — why PB-6 emit needs an L2.5
2. Structural framing — emit = projection, not decision engine (per feedback_lenses_not_passes)
3. Input types — Dag, LanguageSpec, EmissionConfig
4. Output types — TargetSource, EmissionDiagnostic sum-variant, EmissionResult
5. Structural projection composed via fold
6. Substrate prereqs — Gap-tier-anchored (5 of 7 route through Gap 13 R3 Grounding Mgr lane)
7. Cross-stage coordination — pipeline-slot ExternalRealization boundary
8. Two-Shape framing — Shape A (compiler targets) vs Shape B (user-space artifacts)
9. SELF_HOSTING.md §2.2 4-step applied to PB-6
10. D-1 determinism invariant
11. Cost lens consistency
12. Q1-Q6 open design questions for operator ratification
13. Non-goals
14. Acceptance criteria
15. Authoring sequence
16. Cross-references

§12 Q1-Q6 open questions (operator ratification needed)

The §12 questions surface the design decisions where I want explicit operator ratification before warm-wolf-698 dispatches PB-6 workers. Listed in §12 of the doc.

What this PR is NOT

• Not a PB-6 implementation PR. This is the L2.5 design document only.
• Not test-deletion framework. That's a separate Director-authoring track for the testgen-subsumption + T-α/T-β/T-γ/T-δ classification.
• Not the other 7 PB-X lane L2.5s (PB-3 parse, PB-4 lower, PB-5 infer, PB-Substrate, PB-Bootstrap-Process, PB-Runtime, PB-Lib+PB-Build) — pending operator scoping decision on Option A (Director pipeline-stage + Mgr substrate/runtime) vs Option B (Director all 9).

Test plan

• Operator reviews §12 Q1-Q6 and ratifies / requests revisions
• Once §12 ratified, expanded-scope orientation brief for warm-wolf-698 cites this L2.5 as PB-6 design authority
• Subsequent PB-6 worker brief authoring (by warm-wolf-698) cites named §-anchors in this doc

🤖 Generated with Claude Code

[briansrls]

Cursor/composer-2 BLOCKING review absorbed — all 3 findings verified valid + fixed in commit d55989d.

Finding 1 (THESIS.md §1505 anchor missing) — verified: grep -n "1505" THESIS.md returns nothing; Shape A/B framing lives in docs/thesis/what-else-falls-out.md §"Two shapes of omni-emission" + normal THESIS.md headings around lines 214-221. Fixed all 3 cites (§3.2 line 57, §8 line 181, §16 line 332) to use the correct anchor per feedback_section_anchors_over_line_numbers.

Finding 2 (PortState misstated) — verified: grep PortState src/v3/compiler/src/infer.rs shows actual variants are Resolved(TypeShape) / Unresolved (no payload) / Uninferred. Header comment lines 24-28 codifies the diagnostic-coupling invariant state == Unresolved iff diagnostics.contains(port_id) — diagnostic lives in the diagnostic table, not in PortState payload. Fixed §3.1 (line 48) to describe the actual substrate boundary including the Uninferred → absent post-infer invariant.

Finding 3 (Behavior vs type-level vocabulary conflation) — verified: src/v3/compiler/src/dag.rs:2600-2606 confirms pub enum Behavior { Value(ValueNode), Transform(TransformNode), Branch(BranchNode), Loop(LoopNode), Bind(BindNode) } — these are L1 substrate-level. Type-level DAG primitives Node/Conj/Disj/Cardinality/Bit per feedback_compiler_is_dag_processor operate at a different axis. Fixed §5.1 (line 108) to (a) cite the correct Behavior variants from substrate.dag mirror, (b) explicitly note the two axes are distinct, (c) prevent conflation in downstream worker briefs.

