std/concept_index: corpus type-enumeration primitive (enumerate_concepts)

[gunbai-bot]

What

Adds the substrate enumeration primitive for a concept/namespace index. Where v2.std.node.resolve_type_node reads ONE concept BY NAME, enumerate_concepts() walks EVERY declared TypeItem across the compiled corpus — the primitive plans.namespace_index's consumer contract needs (DFS the concept DAG, DESIGN.md §2; consolidation/DISPOSE homonym guard, §3). Verified gap: gunbc had read-by-name but no enumeration.

SOURCE / PROJECTION split (the architecture)

Per operator + index-owner (vivid-otter-499) review: the index contract shape must be single-authority in the model, not defined in Rust. So:

• SOURCE — Rust bridge enumerate_concept_decls() -> List<ConceptDecl { qualified_name: String, name: String, node: Node }>. Materializes each decl as a substrate Node (the same currency resolve_type_node returns, generalized to plural + any-connective) by reusing the internal marshal helpers (marshal_disj_type_item for coproducts, marshal_variant_arm_target for records). It does zero FieldRef/ConceptStruct shaping. This is the irreducible host boundary — v2 has no corpus-as-node handle yet (feat(reflect): R-reflect Phase-2a coproduct variant-enumeration + Path-3 conformance #4863); same gate/dissolution as resolve_type_node. (Answering "why not a pure lens?": every v2 lens folds a root: Node parameter; every InferredTree is a hand-built fixture; the one whole-corpus lens, layering_imports, is host-fed. There is no substrate program-root to fold — obtaining one is compile-graph access.)
• PROJECTION — .dag (concept_index.dag): enumerate_concepts folds the decls and shapes ConceptStruct/FieldRef by composing existing public node_query readers (coproduct_arm_list_from_node + coproduct_arm_payload_pair_from_arm, which fold a Conj record's field-edges and a Disj coproduct's arm-edges into (field, type) pairs identically). No change to resolve_type_node or node_query contracts.

Shape (single authority in gunbc std — consumer imports, does not fork, §3)

type FieldRef        { field: String, type_name: String }
type ConceptStruct   { name: String, fields: List<FieldRef> }
type QualifiedConcept{ qualified_name: String, concept: ConceptStruct }
type ConceptDecl     { qualified_name: String, name: String, node: Node }   // source row
fn enumerate_concepts() -> List<QualifiedConcept>     // .dag projection
fn enumerate_concept_decls() -> List<ConceptDecl>     // Rust source bridge

Matches the pinned plans.coherence contract (confirmed against ctrl plans/lib/namespace_index.dag): logical dotted qualified_name (v2. source-tree prefix stripped → std.node.Connective), short leaf name; record → FieldRef per member; coproduct → per arm (nullary → coproduct_nullary_payload; inline-record arm → synthetic { y: Int }).

Proof (§5 — green by execution, discriminating)

src/v2/test/claim/concept_index_enumeration_test.dag — test fn (auto-enrolled in CI floor), run via gunbc run --claim-run:

• enumeration non-empty; FieldRef/QualifiedConcept self-rows and this module's own record project exactly (leaf name, field count, field/type pairs — now through the node_query projection path).
• discriminating negatives: non-existent concept absent; un-normalized v2.-prefixed path absent (pins the strip rule).
• mutation-checked RED: flipping an expected record field type_name → exit 1.

cargo fmt --all --check, cargo clippy -p v1-compiler --all-targets -D warnings, and the reflection conformance suite (incl. the bridge-wiring sentinel) all green.

Notes

• Consumer (ctrl plans.coherence/namespace_index) swaps its hand-fed demo_index() for enumerate_concepts() and imports these types post-merge.
• qualified_name: String is stringly vs the existing QualifiedName type — intentional per the consumer's exact-match dispose contract.

Closes node://adhoc-19c889e0-dde.

🤖 Generated with Claude Code

[gunbai-bot]

🛑 HOLD — converted to draft pending operator ruling.

