feat(v3): T-Substrate-AnthropicSchemaMirror — provider-domain type mirror for Messages signature

[briansrls]

Summary

T-Substrate-AnthropicSchemaMirror — v3 typed mirror of provider-domain schema types reachable from dsl/extdeps/llm/anthropic.dag's operation Messages signature.

Scope discipline:

• Schema mirror only. Type authority for the provider-domain types reachable from the Messages signature; no data rows, no fn declarations.
• Callable declaration precursor follows. fn anthropic_messages and any Operation row land in the next substrate slice that consumes these types — outside this PR.
• Grounding PR-β (feat(grounding): T-Ground services.dag PR-β — anthropic_operations Phase 1 pilot #1252) remains paused. This PR alone does not unblock Grounding; PR-β consumes the callable from the follow-on, which in turn consumes these types.
• No service-parser ingestion. v3's parser does not handle the v2 service { operation { transport rest { … } } } block; pulling anthropic.dag wholesale into BOOTSTRAP_FIXTURE_PATH_KEYS would lower the service block as ValueBody::Unparsed. This file is a v3-shaped typed twin, not a parse-import.
• No test-only bridge. The mirror is real substrate type authority loaded by Dag::new(); the integration test consumes the production accessor surface.

Adds:

• src/v3/std/anthropic_schema.dag — type authority only. Mirrors:
  • AnthropicChatMessage (UserMessage / AssistantMessage discriminator)
  • AnthropicUserContentBlock (UserTextBlock / UserToolResultBlock)
  • AnthropicAssistantContentBlock (AssistantTextBlock / AssistantToolUseBlock)
  • AnthropicStopReason (closed Anthropic API enum, all 7 variants)
  • AnthropicMessages200TextBlock / AnthropicMessages200Usage / AnthropicMessages200Body
• src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs — 9 ratchets pinning v3 mirror against dsl/extdeps/llm/anthropic.dag. Each ratchet (a) walks the mirrored type's connective in the bootstrap and asserts the expected field/variant label set, then (b) extracts the matching type <name> block from the v2 source and asserts every expected label is present in that block — coupling the expected set to the v2 source itself, not just to constants in the test. Plus a textual optionality ratchet on is_error: Bool? and stop_sequence: String?.

Out of scope (deferred)

• AnthropicErrorShape — covers the 4xx/5xx response slot only; not on the typed return reach for fn anthropic_messages -> AnthropicMessages200Body. Lands with the response/wire lockstep lane (Grounding's PR-β scope).
• AnthropicModel, AnthropicModelSpec, AnthropicToolDefinition, AnthropicApiVersion, SystemPrompt, CacheControl, ThinkingMode, ThinkingConfig — outside the Messages operation's direct signature reach.
• fn anthropic_messages and any Operation data row — next substrate precursor.

Why a v3 typed twin (rather than ingesting anthropic.dag directly)

Per the prior #1246 STOP+PING: pulling dsl/extdeps/llm/anthropic.dag wholesale into BOOTSTRAP_FIXTURE_PATH_KEYS would lower the v2 service { … } block as ValueBody::Unparsed (v3's parser does not handle that DSL today). This file is the typed twin authored on the v3 surface — same staged-fixture discipline as MethodTemplateContract (#1195) and services.dag PR-α (#1246).

Dissolution trigger

When v3 grows a service-DSL parse/load surface that ingests dsl/extdeps/llm/*.dag with full structural fidelity, this twin retires — the v2 source becomes the only authority and this file collapses. Tracked dissolution per modeling-discipline (feedback_parallel_representation_debt).

Local verification commands

cargo run -p v3-compiler --features bootstrap-regen-fresh --bin regen_bootstrap
cargo run -p v3-compiler --features bootstrap-regen-fresh --bin regen_bootstrap -- --verify
cargo test -p v3-compiler refresh_handwritten_parse_snapshot_manifest -- --ignored
cargo test -p v3-compiler --test integration anthropic_schema_lockstep
cargo test -p v3-compiler --test integration parse_stage4_prep::handwritten_parse_snapshot_matches_manifest
cargo test -p v3-compiler --test integration sg0_census

Test plan

• Lockstep ratchets pass (9/9 local).
• Bootstrap regen + manifest refresh committed.
• SG-0 census entry added for the new hand-authored test file.
• CI green once flipped to ready.

