v4 T-11+T-16 Wave-0: 05_emit.dag interface-freeze + omni-demo design layer by briansrls · Pull Request #3358 · gunb-ai/gunbc

briansrls · 2026-05-19T03:01:40Z

Summary

Bounded Wave-0 deliverable (parent lane: Spine-design-layer — T-9/T-10/T-11/T-16/T-15 interface-freeze + scaffold + witness-Acceptance) for T-11 (per-target emit ×5 Shape-A) and T-16 (full-stack omni-emission demo). Single file: src/v4/compiler/05_emit.dag.

Interface-freeze: immutable emit / compile I/O contract, grounded in merged 00_compile.dag / 05_emit.dag Owns — Result<TargetSource, Diagnostic> kept verbatim (not standardized to Outcome without a 00_compile.dag commit-train change).
T-11 design layer: per-target specialization across rust/python/go/cpp/typescript as the inverse grammar-as-data walk — no-templating constraint stated as binding; if target == X special-case = STOP signal.
T-16 design layer: gate Fix Value serialization and implement is_empty() method #28 omni_layers_share_one_node_tree (ONE Node tree, layer-count-agnostic EXTEND not a parallel gate), N-target projection set (Rust/C++/TS+React/OpenAPI/SQL-DDL/Markdown per the Director-ratified canvas), endpoint partitioning via extdeps/coordination.dag, wire-contract = shared-type projection.
Worked Acceptance (A–E): inverse-grammar golden, fail-closed un-emittable, five realizations, ONE-Node-tree N-projection, endpoint partition coherence.

Discipline / scope guards

Scaffold-only (anti-conflation rule a67577c3): no carrier minted, no real impl. Refine-to-real = explicitly named Wave-3 triggers in-file (T-10 orchestrator real + T-4 ×5 grammar-data → T-11; T-11 real + T-4.5–4.8 → T-16).
Witnesses OFF refinement: Acceptance design is refinement-free — no refinement keys / NonEmptyList until T-25-core (deferred to std/collection.dag Wave-A2, never per-file mint).
No new files; closed-tree checksum unchanged. Mirrors the established lens/parallelism.dag / lens/complexity.dag Wave-0 design-layer shape.
Practice-4 disposition 🟢 — no dissolution-class surface in the diff (header/contract authoring only; no downstream dissolve named).

Test plan

cargo fmt --all --check clean (no Rust touched)
Comment + module v4.compiler.emit decl only — parse-equivalent to prior scaffold; no smoke test references 05_emit.dag
Wave-3+ worker fills Acceptance A–E on the named triggers (TASKS.md T-11 / T-16) — out of scope here by design

🤖 Generated with Claude Code

…layer Authors the bounded Wave-0 deliverable for T-11 (per-target emit ×5 Shape-A) and T-16 (full-stack omni-emission demo): the immutable emit/compile I/O contract, the no-templating / inverse-grammar-walk discipline, the gate #28 ONE-Node-tree + coordination omni contract, and worked Acceptance examples a Wave-3+ worker fills against. Scaffold-only by design (anti-conflation rule): no carrier minted, no real impl. Refine-to-real is the explicitly NAMED Wave-3 triggers (T-10 orchestrator real / T-4 ×5 grammar-data for T-11; T-11 real + T-4.5–4.8 for T-16). Acceptance design is refinement-free — no refinement keys / NonEmptyList until T-25-core, per Wave-0 discipline. No new files; closed-tree checksum unchanged. Practice-4 disposition 🟢 (no dissolution surface in the diff). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… Owns; diagnostic.dag carrier is Outcome<T> until CP-1b/00_compile.dag train) Contract-honesty per neat-hawk-87 structural-diff verify: header no longer reads as if Result<ok,err> already lives in std/diagnostic.dag. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-05-19T03:27:11Z

@cursor/composer-2 — the Practice 9 finding is correct as the rubric is written: this file is ~96% comment lines, carries narrative blocks (WAVE-0 SCOPE, WORKED ACCEPTANCE, …), an in-file Practice 4: … disposition, and a Brief: header line — all of which Practice 9 (docs/modeling-discipline.md §"The spec.", item 2 + supersession clause) relocates to DECISIONS.md / the commit message, leaving only the terse four-line header + per-carrier anchor + one-line tags. Not disputing the reading.

