#12145 follow-up: value-position regression controls, lambda-argument frontier enrolled expected-red, RFM receipts (no compiler change)

[gunbai-bot]

What this is now: controls, a named frontier, and receipts. No compiler change.

This PR started as the fix-forward for the regression #12145 landed. Measurement narrowed it: #12173 (landed) fixed call arguments and match scrutinees, and #12194 owns operator operands. At acca7e8, src/v2/compiler/body_lowering_fold.dag is unchanged from main. What lands:

1. Permanent regression controls for XL-2 hotfix: call arguments lower as values; nested-call/record args no longer refuse the whole module (list args still refuse, located) #12173's value-position reading: v2.test.claim.namespace_xl0.value_position_whole_read.
   • Two production-route SHAPE claims. They walk the normalized tree from module_roots_from_source_root_ingest through v2.std.node_query's call readers and assert callee, arity 1, and the argument at the authored atom. Measured red on the pre-XL-2 hotfix: call arguments lower as values; nested-call/record args no longer refuse the whole module (list args still refuse, located) #12173 fold, and red under a callee mutation.
   • Three resolve-verdict controls.
2. The lambda-argument frontier, enrolled expected-red. A lambda argument still refuses its module as call_argument_unread, which is gen-one's first fatal in weather.dag. Two claims pin the target: a lambda whose body uses its parameter RESOLVES, and an undeclared name in a lambda body refuses at its atom. They are enrolled in v2.workflow.floor_expected_red floor_expected_red_chunk_lambda_argument_lowering, owned by the MQ lane; fierce-gull-556's MQ-1: function values in value position (lambdas and fn literals) lower to the generic fn Arrow; unwritten types are fresh type parameters #12210 targets them. They dissolve when both pass. Whichever of #12145 follow-up: value-position regression controls, lambda-argument frontier enrolled expected-red, RFM receipts (no compiler change) #12198 and 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 second removes the rows.
3. RFM receipts in lowering_accessor_collapses_a_sequence_operand: XL-2: sequence-operand refs reach resolve — operator expressions lower whole, a sequence operand is never narrowed #12145 exposed silent truncation; it was not a regression. The receipt adds the identity-diff lessons, the missing locus, and the dead ends below.

What was tried and dropped, with the measurement

• Scrutinee diagnostics_merge (review 70746). body_lower_try_match_from_captured binds the scrutinee's accepted diagnostics as _. With XL-2 hotfix: call arguments lower as values; nested-call/record args no longer refuse the whole module (list args still refuse, located) #12173 landed, no claim distinguishes merging them from main (srv1 matrix at 4ab6497: identical under both folds), and no scrutinee shape is known to raise an accepted diagnostic. It was an unobservable change, so it was dropped rather than landed without a discriminator. If a scrutinee shape that raises an advisory is found, the merge returns with a claim that is red on main.

• Lambda carry. I tried passing a lambda argument through whole as its preserved shell under value_carried_unlowered. On srv1 at 8fc9f27 the module normalizes, but both lambda resolve claims stay red: the carried shell gives the resolver no scope for the parameter. That is a refusal moved from normalize to resolve, not a repair, so I dropped it (quiet-seal-543 / fierce-hawk-274 found the same). Real lambda lowering is the MQ lane's.

• Operator-operand fallback. It was dead. With bare-call and parenthesised-call operands, neither mutation (M1: fallback → refusal; M2: fallback → callee only) turned the claim red. The canary with the fallback removed regressed none of the six operator_operand_unread files: content_hash refuses only list_literal_unlowered, which MQ-1: list literals lower whole as their free-monoid construction, through the one value reader #12208 clears, and the other two stay clear. XL-2 follow-up to #12173: match-scrutinee controls, list/lambda RFM rows (three-baseline identity census posted as a PR comment) #12194 owns operands.

Claims matrix (srv1, neat-boar-16, at 8fc9f27, before the carry was dropped)

claim	fold @ HEAD	fold @ main fd9a659c6c
a_call_argument_lowers_to_the_authored_nested_call_with_arity_one	PASS	FAIL
a_call_scrutinee_lowers_whole_to_the_authored_call	PASS	FAIL
lambda: body uses its parameter → resolves	FAIL	FAIL
lambda: undeclared name in body → refused at its atom	FAIL	FAIL
scrutinee / argument resolve controls (3)	PASS	PASS

The two shape claims' red on main was most likely collateral from the lambda fixture refusing the one shared ingest. At this head they read their own ingest without the lambda fixtures, so their verdict is their own. An earlier mutation run (callee ^vpw_id → ^vpw_leaf) turned exactly the two shape claims red. Pending: a re-run at this head (shape claims green with their own ingest; both lambda rows red as enrolled), and CI's required floor.

Whole-route identity diff (srv1/srv2, at f813dd9, stacked on #12173)

83 of 85 added (cause, path) rows cleared. The 4 removed rows did not return, and the seven's subject became readable. The two stragglers, keyed_roster.dag and sugar.dag (head parse_grammar_choice_overlap_residue), were not in the parent's refusal set: exposure of a parse overlap, outside this PR. Most rows new relative to the parent are parent parse_g0_tokens_remain files moving to a later stage once they parse. Almost all of that clearing is #12173's, which has landed.

🤖 Generated with Claude Code

[gunbai-bot]

Review 70746: confirmed and fixed in 6f6839a. body_lower_try_match_from_captured now binds the scrutinee's accepted diagnostics as sd and returns diagnostics_merge(outer: sd, inner: d) with the arms' diagnostics, so an advisory raised while lowering the scrutinee is counted. That arm came from #12173, which this PR stacks on. The fix lands here because #12173 is already enqueued. — sent from silent-dove-314

[briansrls]

Source review hold on current head da4f5aab6e1d68ba8aa36a93e93fc338086a62dc:

1. The operator-operand fallback still lacks a direct discriminating claim. The PR body says its fixture passes under every fold and relies on the six-row whole-route delta. Please add a boundary/production specimen whose result changes when the body_lower_value_lowered fallback is removed, so this specific consumer is permanently protected.
2. Please correct the failure-mode framing before landing. XL-2: sequence-operand refs reach resolve — operator expressions lower whole, a sequence operand is never narrowed #12145 exposed pre-existing silent truncation; it did not create working-code regressions. The new row may describe the distinct pairing/whole-route-observation gap, but its identity and receipts should not call the fail-closed exposure itself a regression, and the semantic occurrence should remain attached to lowering_accessor_collapses_a_sequence_operand.
3. Rebase after XL-2 hotfix: call arguments lower as values; nested-call/record args no longer refuse the whole module (list args still refuse, located) #12173 lands and rerun the exact seven plus generation-two canary and current required checks on the resulting head.

The whole-value direction and explicit list frontier are otherwise consistent with the pass.