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 1e08a24e · Trigger: schedule
• Comparison: origin/main @ 86961089 ... review/pr-1261-1e08a24e @ 1e08a24e
• Thinking: 23s wall

---

This is a clean substrate addition: typed mirror dag file + bidirectional lockstep ratchet test + regenerated bootstrap. Dissolution trigger documented (v3 service-DSL parser). Type-authority-only with a test asserting no value bodies leak. Set-equality (not subset) on both sides plus per-variant payload + optionality coverage — the ratchet fails closed in both directions.

Verdict: APPROVE — narrow, well-scoped substrate mirror with a fail-closed lockstep test against the v2 authority and a named, bounded dissolution trigger. No INVARIANTS / modeling-discipline / CODING / TESTING violations observed in the diff.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 1e08a24e · Trigger: manual
• Comparison: main @ 86961089 ... session/sharp-raven-604 @ 1e08a24e
• Conversation: View conversation

---

1. Story of the diff

This PR introduces a narrow v3 .dag type mirror for the Anthropic Messages signature rather than importing the existing v2 dsl/extdeps/llm/anthropic.dag wholesale. The new src/v3/std/anthropic_schema.dag explicitly scopes itself to the operation input/output type reach, omits service/operation rows, and names the mirror as a temporary typed twin until v3 can structurally ingest the extdeps service DSL (src/v3/std/anthropic_schema.dag:6-36). The generated bootstrap files are regenerated to include these declarations, the parse corpus manifest is updated for the new .dag, and a new integration ratchet compares the v3 mirror against the v2 source for mirrored type presence, field/variant labels, payload shape, and optionality (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:5-23, src/v3/compiler/tests/integration.rs:40-41).

The load-bearing mechanism is the lockstep test: it treats the v2 file as the source authority and walks the generated v3 bootstrap declarations. That is the right direction for this staged bridge, but the current ratchet does not compare field type targets, so a core part of the “type mirror” can drift while tests stay green.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Finding — BLOCKING. This diff adds substrate-level provider-domain types under src/v3/std/anthropic_schema.dag, and the file correctly states that this is a temporary twin with a dissolution trigger (src/v3/std/anthropic_schema.dag:33-36). However, the bridge’s drift ratchet only compares field/variant labels and optionality: records reduce v2 to labels at src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:314-316, and variants reduce v3/v2 payloads to (field_label, optional) at src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:271-278 and src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:372-390. That means content: List<AnthropicMessages200TextBlock> could become content: String, or input_tokens: Int could become String, without the lockstep test failing as long as the label and ? bit stay the same. For a substrate type mirror, field type identity is part of the modeled fact, not an implementation detail.

1. INVARIANTS.md + modeling-discipline.md.

Finding — BLOCKING. Boundary Discipline / single-authority metadata is only partially enforced. The PR says the v2 file is the authority source (src/v3/std/anthropic_schema.dag:6-15) and that the new test pins mirrored names/labels/variants against it (src/v3/std/anthropic_schema.dag:28-31), but the actual assertions only establish label-set equality and optionality (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:317-347, src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:375-406). The missing type-shape comparison leaves two independently editable authorities for a field’s carrier type.

1. CODING.md.

Compliant. The new Rust test code stays in free-function/data style: helpers such as conj_field_labels, v2_record_fields, v2_disj_variants, and assert_record_lockstep are pure projections/assertions over explicit inputs rather than hidden state or object methods (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:33-78, src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:84-269, src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:312-419).

1. TESTING.md.

Finding — BLOCKING. The new tests are behavior-targeted and wired into the integration suite (src/v3/compiler/tests/integration.rs:40-41), but the central behavior claim is under-tested: “lockstep” for a type mirror must include field type expressions, not only field names and optionality. The record path collects v2_labels from (label, optional) only (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:315-316), while the variant-payload path collects v2_set / v3_set from labels only (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:372-383) and then checks optionality only (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:384-406). This allows semantic type drift to escape the ratchet.

1. LOCKED DESIGN DECISIONS.

Compliant. The PR does not appear to alter a locked thesis/design decision directly; where it introduces a staged mirror, it names the current v3 service-DSL parse/load limitation and scopes the bridge to the Anthropic Messages signature reach (src/v3/std/anthropic_schema.dag:6-26).

1. TRACKED vs UNTRACKED DEBT.

Compliant. The staged mirror has documentation, bounds, and a named dissolution trigger: it is type-authority only for this slice (src/v3/std/anthropic_schema.dag:13-15), bounded to types reachable from operation Messages input/output (src/v3/std/anthropic_schema.dag:17-26), and retires when v3 can ingest dsl/extdeps/llm/*.dag structurally (src/v3/std/anthropic_schema.dag:33-36). The hand-authored test is also registered in the SG0 census (src/v3/compiler/tests/integration/sg0_census_test.rs:232).