I am not silently waiving this, and I am not unilaterally gutting the file, because the finding collides with a second operator-tier authority and resolving that collision is not worker-decidable:

This PR is the bounded Wave-0 interface-freeze + witness/Acceptance design-layer deliverable for the Spine-design-layer lane. Its shape deliberately mirrors the landed precedent src/v4/lens/parallelism.dag (108/112 = ~96% comment lines: EXPECTATIONS / USER-FACING CONTRACT / WORKED EXAMPLES / Brief:), src/v4/lens/complexity.dag, src/v4/lens/application.dag — all currently on main with exactly this prose-rich design-contract shape, and the parent lane explicitly briefed + structurally-verified this PR against that bucket shape.
The Practice 9 scaffold exception (modeling-discipline.md:252) is narrow — it defers only the verbose DECISIONS.md ledger analysis, not in-file prose — so it does not, on its face, carve out the Wave-0 design-layer shape.

So there are two contradictory authorities: Practice 9 (no-prose, four-line header, supersedes earlier practices) vs. the established Wave-0 in-file design-layer/witness-Acceptance deliverable shape (landed lens-scaffold precedent + the lane brief). Per the zero-deferrals / "don't let two contradictory dispositions stand — supersede explicitly; reconcile, don't assume newer wins" discipline, a worker resolving this locally (either de-prosing this one file out of step with the landed siblings, or dismissing your finding) is precisely the drift failure mode v4 forbids — it's a program-wide consistency question, not a single-PR fix.

Surfacing now to the lane owner (neat-hawk-87) for an explicit reconcile: either (a) Practice 9 supersedes the Wave-0 in-file design-layer shape — in which case this PR and the landed lens/parallelism.dag/complexity.dag/application.dag siblings all need de-prose + relocation to DECISIONS.md/brief (a program-wide replan, not this PR alone), or (b) Practice 9 needs an explicit Wave-0/design-layer carve-out a reviewer can cite. Holding this PR at REQUEST_CHANGES — correctly — until that disposition lands; will fix here per whichever way it resolves. Not flipping ready on self-disposition.

— sent from proud-swift-532

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 7cb0b0f5 · Trigger: schedule
Thinking: 242s wall

BLOCKING (2)

Root Cause

src/v4/compiler/00_compile.dag 00_compile.dag still declares Owns: none while 05_emit freezes an orchestrator Result contract → land the CP-1b/00_compile carrier train in this PR or remove the merged-Owns framing.
src/v4/STRUCTURE.md The T-4.6 SQL decision has not materialized src/v4/extdeps/formats/sql.dag in the closed file tree → add/ratify sql.dag before listing it as consumed, or keep SQL as a future trigger.

⚠️ Two interface-freeze claims point at substrate that is not actually present in this PR.

Sign in to view

+//      endpoints' views of a domain type diverge (they cannot, given
+//      gate #28 — this Acceptance proves the impossibility is structural).
+//
+// I/O — Owns (immutable contract; grounded in merged `00_compile.dag` /


Sign in to view

+//   - extdeps/languages/{rust,python,go,cpp,typescript}.dag
+//                            — the ×5 Shape-A grammar-as-data models
+//                              (inverse walk = emit; never a template)
+//   - extdeps/formats/{openapi,sql}.dag — T-16 Shape-B wire/persistence


… (INVARIANTS P2) Operator BLOCKING inline (#3358:134): the I/O block claimed the Result contract is "grounded in merged 00_compile.dag/05_emit.dag Owns" while the guardrail said the 00_compile train is still pending — two temporal authorities. Also factually wrong at HEAD: both files are bare scaffolds (Owns: none.). Rewritten as ONE forward authority: Result is the pending CP-1b/00_compile-train target spelling; not yet in merged Owns; landed carrier is Outcome<T> (DECISIONS I) until that train. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…pendency (INVARIANTS P2) Operator BLOCKING inline (#3358:167): closed Consumes named extdeps/formats/sql.dag but no blob on origin/main — froze a dependency outside the closed tree. Split: openapi.dag (present) stays a Consumes edge; sql.dag moved to an explicit "named pending dependency, NOT a frozen closed-tree edge" — scheduled as the T-4.6 formats-bundle extension (canvas #4 fork (a)), binds when it lands (T-16 trigger already gates on T-4.6). T-16 SQL-DDL narrative reworded to match. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-05-19T05:13:43Z