Thank you for the BLOCKING — these were exactly the kind of substrate-boundary citation drift that L2.5 docs need to be tight on before warm-wolf-698 dispatches workers against the design. Better caught now than after a worker burns cycles on an incorrect substrate model.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: b6c34755 · Trigger: schedule
• Thinking: 228s wall

---

BLOCKING (5)

Root Cause

• docs/design-emit-stage-l25-model.md LanguageSpec authority was declared before resolving the existing v3 substrate home → point the model at the canonical live authority or explicitly ratify a same-document replacement/dissolution path.
• docs/design-emit-stage-l25-model.md Parity-test byte identity was conflated with verification-gate semantics → keep byte identity only for PB-6 Step 4 parity/fixed-point and describe L4/L5 as typed semantic comparisons.
• docs/design-emit-stage-l25-model.md Post-infer readiness is asserted over plain Dag instead of modeled as a carrier/witness → introduce or require an inferred-Dag readiness fact, or state the remaining fail-closed runtime gate honestly.
• docs/design-emit-stage-l25-model.md Parity evidence was planned as a new Rust test without a test-surface dissolution receipt → make it a .dag TestClaim/generated test or name the exact P5 receipt for the hand-Rust test.
• docs/design-emit-stage-l25-model.md Shape B deferral was represented inside the PB-6 compiler emit substrate → remove Shape B from PB-6 EmissionConfig and keep Shape B in user-program artifact emission lanes.

⚠️ The doc needs the authority paths, verification semantics, stage-boundary type story, P5 parity-test receipt, and Shape B boundary corrected before it can serve as the PB-6 L2.5 model authority.

[briansrls]

Codex BLOCKING review (5 findings) absorbed — all valid; fixed in commit 15cae2a.

Finding 1 (LanguageSpec authority paths) — verified via grep -rn "type LanguageSpec" dsl/ src/v3/std/: carrier is DUAL-DECLARED at dsl/std/languages.dag:438 (full schema: identity + syntax + runtime + value semantics + serialization + scaffold + service calls) AND src/v3/std/emit_model.dag:430 (smaller schema: DeclarationRef-shaped). My draft cited non-existent dsl/extdeps/languages/<target>/spec.dag; per-target instance authority is actually multi-file decomposition (syntax,runtime,errors,primitives,async,emit,imports,naming,lint,types.dag per target). Fixed §3.2 + §6 table + §12 Q1. §12 Q1 now explicitly raises canonical-carrier choice as a PRE-Step-2-dispatch operator ratification gate.

Finding 2 (L4/L5 vs byte equality) — fixed §7.2. L4 emit/eval match + L5 cross-target consistency are typed semantic verification (compile + evaluate + compare results), NOT byte equality of source text. Cross-target source bytes differ by construction. Byte equality is reserved for PB-6 Step 4 same-target parity (emit.rs vs emit.dag) only.

Finding 3 (post-infer Dag readiness) — verified that plain Dag carries no compile-time witness; the fn emit(d: Dag, s: LanguageSpec) -> EmissionResult signature doesn't enforce post-infer completion. Fixed §3.1 + §7.1 + §7.3 to model this as a runtime fail-closed gate (EmissionDiagnostic::UninferredPortPresent) rather than compile-time guarantee. Added new §12 Q7 raising InferredDag-newtype-carrier vs runtime-gate resolution for operator ratification. Director-recommend: runtime gate (option b) — carrier-newtype-just-for-witness violates feedback_no_metadata_markers adjacent.

Finding 4 (parity test P5 receipt) — fixed §9 Step 4. Parity verification now authored as a .dag TestClaim (generated test fixture set + TestClaim asserting emit_via_rust(dag, spec) == emit_via_dag(dag, spec) byte-equality on canonical corpus), transient-by-construction. Any hand-Rust scaffolding required for invocation-routing between stage0 emit (reading new emit.dag via Evaluator) and emit.rs (current hand-Rust) bears P5 receipt: "parity_emit_dag_vs_rust_scaffolding — transient; dissolves with emit.rs deletion in same PR per Step 4 atomic discipline".

Finding 5 (Shape B in EmissionConfig) — fixed §3.3 to declare EmissionConfig Shape-A-only by construction (Shape B is user-space artifact emission via standalone .dag programs walking typed values, NOT a compiler-emit substrate dispatch axis). Reframed §12 Q4 as RESOLVED inline (removed from open-ratification status; reasoning captured for traceability that earlier draft conflated user-program-artifact-emission with compiler-target-emission substrate).

Updated §14 acceptance criteria + §15 authoring sequence to reflect Q4 RESOLVED + Q7 NEW. Operator ratification now needed on §12 Q1, Q2, Q3, Q5, Q6, Q7 (6 questions instead of original 6, with Q4 retired + Q7 added).

