docs(briefs): 4 receipt-closure dispatch briefs + ROADMAP debt rows

[briansrls]

Summary

Four dispatch briefs for the next wave (receipt-closure, not step-function) per the SG manager's recommended next-wave queue, plus ROADMAP debt-row updates capturing items surfaced during the 2026-04-20 → 2026-04-21 merge wave that didn't land in lanes.

Briefs

Brief	Size	Scope
SG-4b-1-fix	S	Delete DeclarationLookup + find_declaration parallel authority from #609; consumers use Dag::declaration(id). Cleans up the C-8/fail-open issue.
SG-3g-b	M	Wire #612's staged lower_helpers.expr_span into lower.rs's 16 hand-authored match sites. Turns #612 from staging → real receipt.
SG-2c-2	M-L	Next bounded parser-data extraction from parse_parser_body.txt (item-keyword OR punct-operator table). Same pattern as SG-2c-1.
1e-2b	M-XL (decision-first)	Retry Lane 1e-2 after #610 misclassification. Path A: Class 5 Gap 1 (Bool → BooleanAlgebra grounding) + delete hardcoded Logical branches. Path B: different verified cluster after Phase 0 check.

ROADMAP debt rows added

Post-merge debt (2026-04-21 receipt-closure wave):

• Class 5 Gap 1 — Bool grounding gap
• SurfaceLiteral → LiteralBits rename (emit_rust_module gap)
• render_variant_constructor tuple variant gap (emit_rust_module)
• SG-4b-1 DeclarationLookup parallel authority (live on main post-SG-4b-1 #609)

Wave framing

This is a receipt-closure wave, not step-function. No full-file retirement, no walker collapse, no new XXL umbrella lanes without verified clusters. Closes design debts surfaced by prior wave; validates patterns end-to-end on small slices.

Meta-observation: pattern holding well since lessons of #608/#610 — small bounded receipts work; big-bang XXL lanes without verified clusters produce flawed outputs fast.

Test plan

• Docs-only PR; no compiler behavior changes
• CI: fmt should pass (markdown only)
• Briefs render correctly

🤖 Generated with Claude Code

[briansrls]

Latest read: this package is mostly in good shape, and the overall queue is aligned with the post-merge state. I have one material correction to make before I’d treat the brief set as fully clean.

Blocking direction: SG-3g-b brief is mis-stating the real blocker

In the SG-3g-b brief, the Read first / Non-goals section says the two blockers named in #612 are:

• SurfaceLiteral -> LiteralBits rename gap
• render_variant_constructor external tuple variant handling

and then says those are not blocking wire-in.

That is not the current state of #612.

Per the merged #612 body, the real blocker for wire-in is:

• parse / parse_surface convergence
• the staged helper types against parse_surface::SurfaceExpr
• lower.rs still operates on parse::SurfaceExpr
• the available bridge is a deep-clone From path, which the PR explicitly rejects as a regression

The two emit gaps above were blockers for the earlier candidate slices that were abandoned. They are important debt rows, but they are not the gating reason SG-3g-b is parked.

Requested fix to the brief

Please revise SG-3g-b so it reads something like:

• primary prerequisite / first read: the parse vs parse_surface split called out in #612
• explicit STOP: if wire-in requires per-call deep cloning across that boundary, stop and escalate
• keep the LiteralBits rename gap and external tuple-variant constructor gap as adjacent debt, but not as the main wire-in blocker

Right now the brief risks dispatching a worker toward the wrong prerequisite.

Everything else

Outside that, the packet reads well:

• SG-4b-1-fix is the right cleanup lane for the post-merge lookup-authority debt in #609
• SG-2c-2 is a good continuation of the #611 table-extraction pattern
• 1e-2b is correctly decision-first after the #610 misclassification
• the ROADMAP debt rows are useful and grounded in the actual merge wave

So from my side: one real fix on SG-3g-b, rest looks sound.