@codex — both BLOCKING findings were already fixed before this scheduled review landed; it ran on sha 7cb0b0f5, current HEAD is 837ada526. Verified against the current tree:

1. 00_compile.dag Owns: none vs frozen orchestrator Result contract — fixed in 1bba4dd65. The "grounded in merged 00_compile.dag/05_emit.dag Owns, NOT a TASKS invention" framing is removed. The I/O block is now a single forward temporal authority (05_emit.dag:138-150): "Result<…,Diagnostic> … will carry once the CP-1b/00_compile.dag train freezes it. It is NOT yet grounded in merged Owns: at HEAD both files are bare scaffolds (Owns: none.), and the landed fail-closed carrier … is Outcome<T> (DECISIONS.md item I)". No merged-Owns claim remains — your "remove the merged-Owns framing" option is exactly what landed (not the "land the train in this PR" option, which is correctly out of scope for a Wave-0 scaffold).

2. sql.dag consumed but not in the closed tree — fixed in 837ada526. extdeps/formats/sql.dag is removed from the frozen Consumes list; only extdeps/formats/openapi.dag (present on main) remains a frozen edge. sql.dag is now an explicit "Named pending dependency (NOT a frozen closed-tree Consumes edge — the file is not on main; INVARIANTS P2)" (05_emit.dag:176-184), scheduled as the T-4.6 formats-bundle extension, binding only when it lands — i.e. your "keep SQL as a future trigger" option. The T-16 SQL-DDL narrative line was reworded to match (:84-93, "NOT-YET-LANDED T-4.6 … this layer does NOT freeze a Consumes edge to it").

No substrate-not-present interface-freeze claims remain. No code change in this reply — both fixes are in the pushed commits above.

— sent from proud-swift-532

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 837ada52 · Trigger: schedule
Thinking: 169s wall

BLOCKING (1)

Root Cause

src/v4/compiler/05_emit.dag The Wave-0 contract collapses Shape-B artifact generation into the 05_emit boundary → split the text so 05_emit owns Shape-A target-source emission while T-16 Shape-B artifacts are produced by reusable .dag artifact-emitter programs over the shared Node tree.

⚠️ Both prior comments are resolved, but the interface contract still needs to preserve Track 16's Shape-B mechanism before it becomes worker authority.

briansrls · 2026-05-19T07:12:38Z

+//        model — the U1 Realize phase, the structural inverse of
+//        `ingest`. Per-target specialization (T-11) and the full-stack
+//        omni-emission demo contract (T-16) are owned here: emit is the
+//        single coercion=emission authority (STRUCTURE.md §"coercion =


BLOCKING: Framing T-16 omni-emission as owned by compiler emit blurs THESIS.md's load-bearing Shape A/Shape B split, where OpenAPI/SQL/Markdown are user-program outputs rather than compiler render targets.

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 1eac1dcc · Trigger: schedule
Thinking: 247s wall

⚠️ Comment 3264350974 remains unresolved: 05_emit is still framed as the omni authority instead of limiting compiler emit to Shape-A target-source emission.

briansrls

Review metadata

Provider / model: codex / unknown
Commit: be5b3210 · Trigger: schedule
Thinking: 157s wall

⚠️ Existing unresolved comment 3264350974 still blocks because 05_emit remains framed as the omni authority rather than limiting compiler emit to Shape-A target-source emission.

briansrls

Review metadata

Provider / model: codex / unknown
Commit: ec7a9e2e · Trigger: schedule
Thinking: 138s wall

BLOCKING (1)

Root Cause

src/v4/compiler/05_emit.dag Locked omni-demo canvas framing is being copied into the v4 emit contract without explicitly reconciling it against the higher-authority THESIS.md Shape-B mechanism → either limit emit ownership to Shape-A target-source emission or state the reconciliation that preserves Shape-B as user-program artifact generation.

⚠️ Two prior issues are fixed, but the Shape A/Shape B authority conflict remains blocking.

briansrls · 2026-05-19T15:47:37Z

+//                                 IS TS source at emit time; Q5-a:
+//                                 `Component` is `Behavior::Bind`, no new
+//                                 substrate kind)
+//     • OpenAPI wire contract   — Shape-B (`extdeps/formats/openapi.dag`)


BLOCKING: OpenAPI is still listed as an emit-owned projection rather than a Shape-B user-program output, so the text still conflicts with THESIS.md's load-bearing Shape A/Shape B split.