3. Verdict

REQUEST_CHANGES. The modeling direction is good and the debt is well bounded, but the lockstep ratchet does not yet protect the actual type carriers of the new substrate mirror. Add comparison of v2 field type expressions against v3 resolved field type shapes, including nested List<T> and T?, or explicitly narrow the mirror’s contract so it is not claiming type lockstep.

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: ee78939a · Trigger: schedule
• Comparison: origin/main @ 86961089 ... review/pr-1261-ee78939a @ ee78939a
• Thinking: 137s wall

---

Findings

• BLOCKING: src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:429 and src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:497 add type-expression equality assertions that fail in this PR. cargo test -p v3-compiler anthropic_schema --test integration fails 5/8 tests because v3 canonicalizes String as FreeMonoid<Int> and Int as Int64, while the v2 source text expects String / Int. This violates TESTING.md’s requirement that new tests be behavior-driven, reliable interface checks; the ratchet is not currently passing.

Verdict: REQUEST_CHANGES. The schema mirror itself is documented with a bounded dissolution trigger, but the committed lockstep test fails in a fresh checkout, so the PR cannot land as-is.

[briansrls]

Already fixed on current head. The reviewer was running against ee78939a9 (an intermediate commit during the type-expression-equality landing where v3 canonicalization was unfolding String to FreeMonoid<Int> and Int to Int64); commit b41143fe4 (current HEAD) corrects v3_canonical_ty to prefer the declaration's surface name when present:

fn v3_canonical_ty(dag: &Dag, ty: DeclarationId) -> String {
    let decl = dag.declaration(ty);
    if let Some(name) = decl.name.as_deref() {
        return name.to_string();
    }
    // anonymous Instantiation / Cardinality fallback below…
}

A named declaration (String, Bool, AnthropicStopReason, AnthropicMessages200TextBlock, …) canonicalizes as its surface name even though its underlying connective unfolds to Instantiation (e.g. String = FreeMonoid<Int>); only anonymous use-site instantiations (List<X>, Map<K, V>, T? cardinality carriers) get unfolded. v2 addresses these by name, v3 mirror does too.

Verified locally on b41143fe4:

running 8 tests
test anthropic_schema_lockstep_test::anthropic_assistant_content_block_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_chat_message_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_messages_200_body_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_messages_200_text_block_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_messages_200_usage_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_schema_authors_no_data_rows ... ok
test anthropic_schema_lockstep_test::anthropic_stop_reason_lockstep ... ok
test anthropic_schema_lockstep_test::anthropic_user_content_block_lockstep ... ok
test result: ok. 8 passed; 0 failed

Clippy clean. CI is rerunning on b41143fe4.

— sent from sharp-raven-604

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 365e57f6 · Trigger: manual
• Comparison: main @ bb40a384 ... session/sharp-raven-604 @ 365e57f6
• Conversation: View conversation

---

1. Story of the diff

This PR stages a v3 provider-domain type mirror for the Anthropic Messages signature because v3 cannot yet ingest the existing v2 dsl/extdeps/llm/anthropic.dag service block structurally. The new src/v3/std/anthropic_schema.dag is deliberately scoped to the types reachable from the operation Messages input/output signature, explicitly omits out-of-reach provider types, and declares itself “type authority only” until a later operation-row/callable precursor consumes it (src/v3/std/anthropic_schema.dag:6-25, :27-35).

The safety mechanism is the new anthropic_schema_lockstep_test.rs, which treats the v2 file as the source authority and compares the v3 bootstrap’s mirrored declarations for field sets, variant sets, optionality, and canonical type expressions (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:5-23, :413-531). The PR wires that ratchet into the integration suite, adds the new .dag file to the parse corpus manifest, updates the SG-0 census allowlist for the new hand-authored test file, and regenerates bootstrap artifacts.