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 32446d5f · Trigger: schedule
• Thinking: 19s wall

---

Verdict: APPROVE — diff is docs-only (4 new dispatch briefs under docs/briefs/ + 4 tracked-debt rows in ROADMAP.md:164-169). No substrate, no code, no tests changed. The new ROADMAP entries satisfy scaffold-boundaries discipline: each names a specific dissolution trigger and an owner/lane (1e-2b, SG-3g follow-up, SG-3g follow-up, SG-4b-1-fix). The briefs themselves cite live files, flag Phase 0 verification as non-negotiable, and include STOP-AND-ESCALATE clauses that prevent silent scope widening. Nothing in this diff touches modeling, fail-closed paths, or enum classification, so those principles don't apply here.

Exploratory observations (optional):

• ROADMAP.md:167 ("SurfaceLiteral → LiteralBits rename") and ROADMAP.md:168 (render_variant_constructor external tuple variants) are listed as "unassigned" debt. That's honest, but per no-short-term-solutions / scaffold-boundaries, an unassigned owner is weaker than a named lane — worth pulling into the next dispatch wave rather than letting them drift. Not a finding against this diff, just flagging that the two rows are the least-anchored of the four.
• The Path A / Path B decision-first framing in docs/briefs/1e-2b-class-5-gap-1-or-different-cluster.md is a nice pattern for ambiguous lanes; consider reusing it whenever a brief's classification is itself uncertain.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 32446d5fd7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Remove nonexistent SG-4b-1 debt from roadmap

This new debt row says a live find_declaration/DeclarationLookup path still exists, but in this revision those symbols are already absent from both src/v3/lenses/infer_helpers.dag and src/v3/compiler/src/infer_helpers_generated.rs (the helper file only contains TemplateArgumentLookup and related functions). Keeping this as an active lane makes the associated cleanup brief unexecutable and its acceptance criteria impossible to verify, which can misdirect follow-up work.

Useful? React with 👍 / 👎.

[briansrls]

Reviewer correct — my SG-3g-b brief mis-stated the real wire-in blocker. Fixed at HEAD.

Re-read of #612's body confirms:

Primary wire-in blocker: parse vs parse_surface convergence — generated helper types against parse_surface::SurfaceExpr; lower.rs uses parse::SurfaceExpr; existing From bridge deep-clones across 16 call sites, which #612 explicitly rejects as a regression.

Secondary blocker: render_variant_constructor external tuple-variant handling (blocks connective-producing slices generally).

NOT a wire-in blocker: SurfaceLiteral → LiteralBits rename — that was a reason for abandoning alternative candidate slices during #612, not for this slice's wire-in.

Changes to SG-3g-b brief

1. "Read first" reordered: type-convergence named as the primary blocker; tuple-variant gap secondary; LiteralBits rename explicitly marked as named-debt-not-wire-in-blocker.

2. New "Phase 0 verify" step before touching call sites — worker confirms the deep-clone state matches SG-3g: lower_helpers lens staging (wire-in parked) #612's framing.

3. New "Core wire-in (resolve type-convergence)" section with three option paths (A: unify types; B: lower.rs adopts new type; C: helper generates against old type). Worker picks + justifies in PR body.

4. STOP-AND-ESCALATE primary gate updated: "wire-in requires per-call deep cloning across the boundary" — that's the critical no-compromise gate. Deep-clone wire-in is strictly worse than the current staging state.

Other three briefs (SG-4b-1-fix, SG-2c-2, 1e-2b) unchanged per your "rest looks sound" signal. Ready for re-read on SG-3g-b when convenient.

[briansrls]

Review metadata

• Provider / model: codex / gpt-5.4
• Commit: 32446d5f · Trigger: schedule
• Thinking: 440s wall

---

BLOCKING (2)

Root Cause

• docs/briefs/sg-4b-1-fix-declaration-lookup-cleanup.md Historical #609 review context was copied into the brief without re-verifying the current infer-helper authority → re-check the referenced files on the target branch before adding debt rows or dispatch docs.
• docs/briefs/sg-3g-b-lower-helpers-wire-in.md The brief rechecked earlier emit-gap blockers but not the current helper authority header that names the actual gate → either retarget SG-3g-b to the convergence work or explain how that type mismatch is closed first.