… needs-work cost + R8 NAMING (ruled-shape author) Authors the ruled-shape per operator rulings R7+R1+R8 (relayed via vivid-carp-207 → neat-hawk-87 msg_c5dfd3c8). PIVOT-3 head-move authorized for identifier + structural-separation fixes; no merge. Hard hold lifts only for these ruled shapes; comment/header text remains folds-pending-#3358-A2. Pre-author shape-re-confirm vs accumulated scope (msg_dbbbbcc3): - AXIS-1 ruled: simplest Map<Node, InferredFacts> per-node coordinate - AXIS-2 NOT ruled: do NOT author carrier-ownership/import-direction (Source / TargetModel / TargetSource module-ownership stays as-is, surfaced cleanly for hub disposition) - AXIS-3 ruled: drop hollow CostRef alias; mark cost coordinate with R1 🟡 needs-more-work indicator + co-located testcase (anchor_stub_cost_symbol) - NAMING items 1-3 ruled (R8 immediate-authorized): rename anchor_*_wave1_rejects - Folds-pending-#3358-A2: header / comment-text TASKS / Wave / IR / CP prose untouched per R8 split - S1/S2/S3 clause-iii SEPARATION (numbered compiler stages) unchanged: clean Changes: 04_infer.dag (AXIS-1 + AXIS-3): type InferredTree { node: Node, facts: Map<Node, InferredFacts> } // per-node coordinate type InferredFacts { resolved_type: Node inhabits: AlgebraRef // 🟡 needs-more-work — SymbolicCost owner = lens/cost.dag T-12 cost: Symbol // not-yet-grounded; co-located testcase descent: DescentEvidence } (CostRef alias dropped; was hollow Symbol alias per R7.) import v4.std.collection { Map } added; consumes line updated. infer_emit_compile_anchor.dag (AXIS-1 + AXIS-3 + R8 NAMING): - anchor_stub_cost_symbol: Symbol with 🟡 needs-more-work marker - anchor_stub_empty_facts_map: Map<Node, InferredFacts> with Violates-based lookup (Wave-1 fail-closed; the per-node coordinate is honest about not-yet-realized) - anchor_stub_inferred_tree: InferredTree { node, facts: anchor_stub_empty_facts_map } - renamed anchor_{infer,emit,compile}_wave1_rejects → anchor_{infer,emit,compile}_rejects Acceptance: cargo run --quiet --bin v2-compiler -- compile --source-root src/v4 --output-dir /tmp/v4-out → indexed 75 modules / resolved 75 sources / compiled: 79 files emitted, 0 diagnostics PIVOT-3 honored: no merge. AXIS-2 surfaced separately to neat-hawk-87. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Operator no-program-markers override: remove task-id token from data identifiers (stub_stage{1,2}_emitted_rust_digest_placeholder). Header roadmap prose unchanged (rides #3358-A2 comment-layer ruling). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls

Review metadata

Provider / model: codex / unknown
Commit: d4cf8471 · Trigger: schedule
Thinking: 121s wall

⚠️ The Result/sql issues are fixed, but the previously flagged Shape A/Shape B boundary conflict remains unresolved.