2. Invariant categories

1. LAYER MODEL — Compliant. This touches v3 std/substrate-adjacent modeling, but not core Dag/Behavior structure; the mirror names its v2 authority, limits scope to the Messages signature, and makes the temporary duplication explicit with a dissolution trigger: “when v3 grows a service-DSL parse/load surface… this twin retires” (src/v3/std/anthropic_schema.dag:6-35).
2. INVARIANTS.md + modeling-discipline.md — Compliant. Boundary discipline / single authority is handled by the lockstep ratchet rather than an unguarded parallel representation: the test asserts set equality between v2 and v3, optionality parity, and type-expression parity (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:10-18, :413-531). Coproduct discipline is also addressed: each new provider-domain coproduct is annotated as terminal with source rationale, e.g. AnthropicChatMessage and AnthropicStopReason (src/v3/std/anthropic_schema.dag:37-45, :66-78).
3. CODING.md — Compliant. The Rust test helper code is data + free functions over explicit inputs (&Dag, type names, declaration IDs), not hidden state or object-style mutation; examples include conj_field_labels(dag, name), v3_field_is_optional(dag, owner, field), and assert_record_lockstep(type_name) (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:33-78, :413-465).
4. TESTING.md — Compliant. The diff adds a focused integration ratchet for exactly the boundary this PR creates: v2 authority text versus v3 bootstrap mirror. It is behavior/contract-driven rather than snapshotting generated output: field/variant set equality, optionality, payload-shape parity, and “no data rows” are each asserted directly (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:413-531, :582-601). The integration registration is present (src/v3/compiler/tests/integration.rs:37-42).
5. LOCKED DESIGN DECISIONS — N/A. I do not see this PR altering a locked design surface; it references existing staged-fixture/service-DSL gaps as justification, but does not change the locked substrate reflection or external realization contracts.
6. TRACKED vs UNTRACKED DEBT — Compliant. The mirror is a tracked bridge: documentation names the authority source, the bounds, and a checkable dissolution trigger (src/v3/std/anthropic_schema.dag:6-35). The new hand-authored Rust test is also accounted for in the SG-0 census allowlist (src/v3/compiler/tests/integration/sg0_census_test.rs:229-234).

3. Verdict

APPROVE. The PR introduces a temporary v3 mirror, but it is bounded, source-anchored, ratcheted against the v2 authority, and carries an explicit dissolution trigger. I did not find a changed line that violates the substrate/modeling, coding, testing, locked-design, or tracked-debt rules.

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: 365e57f6 · Trigger: schedule
• Comparison: origin/main @ bb40a384 ... review/pr-1261-365e57f6 @ 365e57f6
• Thinking: 57s wall

---

Findings: None. I read the rubric files from the pinned *-context/ paths and reviewed git diff origin/main...HEAD with focus on the hand-authored surface (src/v3/std/anthropic_schema.dag, anthropic_schema_lockstep_test.rs, integration.rs, manifest line, sg0_census_test.rs entry). The large bootstrap_generated*.rs churn is mechanical regen from the new std module.

Against that slice:

• INVARIANTS / modeling discipline: The mirror is explicitly type-only, scoped to the Messages signature reach, names v2 authority (dsl/extdeps/llm/anthropic.dag) and a named dissolution trigger (retire when v3 can load v2 service DSL with full fidelity). Sum/record shapes carry 🟢 TERMINAL rationale tied to the v2 source — satisfies the “ledger / terminal vs scaffold” expectation for new coproducts on the modeled side. No second competing authority without the lockstep ratchet.

• CODING.md: Data + free functions in the test module; panics/unwrap live in test code where the doc allows it.

• TESTING.md: Integration-style lockstep is appropriate for the interface (“v3 mirror stays set-equal to v2 for these types”), analogous to other corpus/authority tests; corpus manifest update matches the parse-corpus pattern.

The span.file == "src/v3/std/anthropic_schema.dag" filter in anthropic_schema_authors_no_data_rows matches an existing pattern (e.g. services_carrier_shape_test.rs for services.dag) and is not a new class of concern introduced only here.

Verdict: APPROVE — The diff is focused, fail-closed at the ratchet (set equality, optionality, no data rows), and aligns with documented bridge + sunset discipline. I do not see a concrete rubric violation tied to a changed line.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 365e57f6 · Trigger: schedule
• Thinking: 266s wall

---

Non-blocking — Strengths

• src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs The lockstep test checks field sets, variant sets, payload fields, optionality, and canonical type expressions, so the mirror has a useful fail-closed drift ratchet once the cited v2 authority exists.

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

• src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs The type-only ratchet at line 558 only rejects value_body rows, so a future fn in anthropic_schema.dag would lower as an Arrow with value_body None and bypass the guard; extend it in the T-Substrate-AnthropicSchemaMirror follow-up lane if deferred.

