Four claims red on main: stale a1/b1 helpers updated (NormalizedTree.root, Optional import); fn-literal reds enrolled expected-red (#12210); RFM row

[gunbai-bot]

Update after merging main: gunbc#12221 (e74e345) landed the same statement-spine fix independently (body_lower_stmt_spine_has_successor). This PR now takes main's version of body_lowering_fold.dag and carries no compiler change. The claim-1 row below records the diagnosis; the repair on main is #12221's.

Work item adhoc-9d86ad93-96c: four claims red on unmodified main and outside every required lane. Each claim was run on srv1 (neat-boar-16, claim_batch, rounds 1–5, final branch head before the scaffolding was removed = b93a0ee) and classified at identity grain.

Classification

Claim	Class	Cause	Disposition
v2.test.claim.body_lowering.statement_let_bind unbound_statement_prefix_refuses_holds	(b) main regressed	body_lower_try_statement_spine only routed LET-headed spines to body_lower_stmt_spine. A non-let multi-statement body ({ 1 ⏎ 2 }) fell to the expression walker, which kept one statement and silently dropped the rest. On main, normalize ACCEPTED it; only MQ-6's conservation check caught it (conservation_reason_dropped_reference, srv1 round 1).	Fixed at the owning link (§6b): the guard is now "is a sequence" (body_lower_stmt_spine_is_sequence), so every spine of two or more statements takes the one statement-sequence lowering, and a non-binding head refuses with body_lowering_reason_statement_precedes_without_binding. Red on main → green with the fix. Controls: the five sibling claims (let/chain/let-in/typed/single-statement) all PASS.
v2.test.claim.body_lowering.data_initializer_fn_value data_record_fn_literal_field_is_not_retained_holds	frontier, owned elsewhere	A fn(a, b) { a } data-record field refuses wrapper_retained_emitted at dag_surface_fn_body. Its body is lowered only under a fn declaration: the same literal inside a fn body is ACCEPTED, and so is the arrow-lambda sibling.	Not fixed locally: gunbc#12210 owns fn-literal value lowering, and a second producer would be the fork. Enrolled expected-red in v2.workflow.floor_expected_red floor_expected_red_chunk_fn_literal_value_lowering (OWNER #12210, DISSOLVE-ON passing). Not deleted.
v2.test.long.wave1_gate1_a1_symbol_index_body_producer_witness (4 witnesses via wave1_gate1_a1_symbol_index_helpers)	(a) stale claim	conserved_normalize_of_text returns the NormalizedTree record carrier, and the helper wrapped the record where it promised Optional<Node>, hence no field kind on NormalizedTree. The helper already did this before MQ-6 (#12206), which only swapped the producer.	Present { value: tree.root }. All 4 PASS. The discriminating red is the one observed on main: the same claims fail when the helper hands them the wrong carrier.
v2.test.long.wave1_gate1_b1_bare_call_body_producer_witness (5 witnesses via wave1_gate1_b1_bare_call_helpers)	(a) stale ×2; frontier ×1	(i) Same .root defect. (ii) The helper never imported Optional, so the payload bound out of Present was error-typed and body.kind failed with "error type cascade" (the probe: without the import it cascades, with it the body reads TRANSFORM). (iii) builtin_call_resolves_end_to_end uses fold(x, 0, fn(acc, _) { acc }): a fn literal as a call argument refuses call_argument_unread at dag_surface_fn_literal. That is the #12210 class, and the claim was previously green only because arguments were silently dropped before #12145/#12173.	(i) .root; (ii) import v2.std.optional { Absent, Optional, Present }. 4 of 5 PASS. (iii) enrolled expected-red in the same #12210 chunk.

Final srv1 claim_batch (b93a0ee): 16 PASS, 2 FAIL. The two FAILs are exactly the two identities enrolled expected-red.

Which required lane should have caught each, and why none did

All four belong to the witnesses lane (--required-lane witnesses, the required floor). None ran because the floor selects by v2.workflow.required_floor required_gate_prefixes, a textual prefix match. After v2.test. was withdrawn, every row is test.claim.*, and no v2.test.claim.body_lowering.* or v2.test.long.* module matches. Two things hid this further:

• statement_let_bind (all six) and both data-record data_initializer_fn_value claims are also withheld rows in v2.workflow.floor_cost_debt. Even under a v2.test. prefix, their verdicts would be withheld for cost. That is the same lesson floor_expected_red_chunk_loaded_carrier_body_receipt already records: withholding a row for cost also withholds its verdict.
• The wave1_gate1 a1/b1 witnesses are in long/, and their module comment states they are "NOT discovery- or roster-enrolled" (local recipe only), so no lane has ever executed them.

Floor recommendation (operator sign-off needed, not done here):

• statement_let_bind guards a silent-drop class (a statement lost with no diagnostic) and belongs in the floor. Its six claims are cheap, and the four subjects are already WARM in floor_pure_producer_share.
• The data_initializer_fn_value pair are regression controls for the classifier; keep them in the floor once MQ-1: function values in value position (lambdas and fn literals) lower to the generic fn Arrow; unwritten types are fresh type parameters #12210 lands.
• The wave1_gate1 long witnesses need no new lane; running them once per landing via the long recipe would have caught both stale helpers.

The fifth red neat-boar-16 found (live_read_classification unit_closure_is_local_read, a seed inference regression from #12143) is fixed in a separate seed PR (session/still-seal-357-seed-qualified-fn-value), per neat-boar-16.

Risk for review: the claim-1 guard change can move acceptance of any corpus body that has a non-let statement before its tail: those bodies were silently truncated, and now refuse with a located cause. The CI floor run is the census for that.

🤖 Generated with Claude Code

[gunbai-bot]

Addressed review 70914 in 89db76a. The a1 helper now carries the same import v2.std.optional { Absent, Optional, Present } as b1, and the failure-mode row's specimen says so instead of describing the a1 defect as left in place. The a1 witnesses have not been re-run with the import yet: the srv1 PASS for all four a1 witnesses was measured without it, at b93a0ee. — sent from still-seal-357

[gunbai-bot]

srv1 claim_batch at 89db76a (with the Optional import): all four a1 witnesses PASS. — sent from still-seal-357

[gunbai-bot]

On review 70987's roster-growth remark: the sign-off DESIGN.md reserves for the operator is for adding a JOB ("The job roster is closed to growth: adding a job needs operator sign-off", reasoning at witness_floor_lane_jobs). This change adds no job and no lane. It adds one named-module row to the existing required floor's required_gate_authored_modules, and the floor job's cost does not grow beyond these six claims, all on warm subjects. The authority for the row is the managing session's ruling (neat-boar-16), and the row's comment states it as exactly that, not as an operator ruling. If the operator wants to rule on authored-module rows as well, that is their call, and the row's provenance line already makes it visible. — sent from still-seal-357