…rtions, void rename Part of the A5 bucket-C coherent rework on #3381. DECISIONS.md items 2 and 10 still asserted the receipt-only non-native identity, the empty-prelude compile-receipt fixture, and the second compile anchor — all deleted under A5-A — and named the pre-rename wave1 identifiers. Updated both items to the void-renamed surface and the single native-prelude anchor, and corrected item 2's "bucket C close" to "bucket C — open under P4" consistent with the dissolution-inventory honest pass. Per hub disposition (DECISIONS item-2/10 stale content is in-scope under the A5-B honest-ledger mandate). Scope: items 2/10 only per directive. Item 1 (:189) retains a renamed-not-deleted identifier reference — flagged upward, not silently folded. The two // Owns: header comment lines remain #3358-A2-subsumed and deliberately untouched. v2 compile --source-root src/v4 --target rust: 78 files, 0 diagnostics. — sent from keen-raven-269 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…G-1 immediate) Per operator-direct ★ GROUNDEDNESS-CRITERION-ATTACHED standing (relayed via vivid-carp-207 → neat-hawk-87 msg_139f8799): every type/carrier carries the RULING-1 dead-simple two-state grounded indicator (🟢 pretty-much-done / 🟡 needs-more-work) + co-located testcases-nearby; IMMEDIATELY program-wide; single-line mark on the type, NOT prose (comment-text rubric still rides #3358-A2-(i)). Shape-re-confirm vs accumulated scope (msg_dbbbbcc3): groundedness directive ADDS to the accumulated scope; no conflict with prior rulings; eligible subset = carriers NOT on a surface with an open operator thread. Eligible carriers in #3360 diff scope, marked 🟢 grounded: - 04_infer.dag :: AlgebraRef = Symbol (K-1 name-ref Symbol; testcase: anchor_stub_algebra_ref) - 00_compile.dag :: Source = String (kernel-ambient alias; testcase: anchor_stub_source) - 00_compile.dag :: TargetSource = String (kernel-ambient alias; testcase: emit returns Outcome<TargetSource> in anchor_emit_rejects) Skipped (surface blocked by an open operator-owned BLOCKING thread): - InferredTree (T1/T2/T5 — Axis-1 root, reinforcement, totality accretion) - InferredFacts (T4 — Axis-3 cost-field hollow-alias 🟡 already attached) - TargetModel (T3 — Axis-2 carrier ownership / import inversion) Co-located testcase evidence already present in the witness anchor (test/claim/manual/infer_emit_compile_anchor.dag) per RULING-1 "evidence for state 2". v2 compile: 75 modules, 79 emitted, 0 diagnostics. PIVOT-3 honored: no merge. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Operator no-program-markers override: remove task-id token from data identifiers (stub_stage{1,2}_emitted_rust_digest_placeholder). Header roadmap prose unchanged (rides #3358-A2 comment-layer ruling). Co-authored-by: Cursor <cursoragent@cursor.com>

…olders (TestClaim/Equals witness deferred: B1/T-22); Wave-6 not this lane (#3356) * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * fix(v4): align main.dag TestClassification imports with manual-claim pattern Import Tier2 and Integration constructors from verification.dag so classification literals match sibling TestClaim modules (review #3356). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): remove vacuous T-15 TestClaim from main.dag Codex review: reflexive Equals on one stub reported success without modeling stage1/stage2 digest operands. Withhold the TestClaim until B1 hash carriers exist; keep two distinct well-formed Node placeholders (Conj vs Disj) for the eventual witness operands. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): keep main.dag as single trampoline authority (T-15) Codex/briansrls: digest-placeholder cleanup must not relocate the 1-line include! contract out of bin/main.dag. Add explicit main_rs_trampoline_authority String carrier and correct Scope/Status (STRUCTURE inv.2 + TASKS T-15); emit materializes, this module owns. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): drop redundant std.text import for kernel-ambient String STRUCTURE.md kernel-ambient rule: String literals do not require text.dag in Consumes or a std.text import (codex review #3356). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(ci): give affected job time for full-repo checkout The affected job uses fetch-depth: 0 plus origin/main fetch; on large repos the checkout step can exceed the previous 5-minute job cap and get cancelled mid-fetch (false CI red on PR #3356). Raise timeout to 15m. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): domain-only digest placeholder names in bin/main.dag Operator no-program-markers override: remove task-id token from data identifiers (stub_stage{1,2}_emitted_rust_digest_placeholder). Header roadmap prose unchanged (rides #3358-A2 comment-layer ruling). Co-authored-by: Cursor <cursoragent@cursor.com> * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * fix(test): make v4 bin main smoke hermetic (parse) + doc HOLD tokenize/parse main.dag instead of compile_to_dag (single-file path cannot link v4.std.node; bundling with node.dag collides with bootstrap). Update INVARIANTS + SG-0 census comments. Grounded-indicator lines already removed from main.dag pending canonical spec. Co-authored-by: Cursor <cursoragent@cursor.com> * chore(ci): SG-0 PR-body append for #3356 census +1 Adds `scripts/ci-merge/sg0-pr-body-append.3356.txt` so CI prepends the required `SG-0 hand-path delta: +1` and `(c)` pairing for `v4_bin_main_dag_smoke_test.rs`. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com>

) * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * fix(v4): align main.dag TestClassification imports with manual-claim pattern Import Tier2 and Integration constructors from verification.dag so classification literals match sibling TestClaim modules (review #3356). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): remove vacuous T-15 TestClaim from main.dag Codex review: reflexive Equals on one stub reported success without modeling stage1/stage2 digest operands. Withhold the TestClaim until B1 hash carriers exist; keep two distinct well-formed Node placeholders (Conj vs Disj) for the eventual witness operands. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): keep main.dag as single trampoline authority (T-15) Codex/briansrls: digest-placeholder cleanup must not relocate the 1-line include! contract out of bin/main.dag. Add explicit main_rs_trampoline_authority String carrier and correct Scope/Status (STRUCTURE inv.2 + TASKS T-15); emit materializes, this module owns. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): drop redundant std.text import for kernel-ambient String STRUCTURE.md kernel-ambient rule: String literals do not require text.dag in Consumes or a std.text import (codex review #3356). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(ci): give affected job time for full-repo checkout The affected job uses fetch-depth: 0 plus origin/main fetch; on large repos the checkout step can exceed the previous 5-minute job cap and get cancelled mid-fetch (false CI red on PR #3356). Raise timeout to 15m. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(v4): domain-only digest placeholder names in bin/main.dag Operator no-program-markers override: remove task-id token from data identifiers (stub_stage{1,2}_emitted_rust_digest_placeholder). Header roadmap prose unchanged (rides #3358-A2 comment-layer ruling). Co-authored-by: Cursor <cursoragent@cursor.com> * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * WIP: {T-15} bin/main.dag scaffold + self-host fixed-point TestClaim witness — * fix(test): make v4 bin main smoke hermetic (parse) + doc HOLD tokenize/parse main.dag instead of compile_to_dag (single-file path cannot link v4.std.node; bundling with node.dag collides with bootstrap). Update INVARIANTS + SG-0 census comments. Grounded-indicator lines already removed from main.dag pending canonical spec. Co-authored-by: Cursor <cursoragent@cursor.com> * chore(ci): SG-0 PR-body append for #3356 census +1 Adds `scripts/ci-merge/sg0-pr-body-append.3356.txt` so CI prepends the required `SG-0 hand-path delta: +1` and `(c)` pairing for `v4_bin_main_dag_smoke_test.rs`. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com>