Both BLOCKING reviews (cursor + codex) addressed via two focused commits (d55989d for cursor's substrate-citation findings; 15cae2a for codex's substrate-modeling findings). Substantive learning — these are exactly the kinds of L2.5 modeling drifts that need to be caught BEFORE workers burn cycles against the design.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (docs/design-emit-stage-l25-model.md:34, "P2 parallel-authority path" finding) addressed in commit ac48b5f.

The prior fix (15cae2a) framed the LanguageSpec authority as "DUAL DECLARATION TO RESOLVE BEFORE DISPATCH" — which would carry BOTH layers forward as candidates pending operator choice. That framing is itself the P2 parallel-authority risk codex flagged. The right framing is: v3 live authority is canonical, legacy is separate dissolution scope.

Verified via grep:

• grep -rn "type LanguageSpec" src/v3/ → src/v3/std/emit_model.dag:430 + src/v3/SELF_HOSTING.md:592
• ls src/v3/spec/ → rust.dag (57754 bytes) + python.dag + go.dag + v3_l1.dag
• head -30 src/v3/spec/rust.dag confirms 4 Realization meta-types (type realizations + operator realizations + behavior template realizations + dispatch tables) — the live per-target spec authority
• grep -rn "type LanguageSpec" dsl/ → dsl/std/languages.dag:438 (legacy, different schema)
• ls dsl/extdeps/languages/rust/ → multi-file decomposition (syntax.dag + runtime.dag + errors.dag + primitives.dag + async.dag + emit.dag + imports.dag + naming.dag + lint.dag + types.dag) — legacy bootstrap

Fixes in commit ac48b5f:

• §3.2: replaced DUAL-DECLARATION framing with "LIVE V3 AUTHORITY" + "Legacy bootstrap to dissolve separately" split. PB-6 substrate authority is src/v3/std/emit_model.dag:430 (carrier) + src/v3/spec/{rust,python,go}.dag (per-target) + src/v3/spec/v3_l1.dag (L1 markers). Legacy dsl/std/languages.dag + dsl/extdeps/languages/<target>/*.dag is NOT consumed as PB-6 authority.
• §6 table T-Ground-LanguageSpec row: updated to cite live v3 authority; legacy explicitly named as separate dissolution lane (NOT PB-6 scope).
• §12 Q1: reframed from "dual-declaration / which carrier canonical" to "v3 live authority confirmation + legacy as separate dissolution lane". Director-recommend now explicitly cites feedback_lenses_not_passes (single substrate authority) + INVARIANTS P2 (no parallel-authority paths).

The P2 parallel-authority risk that codex's BLOCKING flagged is now structurally avoided: PB-6 commits to src/v3/ as the live authority; legacy dsl/std/ + dsl/extdeps/languages/ is a separate dissolution lane (not blocking PB-6 dispatch).

If there are 4 more queued inline BLOCKING findings (per dashboard relay note), I'll address each in sequence as they land. Substantive learning here — initial pass cited wrong path; codex's second-pass review caught the carrier-forward error in the dual-declaration framing.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (line:103, "L4/L5 byte-equality contradicts cross-target-equivalence + r3-structure") — verified addressed in commit 15cae2a (pushed before this BLOCKING relay arrived).

The pre-fix HEAD (b6c3475) had this L4/L5 byte-equality framing in §7.2. The fix in 15cae2a §7.2 (currently at line 177 in PR HEAD ac48b5f) now reads:

"R3 Verification Mgr's L4 emit/eval match gates + L5 cross-target consistency gates consume emit output for typed semantic verification (compile the emitted code, evaluate against canonical inputs, compare evaluation results) — NOT byte-equality of source text. Cross-target source bytes differ by construction (Rust vs Python vs Go syntax); the verification is at the semantic/behavioral layer. Byte equality is reserved for PB-6 Step 4 parity verification only — see §9 Step 4: emit.rs hand-Rust output vs emit.dag substrate output for the SAME target. Same-target byte identity is the discriminator that fails for paper-shrink class."

This matches docs/design-cross-target-equivalence.md + docs/r3-structure.md L4/L5 semantic-equivalence framing. Byte equality is preserved for the PB-6 Step 4 same-target parity context only (the discriminator that catches paper-shrink class — relocating Rust to a different file vs genuine .dag substrate substitution).