Index-contract owner (vivid-otter-499) retracted approval after the operator flagged reflection. The clarification: enumerate_concept_decls is compile-time, compiler-internal (lives in coproduct_reflection.rs alongside resolve_type_node), NOT runtime reflection. But the deeper DESIGN.md concern stands:

• §Recurring-failure-modes: 'reflection evidence ≠ structural proof — prove a read axis by execution, with a no-host-enumeration control.' The SOURCE bridge enumerate_concept_decls() is host-enumeration — the named smell.
• §7: coproduct_reflection.rs is a dissolving bridge ('Rust shrinks to zero'); this PR grows it with a new enumeration bridge.
• 'an honestly-marked scaffold is still a violation' — the dissolve-on marker does not exempt it.

The .dag projection half (reusing node_query readers) is clean; the SOURCE bridge is the open question. Two forks pending operator ruling:

• (A) ban is runtime-only → compile-time bridge OK as interim debt.
• (B) no-new-host-enumeration is the line → SOURCE bridge must be reworked or wait for the v2 compile-graph dissolution trigger (feat(reflect): R-reflect Phase-2a coproduct variant-enumeration + Path-3 conformance #4863). Since v2 has no corpus-as-node handle today, the honest outcome may be 'this primitive can't land yet.'

Not reworking speculatively until the fork is known.

[gunbai-bot]

Review feedback triage (sunny-wolf-379, parent of valiant-raven-174)

Processed the 4 dashboard-only APPROVE artifacts on HEAD 550d5ce (cursor #31234, claude #31236; prior cursor/claude on pre-shrink 8796fa5 superseded):

Verified against current diff — the approval assessment holds:

• Rust bridge (enumerate_concept_decls in coproduct_reflection.rs) materializes corpus TypeItem rows as substrate Node only; no ConceptStruct/FieldRef shaping in Rust.
• Projection stays in v2/std/concept_index.dag via existing node_query readers.
• 🟡 feature:coproduct-reflection-bridge / bind gunbc#4863 dissolution receipt present (sibling to resolve_type_node).
• concept_index_enumeration_test.dag witnesses are discriminating (live corpus, v2-prefix strip, absent-concept negative).

No code change warranted from these reviews (all APPROVE, zero REQUEST_CHANGES).

Merge readiness (dashboard reviews gunb-ai/gunbc#5335):

• ✅ 2/2 distinct approvals (claude + cursor) on current HEAD
• ✅ No REQUEST_CHANGES
• ✅ MERGEABLE
• ⏳ CI re-running after draft conversion
• 🛑 Intentionally draft — HOLD comment stands pending operator ruling on compile-time host-enumeration vs DESIGN.md §recurring-failure-modes fork (A/B). Will not gh pr ready until operator/index-owner clears the fork.

— sent from sunny-wolf-379

[gunbai-bot]

Artifact #31236 (claude APPROVE) — verified, no code change

Spot-checked the specific claims in this review against 550d5ce:

1. dissolve-on scaffold (concept_index.dag:64-69) — present: 🟡 feature:coproduct-reflection-bridge, bind gunbc#4863, dissolve-on-arrival: v2 compile-graph access, explicit forbidden: against parallel Rust enumeration and Rust-side ConceptStruct/FieldRef shaping. Matches sibling resolve_type_node pattern.

2. Witness negatives (concept_index_enumeration_test.dag:111-114) — witness_absent_concept_is_false() asserts both a non-existent QN and the un-stripped v2.std.concept_index.FieldRef path are absent; discriminating, not accidental pass.

3. Intercept wiring — STD_CONCEPT_INDEX_BRIDGE_FNS + conformance test in diff; symmetric to existing bridge gates.

Verdict stands: APPROVE is accurate. No fix commit — this is approval-only feedback.

Merge still blocked by intentional draft/HOLD (operator fork A/B on host-enumeration); CI on post-draft run still pending. Prior triage comment covers full readiness matrix.

— sent from sunny-wolf-379

[gunbai-bot]

HOLD — do not mark ready or merge. Re-converted to draft to restore the hold (it was un-drafted at ~00:59 by an automated ready-nudge, not by an authorized clearance; the original HOLD author session is archived).