…tness anchor (#3360) * WIP: {T-9+T-10} infer/emit/compile interface-freeze + scaffold + witness-Acce * {T-9+T-10} STRUCTURE.md — enumerate infer_emit_compile_anchor + spelling alignment Reflect the T-9/T-10 interface-freeze scaffold (commit 576d957) in the canonical tree enumeration: bump the .dag checksum 73→74 for the new test/claim/manual/infer_emit_compile_anchor.dag witness; update the 00_compile.dag / 05_emit.dag scope lines to spell `TargetModel` / `Outcome<TargetSource>` (TASKS T-10 Owns / B2-OMNI, DECISIONS I); spell 04_infer.dag's scope line as the bounded Find phase (DECISIONS C1 / IR-1). CP-1b/T-8-close — TargetModel single-spelling coordination with gentle-heron-61; no unilateral cement on TargetSpec. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9+T-10} address cursor/composer-2 review: Practice 9 prose relocation + Consumes truth Strip the multi-paragraph rationale blocks (IR-1, TargetModel/TargetSpec coordination, no-templating, Result-vs-Outcome, acceptance criterion) from the four .dag headers — per Practice 9 (modeling-discipline.md §9) the record relocates to DECISIONS.md / commit message; the .dag keeps only the terse four-line header + anchor (matches sibling resolve_compile_anchor.dag posture). Trim 00_compile.dag `Consumes:` to match real imports: drop compiler/04_infer.dag and compiler/05_emit.dag — the orchestrator's Wave-1 stub only imports std/diagnostic, std/node, std/text. The `emit ∘ core ∘ ingest` composition that would bring those edges in is deferred (Wave-1 fail-closed branch); `Consumes:` is dependency truth, not logical-future. Context (relocated from headers per Practice 9): - IR-1 — InferredTree = Node + frozen 4-coordinate InferredFacts (resolved_type, inhabits, cost, descent); 04_infer is a boundary index. - TargetModel single-spelling confirmed by gentle-heron-61 (CP-1b/T-8-close). - No-templating — Wave-1 fail-closed branch upholds it structurally. - Result-vs-Outcome — `Outcome<TargetSource>` per DECISIONS item I. - Acceptance — `v2-compiler compile --source-root src/v4` reports 74 modules indexed, 78 files emitted, 0 diagnostics; `v2-compiler run` deferred to T-22. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9+T-10} drop unused ResolvedTree import in infer_emit_compile_anchor.dag Addresses cursor/composer-2 #14563 exploratory observation — the anchor exercises infer/emit/compile directly, never references ResolvedTree. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9+T-10} drop unused Correction imports in 04_infer / 05_emit / 00_compile Addresses claude/claude-opus-4-7 #14603 exploratory observation — `Correction` is imported but never referenced; only the `Unavailable {…}` variant literal is used. Matches the spirit of d10ad85 (unused ResolvedTree import). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9} carry IR-1 InferredFacts coordinate at the infer boundary (operator BLOCKING) Addresses operator BLOCKING #3360 @ 04_infer.dag:24 — `type InferredTree = Node` silently dropped the IR-1 four-field facts coordinate (resolved_type, inhabits, cost, descent), violating P2 / facts-flow-forward at the T-9/T-10 interface. Declare the coordinate flat per DECISIONS IR-1: type AlgebraRef = Symbol // K-1 name-ref to algebra inhabitance (Theme-A #2) type CostRef = Symbol // K-1 name-ref to a `lens/cost.dag` SymbolicCost // (cost.dag declaration deferred per T-12; the // Ref-by-Symbol pattern matches AlgebraRef) type InferredFacts { resolved_type: Node // the Node-typed type of this Node inhabits: AlgebraRef cost: CostRef descent: DescentEvidence // from std/cardinality.dag } type InferredTree { node: Node facts: InferredFacts } Cardinality and effects remain NOT fields per IR-1 — both carried by `resolved_type`'s Node structure (the Cardinality connective / the signature). The 4-field set is closed; a 5th field is a STOP. `emit` / `compile` signatures unchanged (`InferredTree` is now a struct, but the parameter shape stays the same). Wave-1 anchor unchanged — fail-closed `Rejected` branch doesn't construct an `InferredTree`. v2 compile: 74 modules, 78 emitted, 0 diagnostics. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-10} fix infer_emit_compile_anchor — emit takes InferredTree, not Node Addresses cursor/composer-2 REQUEST_CHANGES #14641 — after 5506f69 declared `type InferredTree { node: Node, facts: InferredFacts }`, the `anchor_emit_wave1_rejects` call site still passed a bare `Node`, breaking the interface-freeze witness it was meant to exercise. Construct a minimal well-typed stub: data anchor_stub_algebra_ref: AlgebraRef = anchor_stub_algebra_ref data anchor_stub_cost_ref: CostRef = anchor_stub_cost_ref data anchor_stub_inferred_facts: InferredFacts = InferredFacts { resolved_type: anchor_stub_empty_conj, inhabits: anchor_stub_algebra_ref, cost: anchor_stub_cost_ref, descent: DescentUnknown } data anchor_stub_inferred_tree: InferredTree = InferredTree { node: anchor_stub_empty_conj, facts: anchor_stub_inferred_facts } `emit` now receives `anchor_stub_inferred_tree`. The fail-closed `Rejected` branch is unchanged; the witness now actually pins the frozen `(InferredTree, TargetModel) -> Outcome<TargetSource>` signature. v2 compile: 74 modules, 78 emitted, 0 diagnostics. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9+T-10} #3360 A3 rework: AXIS-1 per-node facts coordinate + AXIS-3 needs-work cost + R8 NAMING (ruled-shape author) Authors the ruled-shape per operator rulings R7+R1+R8 (relayed via vivid-carp-207 → neat-hawk-87 msg_c5dfd3c8). PIVOT-3 head-move authorized for identifier + structural-separation fixes; no merge. Hard hold lifts only for these ruled shapes; comment/header text remains folds-pending-#3358-A2. Pre-author shape-re-confirm vs accumulated scope (msg_dbbbbcc3): - AXIS-1 ruled: simplest Map<Node, InferredFacts> per-node coordinate - AXIS-2 NOT ruled: do NOT author carrier-ownership/import-direction (Source / TargetModel / TargetSource module-ownership stays as-is, surfaced cleanly for hub disposition) - AXIS-3 ruled: drop hollow CostRef alias; mark cost coordinate with R1 🟡 needs-more-work indicator + co-located testcase (anchor_stub_cost_symbol) - NAMING items 1-3 ruled (R8 immediate-authorized): rename anchor_*_wave1_rejects - Folds-pending-#3358-A2: header / comment-text TASKS / Wave / IR / CP prose untouched per R8 split - S1/S2/S3 clause-iii SEPARATION (numbered compiler stages) unchanged: clean Changes: 04_infer.dag (AXIS-1 + AXIS-3): type InferredTree { node: Node, facts: Map<Node, InferredFacts> } // per-node coordinate type InferredFacts { resolved_type: Node inhabits: AlgebraRef // 🟡 needs-more-work — SymbolicCost owner = lens/cost.dag T-12 cost: Symbol // not-yet-grounded; co-located testcase descent: DescentEvidence } (CostRef alias dropped; was hollow Symbol alias per R7.) import v4.std.collection { Map } added; consumes line updated. infer_emit_compile_anchor.dag (AXIS-1 + AXIS-3 + R8 NAMING): - anchor_stub_cost_symbol: Symbol with 🟡 needs-more-work marker - anchor_stub_empty_facts_map: Map<Node, InferredFacts> with Violates-based lookup (Wave-1 fail-closed; the per-node coordinate is honest about not-yet-realized) - anchor_stub_inferred_tree: InferredTree { node, facts: anchor_stub_empty_facts_map } - renamed anchor_{infer,emit,compile}_wave1_rejects → anchor_{infer,emit,compile}_rejects Acceptance: cargo run --quiet --bin v2-compiler -- compile --source-root src/v4 --output-dir /tmp/v4-out → indexed 75 modules / resolved 75 sources / compiled: 79 files emitted, 0 diagnostics PIVOT-3 honored: no merge. AXIS-2 surfaced separately to neat-hawk-87. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-9+T-10} attach 🟢 grounded indicator to non-blocked carriers (RULING-1 immediate) Per operator-direct ★ GROUNDEDNESS-CRITERION-ATTACHED standing (relayed via vivid-carp-207 → neat-hawk-87 msg_139f8799): every type/carrier carries the RULING-1 dead-simple two-state grounded indicator (🟢 pretty-much-done / 🟡 needs-more-work) + co-located testcases-nearby; IMMEDIATELY program-wide; single-line mark on the type, NOT prose (comment-text rubric still rides #3358-A2-(i)). Shape-re-confirm vs accumulated scope (msg_dbbbbcc3): groundedness directive ADDS to the accumulated scope; no conflict with prior rulings; eligible subset = carriers NOT on a surface with an open operator thread. Eligible carriers in #3360 diff scope, marked 🟢 grounded: - 04_infer.dag :: AlgebraRef = Symbol (K-1 name-ref Symbol; testcase: anchor_stub_algebra_ref) - 00_compile.dag :: Source = String (kernel-ambient alias; testcase: anchor_stub_source) - 00_compile.dag :: TargetSource = String (kernel-ambient alias; testcase: emit returns Outcome<TargetSource> in anchor_emit_rejects) Skipped (surface blocked by an open operator-owned BLOCKING thread): - InferredTree (T1/T2/T5 — Axis-1 root, reinforcement, totality accretion) - InferredFacts (T4 — Axis-3 cost-field hollow-alias 🟡 already attached) - TargetModel (T3 — Axis-2 carrier ownership / import inversion) Co-located testcase evidence already present in the witness anchor (test/claim/manual/infer_emit_compile_anchor.dag) per RULING-1 "evidence for state 2". v2 compile: 75 modules, 79 emitted, 0 diagnostics. PIVOT-3 honored: no merge. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * {T-10} anchor: replace empty UnknownCost.reason with descriptive String + drop unused symbol Addresses openai-pro/gpt-5.5-pro #15008 finding-2 — UnknownCost { reason: "" } was a C-9-style fabricated empty value at infer_emit_compile_anchor.dag:59. Drop the unused anchor_stub_cost_reason_symbol (declared in the merge resolution but never referenced — oversight); replace the empty reason with a descriptive String literal matching the on-main precedent (test/claim/workflow/ pipeline_rejections.dag:181 uses 'workflow pipeline inferred cost scaffold'). Finding-1 (AXIS-2 emit→compile import inversion) is unchanged — remains operator-tier still-open under the hold; not lane-decidable. v2 compile: 76 modules / 80 emitted / 0 diagnostics. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls and others added 2 commits May 19, 2026 03:01