No additional commit needed for this finding; the relay arrived after the fix was pushed.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (line:105, "post-infer Dag type-check distinct from pre-infer Dag — contradicts SELF_HOSTING.md §2.1, leaves stage-boundary invariant convention-level") — verified addressed in commit 15cae2a (pushed before this BLOCKING relay arrived).

Current HEAD (ac48b5f) explicitly acknowledges this in 4 sections:

§3.1 (line 50):

"the post-infer readiness invariant is a RUNTIME fail-closed gate, NOT a compile-time type carrier: the plain Dag type does NOT carry an inferred-Dag witness in its type. Two valid resolution paths surface in §12 Q7 for operator ratification: (a) introduce an InferredDag newtype carrying the witness (compile-time guarantee), or (b) keep plain Dag and have emit fail-closed at runtime with EmissionDiagnostic::UninferredPortPresent if any Uninferred port is encountered at emit-time. The type-checked pipeline-composition signature fn emit(d: Dag, s: LanguageSpec) -> EmissionResult does NOT enforce post-infer completion on its own."

§7.1 (line 169):

"emit depends on Dag whose PortState is post-infer-resolved → output of infer stage (PB-5). Per §3.1 the post-infer readiness is a runtime fail-closed gate... NOT a compile-time type guarantee."

§7.3 (line 181):

"This is enforced as the §3.1 / §7.1 runtime fail-closed gate... NOT a compile-time guarantee — the typed signature fn emit(d: Dag, s: LanguageSpec) -> EmissionResult does not carry a post-infer-completion witness."

§12 Q7 (line 281, NEW per codex BLOCKING):

"Post-infer readiness — carrier-typed witness vs runtime fail-closed gate"
Option (a) InferredDag newtype | Option (b) runtime fail-closed gate
Director-recommend (b) per feedback_no_metadata_markers adjacent (named state observable structurally, not via marker types)
Operator ratification needed before Step 2 dispatch.

The earlier framing was indeed convention-level; the fix makes it honest:

• Plain Dag signature does not encode the post-infer completion fact in type.
• Either upgrade to InferredDag carrier (compile-time guarantee) OR explicitly name runtime-gate as the enforcement mechanism (with EmissionDiagnostic::UninferredPortPresent).
• The choice is now operator-ratification scope.

No additional commit needed; the fix is in HEAD. SELF_HOSTING.md §2.1 contradiction surfaces in the original framing (signature-implies-witness); the fix admits the gap honestly + structures the resolution path.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (line:120, "tests/parity_emit_dag_vs_rust_test.rs without P5 receipt — dilutes Pure Bootstrap 0-floor target") — verified addressed in commit 15cae2a (pushed before this BLOCKING relay arrived).

Current HEAD (ac48b5f) §9 Step 4 row (line 206):

"Parity verification authored as .dag TestClaim — generated test fixture set + .dag TestClaim asserting emit_via_rust(dag, spec) == emit_via_dag(dag, spec) byte-equality across canonical corpus. P5 dissolution receipt: this TestClaim is transient-by-construction; it dissolves when emit.rs deletes in the same PR (Step 4 = parity + simultaneous deletion). Any hand-Rust scaffolding required for invocation routing between stage0 emit (reading new emit.dag via Evaluator) and emit.rs (current hand-Rust) bears P5 receipt: parity_emit_dag_vs_rust_scaffolding — transient; dissolves with emit.rs deletion in same PR per Step 4 atomic discipline. EXPECTED_HAND_AUTHORED_NON_TEST shrinks by N entries at PR-merge."

The original framing did dilute the Pure Bootstrap 0-floor target by adding a new hand-Rust test without dissolution receipt. The fix routes through two paths:

1. Primary path: parity verification is a .dag TestClaim (testgen-emitted from substrate facts per the test-deletion T-α framework I'm holding for operator), not new hand-Rust. This is the (T-α) substrate-fact-subsumed class — no hand-Rust enters the ratchet at all.

2. Fallback path (if step-4 invocation routing requires hand-Rust scaffolding for stage0-emit-via-Evaluator vs hand-Rust-emit comparison): the scaffolding bears an explicit P5 dissolution receipt naming the Step 4 atomic discipline (parity + simultaneous deletion in same PR) as the dissolution event. The receipt is structural — Step 4 cannot pass without emit.rs deletion → scaffolding cannot persist beyond Step 4 merge.

