v4 T-4 D4: create extdeps/languages/resolver.dag — shared GroundingMap home

[briansrls]

Summary

Encoding-half of operator-ratified D4 (src/v4/DECISIONS.md, PR #3216): creates the new file src/v4/extdeps/languages/resolver.dag, the shared D2-resolver registry.

• type GroundingMap<IRCarrier> { spelling: String } is its first resident — D2's D2a(2) surface-spelling carrier, homed once here instead of re-declared in five language files (parallel-declaration P2 forbids; feedback_import_not_redeclare_carriers).
• STRICT D4 admission rule documented in-header: resolver.dag hosts ONLY structurally-language-INVARIANT resolver types. Spec-varying types (OverflowDisposition, NaN/Inf) stay per-language.
• Zero-import leaf file (String is kernel-ambient).
• STRUCTURE.md: enumeration gains the resolver.dag line; checksum count bumped 64→65 (per invariant Add SVG viz, test helpers, and makegen scaffold #1, enumeration is authoritative).

Sequencing

Prepared in parallel with #3216 per the dispatch brief. Cites D4 as ratified; lands after #3216 merges the D4/D5 rows to main. This PR is the encoding-half of D4 — NOT a downstream consumer.

Out of scope (separate fan-out, canonical rust.dag first): the per-language import … resolver { GroundingMap } lines + per-primitive data <lang>_<prim>_grounding instances. rust.dag's current provisional 🟡 TRACKED-SCAFFOLD GroundingMap declaration (rust.dag:1718) relocates here in that fan-out.

Test plan

• cargo test -p v3-compiler --test integration v4_extdeps_resolver — new 0-diagnostic compile_to_dag smoke passes (mirrors the typescript.dag smoke).
• Confirm v4 D4 + D5 — ratify GroundingMap-home (resolver.dag) + operator-driven frozen-header reconcile #3216 merges D4 to main before this lands.

🤖 Generated with Claude Code

SG-0 hand-path delta: +1 (EXPECTED_HAND_AUTHORED_TEST +1; EXPECTED_HAND_AUTHORED_NON_TEST 0, EXPECTED_HAND_AUTHORED_FRAGMENTS 0)

SG-0 pairing: (b) operator-ratified budget. DECISIONS.md row D4 (operator-ratified 2026-05-16) mandates this exact hand-path add: D4's encoding targets explicitly include the extdeps/languages/* P5(b) receipt (the INVARIANTS.md row + sg0_census_test.rs EXPECTED_HAND_AUTHORED_TEST entry + the compile_to_dag smoke harness — the mandated triple). The +1 is operator-budgeted, not a discretionary census add. Authorization: #3216 (the operator-ratified D4/D5 ratification PR).

[briansrls]

Re the cursor/composer-2 review's prose nit (resolver.dag header "2026-05-16" vs STRUCTURE.md "2026-05-17"):

Not an inconsistency — the two dates intentionally mark two distinct events, and resolver.dag already states both:

• resolver.dag:8 — "DECISIONS.md row D4 (operator-ratified 2026-05-16)" — the date D4 was ratified (matches the D4 row's own operator-ratified 2026-05-16).
• resolver.dag:48 — "file created 2026-05-17" — the date this file was added.
• STRUCTURE.md ledger — "+1 .dag 2026-05-17: resolver.dag — operator-ratified D4" — the file-addition date, with "operator-ratified D4" as the why-clause. Same convention as the dag.dag ledger line directly above it.

A decision ratified one day and encoded the next is expected (ratified ≠ encoded ≠ landed is the standing v4 discipline). No fix needed; leaving as-is.

The exploratory point about the PR-description P5(b) receipt: the two SG-0 lines + the in-repo INVARIANTS row are present in the PR body and the diff respectively; CI's SG-0 gate checks the PR-body receipt mechanically.

— sent from keen-gull-671

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 713d5f76 · Trigger: schedule
• Thinking: 294s wall

---

Non-blocking — Strengths

• src/v4/extdeps/languages/resolver.dag GroundingMap is a product carrier with the language-varying spelling left to per-language data instances, so it avoids a shared coproduct or parallel per-language type authority.

ROADMAP — Verified

• T-PB-B / pb_rust_tests_outside_residual_zero: The added Rust smoke test is paired with the SG-0 census entry and INVARIANTS P5 receipt naming the existing zero-Rust-test lane and dissolution trigger.

✅ Mixed .dag/docs/test change looks consistent with the thesis, modeling discipline, and P5 receipt requirements; no blocking concerns.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: a2f941e3 · Trigger: manual
• Comparison: main @ ba9337f1 ... session/keen-gull-671-resolver-dag @ a2f941e3
• Conversation: View conversation

