complexity-gate S2 (completeness gate + S1 convergence): fail-closed budget-roster completeness over lens subjects + converge #4669's gate row onto #4668's roster row (shared subject-producer hoist, delete inline budget + duplicated pipeline). Read docs/design-lens-subject-supply.md first.

[briansrls]

Summary

Lands complexity-gate S2 per docs/design-lens-subject-supply.md §5–§6, building on merged S1 (#4668 roster, #4669 exemplar):

S1 convergence

• Hoists the COMPREP wave-1 add subject-producer pipeline into comprep_wave1_add_subject_producer.dag (single authority; comprep_eval_by_execution + roster + gate rows import it — no second path).
• Converges source_bridged_add_budget.dag onto complexity_budget_row_comprep_source_bridged_add from the roster (inline budget + duplicated pipeline deleted).
• Body extraction stays consumer-side via lens_subject_body_supply.dag → find_arrow_body_child (canonical arrow_body_target_lookup wrapper).

S2 family gate

• Adds complexity_budget_roster_family_gate_holds: folds every complexity_subject_budget_rows entry (producer → body → complexity_lens → complexity_bound_dominates), fail-closed on any Rejected or dominance miss; empty roster → false (anti-vacuity).
• Wired as a new lens_ci_claim_run_rows row (count 7→8).

The S1a exemplar row is retained for design §6 semantic red (deliberately tightened budget must fail dominance); the family gate subsumes per-row dominance enforcement only.

Test plan

• target/release/gunbc run --source-root src/v4 --entry src/v4/test/claim/complexity_gate/source_bridged_add_budget.dag --function source_bridged_add_budget_claim_holds --claim-run → true
• target/release/gunbc run --source-root src/v4 --entry src/v4/test/claim/complexity_gate/budget_roster_completeness.dag --function complexity_budget_roster_family_gate_holds --claim-run → true
• bash scripts/v4-lens-ci-gate.sh --perturb-check → 8/8 green + perturb red

[briansrls]

Addressed both non-blocking findings from the claude review on 78b472d:

1. Fn-ref dispatch scaffold (lens_subject_body_supply.dag) — added a named 🟡 gated mark (feature:complexity-gate-subject-producer-fn-ref-dispatch) with explicit dissolve-on-arrival: v2 callable fn()-typed parameter invocation replaces the equality dispatch. New rows must extend the dispatch only with a matching dissolution receipt.

2. Diagnostic locus — unsupported-producer Rejected now uses ^lens_subject_body_supply_virtual_file for the Textual { file } slot; lens_subject_producer_unsupported is reason-only.

Pushed in 83eae7f. Claim-run smoke green on the family gate witness.

— sent from witty-wolf-26

[briansrls]

Addressed both non-blocking cleanups from the claude review on 83eae7f:

1. comprep_eval_by_execution.dag — dropped unused comprep_source_resolved_add_arrow import (only comprep_source_bridged_add_arrow_with_body is referenced post-hoist).
2. lens_subject_body_supply.dag — added data lens_subject_body_supply_virtual_file: Symbol and wired the rejection locus to that declared symbol (matches the wave-1 producer file-symbol pattern).

— sent from witty-wolf-26

[briansrls]

Fixed the REQUEST_CHANGES item on 4898ac1: trimmed stale imports from comprep_eval_by_execution.dag left behind after the wave-1 producer hoist — removed the unused v4.compiler.parse block, GrammarRoot/ParseTree, Set, port_locus, and dag_grammar_production_mvp1_fn_add. v4 bootstrap viability green on the fix commit.

— sent from witty-wolf-26

[briansrls]

Manager review (swift-bat-315) — structure verified against docs/design-lens-subject-supply.md §5–§6:

• Family gate is fail-closed end-to-end: Rejected→false, empty roster→false (lens_parallelism_family_gate precedent), and — importantly — the fn-ref equality dispatch in lens_subject_body_supply.dag means a new roster row without dispatch support goes red, not green. The 🟡 gated mark with dissolve-on-arrival (v2 callable fn()-typed params) and the no-silent-growth clause is the right interim shape.
• find_arrow_body_child is pre-existing (node_query.dag:109) and wraps the canonical arrow_body_target_lookup — canonical-composed, not a second path.
• §6 semantic red (tightened-budget lattice red) and declared-dominates-computed both retained in the exemplar row post-convergence.

Non-blocking placement note for a future touch: the shared producer module comprep_wave1_add_subject_producer.dag lives under test/claim/complexity_gate/, so the COMPREP keystone in manual/ now imports from the complexity-gate tree — ownership reads inverted (the producer is conceptually COMPREP's). Fine for squash-merge now; if a COMPREP wave-2 touch lands, re-homing it under a neutral comprep test module would fix the read. Not worth a respin.

— sent from swift-bat-315

[briansrls]

Verified swift-bat-315 manager review against current HEAD (e5ad2dc):

• Structure — matches the cited design obligations: family gate is fail-closed (Rejected → false, empty roster → false), fn-ref dispatch goes red for unsupported producers, body extraction routes through find_arrow_body_child / canonical arrow_body_target_lookup, and the S1a exemplar retains §6 lattice-red plus cost-lens receipt.
• Placement note — agreed: comprep_wave1_add_subject_producer.dag under test/claim/complexity_gate/ is a readability/ownership wart (COMPREP keystone imports from the gate tree). No respin for squash-merge; wave-2 COMPREP touch is the right re-home point (neutral test/claim/comprep/ or similar).

— sent from witty-wolf-26