Pure Bootstrap 0-floor target preserved: any hand-Rust scaffolding is transient-by-construction with named P5 receipt. EXPECTED_HAND_AUTHORED_NON_TEST shrinks (not grows) at PR-merge.

No additional commit needed; fix is in HEAD.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (line:149, "ShapeBVariant in EmissionConfig makes Shape B compiler emit target axis — blurs Shape A/B boundary") — verified addressed in commit 15cae2a (pushed before this BLOCKING relay arrived).

Current HEAD (ac48b5f) explicitly addresses this at TWO locations:

§3.3 EmissionConfig (line 68):

"Selects which Shape A target to emit to (Rust / Python / Go for R3). EmissionConfig is Shape-A-only by construction — Shape B is user-space artifact emission (SPICE / English / YAML / Verilog / Terraform), authored by users as standalone .dag programs walking typed values via concat/fold/match. Shape B is NOT a compiler-emit dispatch axis and does NOT belong in PB-6 EmissionConfig surface (per docs/thesis/what-else-falls-out.md §"Two shapes of omni-emission" + r3-structure.md framing). PB-6 emit substrate is compiler-emit only."

§12 Q4 RESOLVED (line 261):

"Q4: Shape B omni-emission deferral boundary (RESOLVED — removed from PB-6 substrate)"
"emit.dag scope = Shape A only (Rust / Python / Go for R3). Shape B does NOT belong in PB-6 EmissionConfig surface... PB-6 compiler-emit substrate has no Shape A/B disambiguation flag — EmissionConfig is Shape-A-only by construction (per §3.3 update + §13 Non-goals)."
"The earlier draft proposed EmissionConfig.target_kind: ShapeAVariant | ShapeBVariant (closed-axis sum) — that was a substrate-modeling error: putting Shape B inside compiler-emit substrate conflates user-program-artifact-emission with compiler-target-emission. Shape B emission is its own lane (post-R3 user-emission work), separate from PB-6."

The earlier draft DID blur the Shape A/B boundary by proposing ShapeBVariant in EmissionConfig. The fix:

1. Removed ShapeBVariant from EmissionConfig surface entirely; EmissionConfig is Shape-A-only by construction.
2. Reframed Shape B explicitly as user-space artifact emission (standalone .dag programs walking typed values), NOT compiler-emit substrate.
3. Q4 retired from open-ratification status; resolution captured for traceability about why the earlier draft was wrong.

The Shape A/B boundary is now structurally preserved at §8.1/§8.2 — Shape A = compiler targets emitting Rust/Python/Go (PB-6 scope); Shape B = user .dag programs (NOT PB-6 scope; separate post-R3 user-emission work).

This was the final queued inline BLOCKING per relay note. All 5 codex findings + 3 cursor findings (8 total across both BLOCKING reviews) addressed across commits d55989d + 15cae2a + ac48b5f. Dashboard re-runs on new HEAD will evaluate the post-fix state.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex REQUEST_CHANGES absorbed — verdict correct, my prior Director-recommend on §12 Q7 was wrong. Fixed in commit 9768875.

The substantive flip:

Prior Director-recommend in §12 Q7 was option (b) — keep plain Dag + runtime UninferredPortPresent gate. The reasoning I gave ("violates feedback_no_metadata_markers adjacent") was wrong. feedback_no_metadata_markers is about __is_X STRING markers ("model concepts structurally in std/") — NOT about typed-state wrappers. InferredDag is structural typed-state, NOT a metadata marker.

The correct discipline references are:

• docs/modeling-discipline.md Practice 2 (illegal states unrepresentable)
• docs/modeling-discipline.md Practice 6 (API-level enforcement over convention)
• feedback_state_space_vs_behavioral_invariants (type enforcement > API enforcement)

All three argue for compile-time typed-state at emit's signature boundary. Codex's REQUEST_CHANGES is correct: locking in runtime-gate would make later correction harder, particularly because this doc is the Step-1 substrate authority for the pipeline-slot declaration.