---

1. Story of the diff

This PR creates a shared v4 language-resolver home for facts that are invariant across target-language specs: src/v4/extdeps/languages/resolver.dag declares module v4.extdeps.languages.resolver and introduces GroundingMap<IRCarrier> { spelling: String } as the common carrier for per-primitive surface spellings (src/v4/extdeps/languages/resolver.dag:68, src/v4/extdeps/languages/resolver.dag:86-88). The file is intentionally narrow: it admits only structurally language-invariant resolver types and explicitly keeps spec-varying facts such as overflow disposition and NaN/Inf handling out of the shared registry (src/v4/extdeps/languages/resolver.dag:15-27). The PR then updates the v4 structure ledger/checksum to include the new file (src/v4/STRUCTURE.md:35, src/v4/STRUCTURE.md:38, src/v4/STRUCTURE.md:114-124) and adds a tracked Rust integration smoke test that compiles the new .dag file and requires zero diagnostics (src/v3/compiler/tests/integration/v4_extdeps_resolver_dag_smoke_test.rs:13-25). Because that smoke test is new hand-authored Rust, the diff also registers it in integration.rs, the SG-0 hand-authored test census, and the INVARIANTS debt ledger (src/v3/compiler/tests/integration.rs:276-277, src/v3/compiler/tests/integration/sg0_census_test.rs:830-833, INVARIANTS.md:345).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation). — Compliant. This is .dag model/substrate-adjacent authoring rather than a Rust implementation mutation: the new authority is a v4 extdeps language module (src/v4/extdeps/languages/resolver.dag:68) with one shared carrier (src/v4/extdeps/languages/resolver.dag:86-88), and it does not touch dag.rs or mutate the compiler’s runtime substrate types.
2. INVARIANTS.md + modeling-discipline.md. — Compliant. P2 single-authority discipline is handled explicitly: the file says these are “one declaration, imported by each language file, never re-declared” (src/v4/extdeps/languages/resolver.dag:3-7), fences the admission rule to invariant shared types (src/v4/extdeps/languages/resolver.dag:15-27), and names the existing provisional Rust-local declaration as something that relocates here in the fan-out (src/v4/extdeps/languages/resolver.dag:51-56). P3 fail-closed is covered by the smoke test’s compile_to_dag result split and zero-diagnostics assertion (src/v3/compiler/tests/integration/v4_extdeps_resolver_dag_smoke_test.rs:15-25).
3. CODING.md. — Compliant. The new Rust test stays simple and explicit: it has no hidden mutable state, passes the source/path directly to compile_to_dag, and matches structured CompileError cases rather than probing strings (src/v3/compiler/tests/integration/v4_extdeps_resolver_dag_smoke_test.rs:7-25). The panic! surface is confined to test code, which is consistent with the style guidance.
4. TESTING.md. — Compliant. The test is correctly integration-level because the behavior under test is whether the new .dag authority can pass the v3 compile/lower/infer surface with zero diagnostics (src/v3/compiler/tests/integration/v4_extdeps_resolver_dag_smoke_test.rs:3-5, src/v3/compiler/tests/integration/v4_extdeps_resolver_dag_smoke_test.rs:13-25). It is also one claim: v4_extdeps_resolver_dag_compiles only proves clean compilation of resolver.dag.
5. LOCKED DESIGN DECISIONS. — N/A. The diff does not alter a locked thesis/design document. It references an operator-ratified D4 placement decision and follows it by anchoring the shared home here (src/v4/extdeps/languages/resolver.dag:8-13, src/v4/extdeps/languages/resolver.dag:58-65), but there is no diff-cited divergence from a locked design.
6. TRACKED vs UNTRACKED DEBT. — Compliant. The new hand-authored Rust smoke harness is tracked: the INVARIANTS row documents the reason, the same-PR SG-0 census line, the dissolution trigger, and the interim ratchet (INVARIANTS.md:345), while the census actually includes the path in this PR (src/v3/compiler/tests/integration/sg0_census_test.rs:830-833). The staged pre-fan-out state is also bounded: resolver.dag says per-language imports and data instances are separate fan-out work, with the current provisional Rust-local GroundingMap relocating here in that fan-out (src/v4/extdeps/languages/resolver.dag:51-56).

2.5. Top-down PM intent review

Compliant. The highest-level intent is preserved: target-language grounding remains .dag data, not a new hand-written compiler path, and the shared resolver home reduces duplicate authority rather than creating a permanent parallel representation. The potentially risky part — spelling: String — is scoped as a source-syntax boundary value for emit/ingest, while the typed carrier remains GroundingMap<IRCarrier> rather than an untyped lookup table (src/v4/extdeps/languages/resolver.dag:34-42, src/v4/extdeps/languages/resolver.dag:74-88). I do not see a diff-cited case where this PR makes a must-have target optional, turns a dissolution goal into permanent scaffolding, or adds hand-written implementation beyond the tracked smoke receipt.

3. Verdict

APPROVE. The PR is narrow and disciplined: it creates the shared v4 resolver authority, updates the v4 structure ledger, and registers the temporary Rust smoke test with the required SG-0/P5 tracking. I found no blocking invariant or PM-intent mismatch in the diff.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: a2f941e3 · Trigger: schedule
• Thinking: 148s wall

---

✅ No new concerns in this re-review.

[briansrls]

Closing per operator-ratified Phase 0 of the D2 foundation reversal.

The operator declared D2 (the alias-identity primitive-inhabitance form) WRONG; a fact-bundle model replaces it. resolver.dag exists almost entirely as the shared D2-resolver registry (GroundingMap<IRCarrier>, the D4 "completes-D2" home) — per the verified D2-impact map this is the (a) close disposition: the file's reason to land is removed by the reversal.

This is NOT a git-revert of any merged commit — it is a clean close of an unmerged PR. The resolver.dag work is not lost: the closed branch is preserved, and the genuinely-durable fact (surface-spelling is a real per-language fact) re-homes in the forthcoming fact-bundle design (the substrate-first Phase-1+ rework). The binding-gate four-check this PR cleared was correct work; the foundation it encoded is what reversed, above the PR.

Operator-ratified; T-4 mgr executing Phase 0 off the verified impact map. — sent from vivid-carp-207