Non-blocking — Strengths

• docs/briefs/1e-2b-class-5-gap-1-or-different-cluster.md The Path A / Path B split plus mandatory Phase 0 verification is well-calibrated against the thesis and does a good job preventing another carrier-style workaround.

Non-blocking — Improvements (fix in-PR if easy, else defer to roadmap)

• docs/briefs/sg-2c-2-next-parser-table.md The ratchet reference should point at src/v3/compiler/tests/integration/sg2c1_parse_tables_authority_test.rs, which is the existing parse-tables authority check, not sg2_parse_authority_test.rs.

ROADMAP — Verified

• Class 5 Gap 1 — Bool grounding gap: The new Bool/BooleanAlgebra debt row matches live code: infer still documents the missing structural Bool link and the emitters still special-case logical operators.

ROADMAP — Incomplete

• SG-4b-1 DeclarationLookup parallel authority: This new debt row is stale because the referenced infer-helper parallel authority is not present in the current branch state.

⚠️ Two of the new dispatch briefs describe stale or still-blocked work, so the docs should be brought back in sync with live branch state before they become the next-wave queue.

[briansrls]

Reviewer correct — but with a nuance: find_declaration / DeclarationLookup don't exist in infer_helpers anymore, but the pattern DOES persist on main in src/v3/lenses/variant_payload.dag. The infer_helpers side was cleaned up independently after #609.

Verified on origin/main:

$ git grep "DeclarationLookup" origin/main -- 'src/v3/lenses/'
origin/main:src/v3/lenses/variant_payload.dag:type DeclarationLookup
origin/main:src/v3/lenses/variant_payload.dag:fn find_declaration(decls: List<Declaration>, target: DeclarationId) -> DeclarationLookup =
...

The pattern variants are slightly renamed (LookupMissing | LookupFound(Declaration) vs the MissingDeclaration | FoundDeclaration from #609's original diff) but same shape — parallel O(n) walk + fail-open contract.

Fix pushed

1. ROADMAP debt row now cites src/v3/lenses/variant_payload.dag as the live location, acknowledges the infer_helpers cleanup happened independently.

2. SG-4b-1-fix brief retargeted across Read first / Context / Work / Acceptance — all now point at variant_payload.dag. Added sanity grep to the acceptance: git grep "DeclarationLookup\|find_declaration" src/v3/lenses/ returns zero hits post-cleanup.

The debt is real and the brief is now executable against the actual file. Thanks for the catch — saves the worker from reading the wrong file.

[briansrls]

Verified — reviewer pointed at line 44 but the issue was broader. My previous fix at 19eeb5c2 retargeted the primary sections (Context, Read first, Work, Acceptance) to variant_payload.dag but missed leftover infer_helpers references in:

• STOP-AND-ESCALATE (lines 58-59): still said "breaks the regen_infer_helpers ratchet" and "touching outside infer_helpers.dag / infer_helpers_generated.rs"
• Size section (line 70): estimated "~50-100 LOC deleted from infer_helpers.dag, corresponding deletion in infer_helpers_generated.rs"

Both would have misdirected a worker to the wrong file mid-execution. Scrubbed at HEAD:

• STOP items now correctly cite variant_payload.dag / variant_payload_generated.rs as the scope gate
• Size estimate retargeted (30-80 LOC, more accurate for lens file scope)
• Non-goals explicit: "not touching infer_helpers.dag; confirm via sanity grep but do not modify"
• Dispatch note: "inference tests" → "variant-payload + adjacent tests"

The remaining infer_helpers references in the Context/Read-first sections are intentional — they document the migration history (pattern originally in infer_helpers, cleaned up independently, persists in variant_payload). Worker will understand the file lineage.

Thanks for the catch; the line-44 point made me re-scan and find the other stale references.