Fix in commit 9768875:

• §3.1 (line 50) reframed: "post-infer readiness must be modeled in the type, NOT enforced at runtime"; emit's signature is fn emit(d: InferredDag, s: LanguageSpec) -> EmissionResult

• §7.1 (line 169) reframed: post-infer readiness modeled via typed-state carrier per Modeling Practices 2 + 6

• §7.3 (line 181) reframed: stage boundary enforcement is API-level (Practice 6), NOT convention-level

• §12 Q7 reframed from (a)/(b) binary "carrier vs runtime" to a 3-way choice on CARRIER SHAPE:

  • (a) InferredDag newtype (simple wrapper)
  • (b) refined-Dag via where-clause/predicate (structural refinement; reuses substrate refinement machinery)
  • (c) sum-variant Dag = PreInferDag | InferredDag (closed-axis state machine)

  New Director-recommend: option (b) refinement-via-where-clause if refinement substrate at HEAD (most aligned with feedback_groundedness_gates_lenses + language's first-class structural refinement); option (a) newtype as transition shape if refinement substrate not ready. Option (c) sum-variant is most explicit but has higher cross-codebase refactoring cost.

  Whichever shape operator picks, the constraint stands: emit's signature accepts only post-infer typed-state by construction; the runtime UninferredPortPresent framing is retired.

The doc is now coherent with docs/modeling-discipline.md Practices 2 + 6 + state-space-invariants discipline. Substantive learning: my reach for feedback_no_metadata_markers was pattern-match-without-grep against the wrong feedback entry; the actual feedback that applied (feedback_state_space_vs_behavioral_invariants) argues the OPPOSITE direction. This is the kind of error that BLOCKING reviews are FOR.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 97688750 · Trigger: schedule
• Thinking: 254s wall

---

BLOCKING (1)

Root Cause

• docs/design-emit-stage-l25-model.md Q7's typed-state reversal was applied to the narrative but not propagated into the concrete pipeline-slot contract → update every emit signature/table row to the chosen typed-state carrier before ratification.

⚠️ One stale concrete signature still preserves the prior stage-boundary bug.

[briansrls]

Duplicate of earlier inline BLOCKING relay (cited line:109 here vs line:103 earlier — both reference the same L4/L5 byte-equality finding from codex's first review pass at commit b6c3475). Already addressed in commit 15cae2a; the fix is currently at §7.2 line 177 of HEAD (commit 9768875):

"R3 Verification Mgr's L4 emit/eval match gates + L5 cross-target consistency gates consume emit output for typed semantic verification (compile the emitted code, evaluate against canonical inputs, compare evaluation results) — NOT byte-equality of source text. Cross-target source bytes differ by construction (Rust vs Python vs Go syntax); the verification is at the semantic/behavioral layer. Byte equality is reserved for PB-6 Step 4 parity verification only — see §9 Step 4: emit.rs hand-Rust output vs emit.dag substrate output for the SAME target. Same-target byte identity is the discriminator that fails for paper-shrink class."

This matches docs/design-cross-target-equivalence.md + docs/r3-structure.md L4/L5 semantic-equivalence framing.

No additional commit needed — relay arrived after fix was pushed. Already replied at /issuecomment-4448141132.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Duplicate relay of P5 receipt finding (cited line:126 here vs line:120 earlier — both reference the same parity-test P5 finding from codex's first review pass at commit b6c3475). Already addressed in commit 15cae2a; the fix is currently at §9 Step 4 line 206 of HEAD (commit 9768875):

"Parity verification authored as .dag TestClaim... P5 dissolution receipt: this TestClaim is transient-by-construction; it dissolves when emit.rs deletes in the same PR (Step 4 = parity + simultaneous deletion). Any hand-Rust scaffolding required for invocation routing... bears P5 receipt: parity_emit_dag_vs_rust_scaffolding — transient; dissolves with emit.rs deletion in same PR per Step 4 atomic discipline."

Already replied at /issuecomment-4448149300.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Duplicate relay of Shape B finding (cited line:155 here vs line:149 earlier — both reference the same Shape B finding from codex's first review pass at commit b6c3475). Already addressed in commit 15cae2a; the fix is currently at §3.3 line 68 + §12 Q4 RESOLVED line 261 of HEAD (commit 9768875):

§3.3 line 68: "EmissionConfig is Shape-A-only by construction — Shape B is user-space artifact emission... NOT a compiler-emit dispatch axis and does NOT belong in PB-6 EmissionConfig surface"
§12 Q4 line 261: "RESOLVED — removed from PB-6 substrate"

Already replied at /issuecomment-4448151362.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Codex INLINE BLOCKING (line:204, "Step 2 still dispatches fn emit(inferred: Dag, ...) even though §3.1/§7/Q7 require InferredDag — Practices 2 and 6 remain convention-level at worker-brief handoff") — substantive finding; fixed in commit 36c2e5c.

