Reshape the fn-arrow body skeleton so a call is a call: the marshal now emits the Transform carrier node_is_call accepts

[gunbai-bot]

Every node the fn-arrow body skeleton marshal produced was a TypeNode { connective: Conj }
-- calls included -- while v2.std.node_query node_is_call accepts only a
ComputationNode { behavior: Transform } whose first positional child is a callee reference,
the carrier v2.compiler.body_lowering_fold produces for a lowered body. A Conj type node
cannot match a Transform computation node, so across the WHOLE live skeleton substrate
node_is_call was false, node_callee_symbol was Absent, and every call-shaped lens over it
-- v2.lens.determinism transitive findings, v2.lens.machine_shape, the origin probe's call
walk -- was GREEN BY CONSTRUCTION. That is DESIGN section 4b's decoration: a check whose RED is
unauthorable in the population it ranges over, worse than absent because it is cited as coverage.

THE REPAIR IS IN THE PRODUCER, NOT IN THE CHECK. Widening node_is_call to accept the Conj
spelling would have ratified the fork and made the substrate's own definition of a call
advisory. marshal_generic now returns transform_record for an ExprCall with a non-empty
authored name -- exactly the case where the callee identity leaf is already edges[0] -- and
conj_record for everything else, including a call through an unnamed callee expression, which
has no callee reference to be first and which node_is_call would refuse anyway.

THE SECTION 3 RECEIPT WAS ALREADY STANDING AND IS NOW DISSOLVED.
v2.lens.no_dual_representation_test.scan carried skeleton_is_call,
skeleton_is_callee_atom and skeleton_call_argument_targets -- private copies of
node_is_call / call_callee_target / call_argument_targets respelled against Conj, because
the shared authority could not read its own substrate. One question with two answers. The
copies are deleted, not parameterised, and that module now calls the authority. Its
replace_param_in_output descends every non-atom kind rather than Conj alone: a Conj-only arm
would have returned bodies containing calls UNSUBSTITUTED, answering "not alpha-equivalent" for
exactly the bodies the check is about.

EXECUTED EVIDENCE, with the delete arm confirmed to have deleted.
marshaled_call_carries_the_transform_call_carrier and its control
marshaled_non_call_stays_a_conj_type_node run on the required cargo test -p v1-compiler --lib path. Reverting marshal_generic to the unconditional conj_record reddens EXACTLY the
first of them (4 passed, 1 failed) and restoring returns 5/5 green -- the mutation asserted its
own substitution before the verdict was read.

In .dag, the_lens_fires_on_a_transform_shaped_call and
the_lens_is_blind_to_the_conj_shape_the_marshal_no_longer_emits are enrolled as a pair
differing in ONE field, the node kind: one green alone cannot separate a correct lens from an
inexpressible subject. The blind row stays enrolled per section 4b(4) as the regression
control. The determinism fixture's seed body is now the call carrier itself rather than a shape
no marshal emitted.

Files one recurring_failure_mode row,
fixture_shape_diverges_from_the_producer_shape, bounded against
executed_conjunct_discriminates_nothing (there the RED was never authored; here the fixture
EXISTS and is what hides the defect) and against the sibling v2.lens.effect_reach disjunction
defect at adhoc-f0508f41-ae9, which presents identically and is repaired in the check rather
than in the producer.

NOT REGENERATED HERE: docs/design-failure-modes.md. Resolve refuses on this tree for reasons
that predate this change -- absence_classifier_default_bucket and
green_reported_over_a_population_the_instrument_does_not_own are each declared twice in
gunbc.recurring_failure_mode on main, plus unresolved Plan/NonEmptyStr types under
dag/gunbc -- so no projection can be produced from this authority until gunbc#10278 lands. The
.dag authority carries the row; the projection is a generated artifact and heals downstream.

🤖 Generated with Claude Code

https://claude.ai/code/session_011vmMNtRoPfgkR75q8feHYg

[gunbai-bot]

Both findings in review 59532 verified against the current head (1e188e8). One is answered with a receipt, one is disputed on the authority it cites. Neither is fixed by a code change and I say why.

Advisory — seed census for the hand-written Rust (accepted, receipt below). Measured on this diff, not estimated: coproduct_reflection.rs is +141/-1. The split is 22 comment lines, 3 production functions (nullary_behavior_variant, node_kind_computation_node, transform_record — 24 lines together), and the remaining ~95 lines are inside #[cfg(test)]: four helpers and the two discriminating tests. Production growth in the seed is 24 lines. NO SCAFFOLD PATH IS DELETED BY THIS DIFF and I am not claiming one. What this touches is a path ALREADY DECLARED a §7 scaffold with a dissolution trigger that predates me — v2.std.fn_index fn_arrow_skeleton_g2_marshal_host_scaffold_dissolution_trigger, which dissolves when fn-arrow body projection is a modeled substrate fold emitting the same skeleton facts, retiring the hand-Rust marshal expansion. That trigger is UNCHANGED and UNWEAKENED by this change: reshaping which node kind the marshal emits does not make the marshal any less scheduled for deletion, and the three new functions die with it. Deferral lane: the same #5364 corridor the existing trigger names. Tracking row: the trigger declaration itself, which is the carrier for this, not a new row — filing a second one would be the ledger-layer nicknaming the repo already refuses.

Blocking — §4c on the receipts: List<String> entries (disputed, and the citation does not reach this carrier). §4c's rule is about UNCLASSIFIED PROSE: "an ordinary String declaration whose sole purpose is commentary is misplaced or dead data", and its measured occasion was 215 dead prose rows hoisted into data …: String after // was made unwritable. The prose in a RecurringFailureMode row is not commentary beside program data — it IS the program data. gunbc.recurring_failure_mode is the authority DESIGN §4b names for this ledger, gunbc.design_document folds .authored over exactly these receipts, and the result is docs/design-failure-modes.md. Delete the strings and the projection is empty. That is the opposite of dead data: it is read, folded, and rendered.

The shape is also not mine to choose. receipts: List<String> is the carrier main established at 1af8892 for all ~90 rows, and its own header records the classification debt the review is describing — the analytic fields (misleading surface, actual question, failure mechanism, required guard) are "DELIBERATELY ABSENT", because "filling them for seventeen classes means authoring analysis that does not exist in the text today, which is semantic rewriting inside a structural cut and has no oracle." My row conforms to the declared shape. Typing the fields for ONE row would fork the type or leave one row shaped unlike its 90 siblings, which is the §3 defect this PR exists to remove, reintroduced one layer up in the same session.

So the finding is real as a CLASS OBSERVATION about the carrier and I am not disputing that the debt exists — it is filed on the carrier, by the carrier, today. It is not a defect in this row, and repairing it here would mean redesigning RecurringFailureMode and migrating every row, which is its own subject with its own oracle. Happy to open that as a work item; it should not ride on a marshal reshape.

— sent from cool-fox-470