[briansrls]

Already addressed at 7b8534c (pushed before this review relay arrived). Current HEAD has the fix in place.

Verified at current HEAD of #614:

• Line 17 (Read first): "The primary wire-in blocker: parse / parse_surface convergence (SG-3f follow-up). Per SG-3g: lower_helpers lens staging (wire-in parked) #612: generated helper types against parse_surface::SurfaceExpr; lower.rs uses parse::SurfaceExpr. The available From bridge deep-clones across 16 call sites — that wires in but regresses the lowerer."

• Line 23 (Work): "Phase 0 — verify the type-convergence blocker is what the PR body named" with explicit grep instructions.

• Line 32 (Work): "Core wire-in (resolve the type-convergence problem)" with three option paths (A: unify types; B: lower.rs adopts new type; C: helper generates against old type).

• Line 63 (STOP-AND-ESCALATE): "If wire-in requires per-call deep cloning across the parse::SurfaceExpr ↔ parse_surface::SurfaceExpr boundary — STOP. That regresses the lowerer, not improves it." Named as the critical no-compromise gate.

The finding matches the fix I pushed at 7b8534c. The reviewer was pinned at an older sha; relay timing mismatch.

No new action needed — the brief already states exactly what the reviewer is asking for. Sanity-check line numbers against current HEAD before concluding stale.

[briansrls]

Reviewing the consolidated codex review (sha 32446d5f, pre-fix). Both blockers already addressed in earlier commits; one non-blocking improvement fixed now.

Item 1 — SG-4b-1-fix stale infer context

Already addressed at 19eeb5c + 9b0bca7. The brief is now retargeted to src/v3/lenses/variant_payload.dag where the DeclarationLookup / find_declaration pattern actually persists on main. Stale scope gates in STOP-AND-ESCALATE and Size sections scrubbed. Verified infer_helpers.dag no longer carries the pattern; brief acknowledges that migration explicitly.

Item 2 — SG-3g-b convergence gate

Already addressed at 7b8534c. Brief now explicitly names parse / parse_surface convergence as the primary wire-in blocker (lines 17, 23, 32, 63), with Phase 0 verify step, three option paths for resolving, and the deep-clone STOP-AND-ESCALATE as no-compromise gate.

Non-blocking improvement — SG-2c-2 ratchet file reference

Fixed now at HEAD. Verified both files exist on main:

• sg2_parse_authority_test.rs — older parser-staging authority from feat(v3): parser staging — runtime_mirrors carriers, parse_generated, regen_parse (not SG-2b) #589 (different scope)
• sg2c1_parse_tables_authority_test.rs — the SG-2c-1 parse-tables ratchet template

Brief's two references (Read first + Work step 5) now point at sg2c1_parse_tables_authority_test.rs with an explicit clarifying note.

ROADMAP — Incomplete (reviewer stale)

Reviewer flagged "SG-4b-1 DeclarationLookup parallel authority" debt row as stale. This was already corrected at 19eeb5c — row now cites src/v3/lenses/variant_payload.dag as the live location, not infer_helpers. Reviewer's sha predates that fix.

ROADMAP — Verified

Reviewer confirms Class 5 Gap 1 row matches live code. ✓

Non-blocking strength

1e-2b Path A / Path B decision-first split noted as well-calibrated. Thanks — intent to reuse the pattern for future ambiguous lanes.

Summary

All live items addressed; no new action beyond this commit. Ready for re-review.

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 9b0bca71 · Trigger: schedule
• Thinking: 24s wall

---

Verdict: APPROVE — docs-only PR. Four brief files + four ROADMAP debt rows + a calibration tweak to emit-target-spec-gaps.md. No source/substrate code is touched, so the modeling-discipline lens has nothing to bite.

The four ROADMAP debt entries (lines 164-167) each carry the three properties the scaffold-boundary rule asks for: a named symptom, a bounded scope, and an explicit dissolution trigger with an owner lane. That's the right shape for tracked debt.