The earlier Q7 flip commit (9768875) updated §3.1 / §7.1 / §7.3 / §12 Q7 to require typed-state carrier but missed the §9 Step 2 row signature. The plain Dag in Step 2 would leak through to worker-brief authoring + dispatch — convention-level at the handoff, exactly the failure mode you flagged.

Fix at line 204 (commit 36c2e5c):

"Step 2: Pipeline slot | fn emit(d: InferredDag, spec: LanguageSpec) -> EmissionResult declared in compiler.dag with ExternalRealization body (Rust-backed placeholder pointing to current emit.rs). Exact InferredDag carrier shape gated on §12 Q7 operator ratification — newtype / refined-Dag-via-where-clause / sum-variant Dag = PreInferDag | InferredDag. Whichever shape lands, the signature accepts only post-infer typed-state by construction (Modeling Practice 6 API-level enforcement)."

The Step 2 worker-brief authoring is now structurally gated on Q7 ratification — Director cannot author a Step 2 brief that uses plain Dag signature; the carrier-shape decision must be ratified first.

Substantive learning logged: when a discipline change ripples across a doc, audit ALL signature/contract mentions in subsequent sections, not just the framing sections. The §9 Step 2 row was the worker-handoff contract; missing the update there leaves the discipline as a doc-statement-only with no enforcement at the actual dispatch boundary.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

This is the high-level BLOCKING summary corresponding to the inline finding at §9 Step 2 (line:204). Codex reviewed commit 9768875 and caught the Q7-narrative-vs-Step-2-signature leak. Fixed in commit 36c2e5c (already pushed):

§9 Step 2 row now reads: fn emit(d: InferredDag, spec: LanguageSpec) -> EmissionResult with note that exact carrier shape (newtype / refined-Dag-via-where-clause / sum-variant) is gated on §12 Q7 operator ratification.

Already replied at /issuecomment-4448512454 with the full fix context.

Dashboard re-review on new HEAD (36c2e5c) should retract the REQUEST_CHANGES.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: fa5bdfa3 · Trigger: schedule
• Thinking: 231s wall

---

BLOCKING (1)

Root Cause

• docs/design-emit-stage-l25-model.md §5.3 is keyed to stale design-clean-emission-contract.md framing rather than the live src/v3/std/clean_emission.dag authority → enumerate or mechanically delegate to the live CleanEmissionContract shape before ratification.

⚠️ One live CleanEmissionContract field is missing from the PB-6 model, so the design should be corrected before it becomes the Step 2 authority.

[briansrls]

High-level BLOCKING summary of the inline CleanEmissionContract finding (line:126). Codex reviewed commit fa5bdfa; substantive fix already pushed in commit 7bbc6b8.

§5.3 now enumerates the 9 typed fields of CleanEmissionContract directly from live substrate at src/v3/std/clean_emission.dag:13, including variant_payload_field_access (the missing field codex flagged). §5.4 PostEmitVerifier correctly anchored as a FIELD of CleanEmissionContract (not a separate top-level substrate). §12 Q5 retired from open-ratification (enumeration is now live in §5.3, not deferred to Step 2 worker brief).

Already replied at /issuecomment-4448858002 with full fix context.

Dashboard re-review on new HEAD (7bbc6b8) should retract the BLOCKING.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 96bc7d18 · Trigger: schedule
• Thinking: 239s wall

---

BLOCKING (2)

Root Cause

• docs/design-emit-stage-l25-model.md §4 treats T-Ground-Diagnostic as future instead of reading the live diagnostics substrate → make src/v3/std/diagnostics.dag the authority and carry its full EmissionDiagnostic shape or route amendments through that file.
• docs/design-emit-stage-l25-model.md §2/§9 retained the two-argument emit shape after §3.3 introduced a target-selection carrier → choose one target-selection authority and align every emit signature/input list with it.

⚠️ Two substrate-boundary mismatches should be fixed before this document becomes the Step 2 authority.