⚠️ Prior blocking comments 3167442700 and 3167442703 remain unresolved; I found no new blocking concerns beyond them.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 6613e080 · Trigger: manual
• Comparison: main @ bb40a384 ... session/sharp-raven-604 @ 6613e080
• Conversation: View conversation

---

1. Story of the diff

This PR stages a v3 provider-domain mirror for the Anthropic Messages signature rather than trying to ingest the v2 service { operation ... } source directly. The new src/v3/std/anthropic_schema.dag explicitly names dsl/extdeps/llm/anthropic.dag as the authority, limits scope to the input/output types reachable from operation Messages, and defers callable/operation rows to a later slice while this PR remains type-authority only (src/v3/std/anthropic_schema.dag:6-26). The actual mirror adds closed provider-domain coproducts for chat/content/stop reasons plus response records for the 200 body (src/v3/std/anthropic_schema.dag:43-116).

The key guardrail is the new integration ratchet: it extracts v2 type blocks from the v2 authority text and compares them against the full v3 bootstrap declarations for field/variant set equality, optionality, and canonical type expressions (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:5-23, :413-530). It also asserts the new .dag file stays type-only by rejecting any lowered data row or fn declaration from that file (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:582-610). The rest is wiring: the test module is registered, the parse corpus manifest includes the new .dag, the SG0 census accounts for the new hand-authored test, and generated bootstraps are refreshed.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — this does touch Dag-visible substrate/domain types under src/v3/std, but it keeps them as declared types only: the file states “type authority only” and defers fn anthropic_messages plus Operation data rows to a later consuming precursor (src/v3/std/anthropic_schema.dag:13-15). The model also names its upstream authority and bounds scope to the signature-reachable types (src/v3/std/anthropic_schema.dag:6-26).

1. INVARIANTS.md + modeling-discipline.md.

Compliant — boundary discipline / single authority is handled as a tracked mirror rather than silent parallel authority: the file names the v2 authority, adds a lockstep ratchet, and names the dissolution trigger when v3 can ingest dsl/extdeps/llm/*.dag structurally (src/v3/std/anthropic_schema.dag:28-36). Illegal states are also pushed into type shape: AnthropicChatMessage separates user vs assistant content blocks structurally rather than permitting a shared free-form content payload (src/v3/std/anthropic_schema.dag:38-45).

1. CODING.md.

Compliant — the new Rust test is data + pure helper functions over explicit inputs: helpers take &Dag / type names and return structured projections (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:33-78, :93-126), with no hidden global state beyond a compile-time include_str! of the named authority (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:29). Panics are confined to test assertions / ratchet failures, not production library paths.

1. TESTING.md.

Compliant — the PR adds behavior-driven lockstep tests at the right level for a bootstrap-visible mirrored schema: set equality, optionality, type-expression equality, and type-only enforcement are explicit test claims (src/v3/compiler/tests/integration/anthropic_schema_lockstep_test.rs:10-17, :418-463, :472-530, :582-610). The integration layer is justified because the contract being pinned is “v3 full bootstrap reflects this .dag mirror in lockstep with the v2 authority,” not a narrow helper function.

1. LOCKED DESIGN DECISIONS.

N/A — the diff references the known service-DSL parse/load gap and future service ingestion path, but does not edit or contradict a locked design document. The divergence is explicitly scoped as a temporary typed twin until service DSL ingestion exists (src/v3/std/anthropic_schema.dag:6-15, :33-36).

1. TRACKED vs UNTRACKED DEBT.

Compliant — the mirror is a temporary parallel representation, but it has all three bridge properties: documentation of why it exists (src/v3/std/anthropic_schema.dag:6-15), bounded scope (src/v3/std/anthropic_schema.dag:17-26), and a named dissolution trigger (src/v3/std/anthropic_schema.dag:33-36). The new hand-authored Rust ratchet is also accounted for in the SG0 census (src/v3/compiler/tests/integration/sg0_census_test.rs:229-233).

3. Verdict

APPROVE

The PR introduces a substrate-visible mirror, but it is narrowly scoped, structurally typed, ratcheted against its named authority, and explicitly marked as dissolving once v3 can ingest the upstream service DSL directly. I did not find a diff-local invariant violation that would justify blocking or comments.

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: 6613e080 · Trigger: schedule
• Comparison: origin/main @ bb40a384 ... review/pr-1261-6613e080 @ 6613e080
• Thinking: 125s wall

---

Verdict: APPROVE

The diff is narrowly scoped: a documented v3 provider-domain mirror, generated bootstrap updates, and a lockstep ratchet against the v2 authority. I found no concrete violations of the pinned invariants or testing/coding guidance. The temporary duplicate authority is documented, bounded, tested bidirectionally, and has a named dissolution trigger.

Verification run: cargo test -p v3-compiler anthropic_schema_lockstep --test integration passed all 8 targeted tests.