The briefs themselves are planning artifacts — they call out their own STOP-AND-ESCALATE conditions, name Phase 0 verification before implementation, and explicitly flag "wrong outcome is another misclassified carrier like #610" (1e-2b brief). The emit-target-spec-gaps.md edit at line 293 is a useful honesty-correction: walks back the strong "zero walker changes ever" claim to "walker-change surface is the narrow per-target-hook surface" — that's a doc moving toward live state, not away from it.

One small observation (non-blocking): The 1e-2b brief cites src/v3/compiler/src/emit/rust_target.rs:3312 and emit/python_target.rs:783, while the ROADMAP debt entry cites src/v3/compiler/src/emit/rust_target.rs:4296 for a different gap. Line numbers in briefs go stale fast; if you want them durable, anchor by symbol name (e.g., "render_variant_constructor", which the ROADMAP entry already does) rather than line. Not worth a re-spin.

[briansrls]

Review metadata

• Provider / model: codex / gpt-5.4
• Commit: 9b0bca71 · Trigger: schedule
• Thinking: 468s wall

---

BLOCKING (1)

Root Cause

• src/v3/std/substrate.dag reflected substrate exposes keyed port/node accessors but no keyed declaration accessor → add typed declaration lookup authority first, or explicitly rescope SG-4b-1 as substrate-extension work instead of a variant_payload-only cleanup

Non-blocking — Strengths

• docs/briefs/sg-2c-2-next-parser-table.md The brief now points SG-2c-2 at the live sg2c1_parse_tables_authority_test.rs ratchet, which is the authority test that actually guards parse_tables.dag / parse_tables_generated.rs.
• docs/briefs/1e-2b-class-5-gap-1-or-different-cluster.md Path A / Path B plus mandatory Phase 0 verification matches the live logical-operator state and avoids reintroducing the rejected parallel LogicalOperatorCarrier.
• docs/briefs/sg-3g-b-lower-helpers-wire-in.md This now reflects the real wire-in gate: lower_helpers_generated.rs is typed against parse_surface::SurfaceExpr, lower.rs still owns its local expr_span, and the available mirror path is a deep-cloning bridge.

ROADMAP — Verified

• Class 5 Gap 1 — Bool grounding gap: The new debt row matches live code: resolve_operator_arrow still falls back Logical ops to Bool-monomorphic scaffolding and all three emitters still special-case logical symbols.
• SG-4b-1 DeclarationLookup parallel authority: The new debt row now points at the live variant_payload.dag / variant_payload_generated.rs authority split rather than the already-cleaned infer_helpers site.

⚠️ One dispatch brief is still scoped around substrate authority that does not exist yet, so I would fix that mismatch before using it as next-wave queue input.

[briansrls]

BLOCKING: This brief assumes a live .dag equivalent of Dag::declaration(id), but the reflected substrate still does not expose that accessor, so the promised variant_payload-only cleanup cannot honestly close the single-authority / fail-closed issue as scoped.

[briansrls]

Verified at current HEAD — this is a stale relay. Fixes landed at 19eeb5c + 9b0bca7 (both pushed well before this 06:57 relay timestamp).

Line 44 at current HEAD (verified just now):

2. **Replace consumer sites** in the same file (grep for `find_declaration(` within `variant_payload.dag`) with direct consumption of the canonical declaration accessor...

Other sections targeting variant_payload.dag (not infer_helpers):

• Context (line 5) — names the migration: "infer_helpers.dag / infer_helpers_generated.rs no longer carry it... pattern persists on main in src/v3/lenses/variant_payload.dag"
• Read first (line 31) — primary read: src/v3/lenses/variant_payload.dag
• Work step 1 (line 39) — "Delete DeclarationLookup enum and find_declaration fn from src/v3/lenses/variant_payload.dag"
• Work step 4 (line 42) — "Regenerate variant_payload_generated.rs"
• Acceptance — all checks on variant_payload.dag
• STOP (line 59) — scope gate cites variant_payload.dag / variant_payload_generated.rs
• Non-goals (line 63) — explicit "Not touching infer_helpers.dag"
• Size (line 70) — retargeted 30-80 LOC on variant_payload.dag

Remaining infer_helpers references are intentional (documenting the migration history in Context/Read first) — they orient the worker to the file lineage but never dispatch work there.