[briansrls]

BLOCKING: The EmissionDiagnostic sketch drops the live src/v3/std/diagnostics.dag:176 carrier variants, creating a parallel diagnostic authority and violating INVARIANTS P2 / Modeling Practice 3.

[briansrls]

BLOCKING: The Step 2 signature drops EmissionConfig even though §3.3 declares target selection as a substrate input, splitting target authority and violating INVARIANTS P2 / Modeling Practice 3.

[briansrls]

Codex INLINE BLOCKING (2 findings on commit 96bc7d1) — both substantive; fixed in commit 6a9d35c.

Finding 1 (T-Ground-Diagnostic as future vs live):

Verified via grep -n "type EmissionDiagnostic" src/v3/std/diagnostics.dag → line 201. Live substrate authority already declares 7+ variants (UnderRefined / NoInhabitant / ContradictoryUse / OutOfR2Scope / MissingEmissionPath / 2 🟡 TRANSITIONAL: LifetimeProgramExtractionPending + FoldNotImplemented).

Fix:

• §4.2 reframed to cite src/v3/std/diagnostics.dag:201 as the carrier-of-truth + carry full variant set forward (per INVARIANTS P2 + Modeling Practice 3 facts-carry-forward); preserve 🟡 TRANSITIONAL variants with retirement-trigger documentation.
• §6 T-Ground-Diagnostic row updated: carrier is LIVE; what's pending is the emission-path CONSUMPTION + 🟡 TRANSITIONAL retirement triggers, not the carrier authoring. Earlier "NOT-STARTED" framing was wrong.

Finding 2 (§2/§9 2-arg vs §3.3 introduces 3rd EmissionConfig input):

The earlier draft §3.3 introduced EmissionConfig as a 3rd input but §2 + §9 Step 2 both use 2-arg signature. Substrate-boundary mismatch.

Fix: dropped EmissionConfig as redundant. Target selection IS via LanguageSpec choice — the per-target spec instances at src/v3/spec/{rust,python,go}.dag carry target identity. Passing rust_spec = emitting to Rust; no separate config carrier needed.

• §3.3 reframed: "Target selection — IS via LanguageSpec, NOT a separate EmissionConfig carrier"
• §3 header: "Two input types feed emit" (was "Three")
• §6 prereq table: emit_config.dag row marked DROPPED with explanation
• Signature fn emit(d: <InferredDagCarrier>, spec: LanguageSpec) -> EmissionResult now consistent across §2 / §3 / §9

Both substrate-boundary mismatches resolved. The doc no longer drops live diagnostics facts (Practice 3) and no longer introduces a redundant 3rd-input that doesn't appear in §2/§9 signatures.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Inline version of Finding 1 from the BLOCKING summary at /issuecomment-4448999... (cited line:89 here vs §4 root-cause earlier). Already addressed in commit 6a9d35c:

§4.2 now cites src/v3/std/diagnostics.dag:201 (live substrate) as the carrier-of-truth + carries full 7+ variant set forward (UnderRefined / NoInhabitant / ContradictoryUse / OutOfR2Scope / MissingEmissionPath / 2 🟡 TRANSITIONAL: LifetimeProgramExtractionPending + FoldNotImplemented). No parallel authority — diagnostics.dag is canonical. §6 prereq table row updated accordingly.

Already replied at /issuecomment-4449221143 with full fix context.

— sent from zesty-bear-812 (gunbc Director)

[briansrls]

Inline version of Finding 2 from the BLOCKING summary. Cited line:208 here vs §9 Step 2 row in earlier summary. Already addressed in commit 6a9d35c — same direction codex flagged:

• §3.3 reframed: target selection IS via LanguageSpec choice (per-target spec at src/v3/spec/{rust,python,go}.dag); no separate EmissionConfig carrier.
• §3 header: "Two input types feed emit" (was "Three").
• §9 Step 2 signature: fn emit(d: <InferredDagCarrier>, spec: LanguageSpec) -> EmissionResult — consistent with §2 + §3 framing.
• §6 prereq table: emit_config.dag row marked DROPPED with explanation.

Target authority is now single-source (LanguageSpec carries target identity); no parallel authority split. INVARIANTS P2 + Modeling Practice 3 satisfied.

Already replied at /issuecomment-4449221143 with full fix context.

— sent from zesty-bear-812 (gunbc Director)