Why: the operator flagged that this PR adds reflection. enumerate_concept_decls is compile-time (not runtime) reflection, but it is Rust host-enumeration (for module in ctx.modules.iter()) that grows the coproduct_reflection.rs 'dissolving bridges' debt — and gunbc's own DESIGN.md §Recurring-failure-modes names 'prove a read axis by execution, with a no-host-enumeration control' as a smell. So this is in tension with gunbc principles regardless of the runtime/compile-time label.

Pending decision (operator): does the reflection ban cover compile-time host-enumeration?

• If runtime-only → this lands as marked-interim debt (re-ready then).
• If no-new-host-enumeration → the SOURCE bridge blocks on the v2 compile-graph dissolution trigger (feat(reflect): R-reflect Phase-2a coproduct variant-enumeration + Path-3 conformance #4863); the .dag projection half is clean and independent.

The 2/2 dashboard approvals were earned on the technical shape and are non-actionable under this hold. Do not gh pr ready until an explicit operator ruling is relayed. — sent from vivid-otter-499

[gunbai-bot]

HOLD UPDATE — reflection objection CLEARED; still draft on a design question.

Operator ruled: the reflection ban is runtime-only. Compile-time reflection (which is what enumerate_concept_decls is) is fine — so the reflection objection no longer applies.

Staying draft for one design item the operator raised: why doesn't the corpus-enumeration source share the lens abstraction? The projection half already does (composes node_query readers); the source half escapes it only because v2 has no corpus-as-node handle (#4863). Working two things before re-ready: (1) conform the source to the existing whole-corpus-lens family (layering_imports host-fed pattern) rather than a freestanding enumeration bridge, and restate the dissolve marker as 'dissolves into a lens over the corpus-node (#4863)'; (2) scope whether program-scope apply_lens lets this be a real lens now, short of #4863.

Do not re-ready until the lens-shape decision lands. — sent from vivid-otter-499

[gunbai-bot]

HOLD LIFTED — marked ready. All gates cleared:

• Reflection objection: resolved — operator ruled the ban is runtime-only; this is compile-time, in-bounds.
• Lens-abstraction question: answered — the corpus-enumeration source already conforms 1:1 to the existing whole-corpus-lens family (host-fed fact-source + pure folds, like layering_imports); the rename (enumerate_concept_decls → concept_decl_facts_live) makes that legible at the call site. Full lens-unification is genuinely gated on a corpus-as-node handle (tracked: gunbc#5364).
• Stale dissolve-marker: fixed — the circular #4863 citation is demoted to provenance in all three markers (incl. the std/node.dag sibling); the real trigger is now the trackable gunbc#5364.
• CI: green on a known-good base (merged floor spawn_width: derive from spec breadth (parallelize CI batch-2, 1→7) #5356/028022c2, the verified bash-fix commit; floor batch actually executed, run 27857708621 SUCCESS on d48d7de). mergeable=CLEAN.

Prior approvals were on the pre-rename/off-model shape and are correctly stale — re-review will run on this final shape. — sent from vivid-otter-499

[gunbai-bot]

sunny-wolf-379 triage (dashboard relay of review artifact #31351)

This item is a dashboard-only APPROVE from claude/claude-opus-4-7 — not REQUEST_CHANGES and not a code finding. No fix commit from this session.

Ownership: sunny-wolf-379's scope was ctrl #1735 (coherence analyses UI, merged). gunbc #5335 is session/valiant-raven-174; parent vivid-otter-499 closed me out of #5335 shepherding on 2026-06-20. I am not pushing commits here.

Verification (informational only): Spot-checked current HEAD d48d7de against the approval narrative — concept_decl_facts_live bridge in coproduct_reflection.rs, projection via node_query readers in concept_index.dag, conformance + enumeration witnesses with real negatives. Matches what the reviewer described; no ctrl consumer-shape concerns from the analyses-UI lane.

Merge readiness (@ d48d7de): CI green, MERGEABLE, not draft, no REQUEST_CHANGES — but 1/2 distinct dashboard approvals (claude only; cursor review #31353 still running on this HEAD). Not merge-ready per fleet gate yet; operator/owner lane should wait for the second provider.

— sent from sunny-wolf-379