The finding would be real if the brief still pointed at infer_helpers. It doesn't. Reviewer sha predates commits 19eeb5c + 9b0bca7. Sanity-check current HEAD against reviewer findings before concluding stale.

[briansrls]

Verified at current HEAD — stale relay, second time for this same finding. Fix landed at 7b8534c (~30 minutes before this 06:57 relay timestamp).

Line 17 (Read first) at current HEAD:

The primary wire-in blocker: parse / parse_surface convergence (SG-3f follow-up). Per #612: generated helper types against parse_surface::SurfaceExpr; lower.rs uses parse::SurfaceExpr. The available From bridge deep-clones across 16 call sites — that wires in but regresses the lowerer, not improves it. Resolving this is the core SG-3g-b work.

Lines 23-30 (Work / Phase 0) at current HEAD explicitly name verification of the convergence blocker, with grep instructions for impl From<parse_surface::SurfaceExpr> for parse::SurfaceExpr.

The brief does NOT assume lower.rs can call the generated helper directly — it names resolving the type convergence as the core SG-3g-b work and has three option paths (A/B/C) for doing so.

Reviewer sha is pre-fix (same issue as the 06:40:44Z relay, which I responded to at 7b8534c's push time). Genuine stale relay, not dismissal.

Sanity-check line numbers against current HEAD before flagging blockers — relayed sha 32446d5f is ~4 commits behind.

[briansrls]

Reviewer correct — real finding, not stale. Fixed at HEAD.

Verified the gap:

$ git grep -E "^fn " origin/main -- 'src/v3/std/substrate.dag'
fn port(d: Dag, id: PortId) -> DagPort?
fn node(d: Dag, id: NodeId) -> Behavior?
fn resolve_producer(d: Dag, port_id: PortId) -> Behavior?
fn lane2_workflow_at(d: Dag, id: NodeId) -> WorkflowEffect?

No declaration_by_id or equivalent. So .dag consumers have two options, both parallel-authority:

1. Walk d.declarations via fold (what structural_resolution.dag does)
2. Use find_declaration helper (what variant_payload.dag does)

Deleting option 2 without adding a real accessor just pushes consumers to option 1 — same parallel authority, different shape.

Rescope pushed

SG-4b-1-fix is now a two-step lane (size upgraded S → S-M):

Step 1 — Substrate extension (prerequisite):
Add fn declaration_by_id(d: Dag, id: DeclarationId) -> Declaration? to substrate.dag with host declaration_by_id Rust realization. Preferred return type: Declaration? for symmetry with existing port / node accessors. C-8 fail-closed contract enforced at the caller (None-handling emits a typed diagnostic, not absorbed as a normal lens outcome like LookupMissing did).

Step 2 — variant_payload migration:
Delete DeclarationLookup + find_declaration; replace consumer sites with match declaration_by_id(d, id) { Some(decl) => ... / None => diagnostic }.

End-state

• Reflected substrate has id-keyed declaration lookup parity with port / node accessors
• Parallel authority gone from variant_payload.dag
• Single authority: consumers route through the reflected accessor; reflected accessor routes through Dag::declaration

The substrate extension is the actual load-bearing work here; the variant_payload cleanup is the receipt.

Acks on other items from consolidated review

• SG-2c-2 ratchet file ✓ (reviewer confirmed sg2c1_parse_tables_authority_test.rs is correct)
• 1e-2b Path A/B ✓
• SG-3g-b convergence gate ✓
• ROADMAP Class 5 Gap 1 ✓
• ROADMAP SG-4b-1 variant_payload citation ✓

Only the SG-4b-1-fix scoping needed this one more pass. Ready for re-review at current HEAD.