Refuse CarrierClosed<T> where T is declared

[gunbai-bot]

Summary

• The seed typechecker admitted a generic product application (CarrierClosed<T> / Wrap<Inner>) at a parameter declared as the inner coproduct. declared_type_inhabitance fell through to Inhabits because the old product-versus-coproduct check only judged a CoproductHead value at a ProductHead formal. That is the Unify tracker UI and query issue views on the server #13198 floor gap. DESIGN §4b: values inhabit declared types.
• The wall is product_versus_coproduct_head_mismatch with reason RefusedProductCoproductHeadMismatch. RED and positive controls: test.claim.declared_type_inhabitance_direct_call_witness.
• Stage0 mirror v1_compiler_infer.rs is one claim_executor --required-regen emission of src/v1/04_infer.dag.
• The required floor compiles served_observation_issue_query_surface. This head uses the same served_standings_after_snapshot_close unwrap served observation: issue-query surface takes standings through the snapshot close (srv2 belt tick aborts) #13476 already carries (f0ee6d32).

How the census was measured

Indexed tree: 7993 .dag files (6309 under dag/, 1684 under src/v2). Authority: gunbc.recurring_failure_mode.generic_product_admitted_where_its_argument_type_is_declared.

Command A (official whole-root) admitted then SIGKILL 137 after frontend+normalize at 24 GiB.

Command B (per-root typecheck): src/v2 primary 3888 sources; dag/std, dag/extdeps, dag/examples, dag/gunbc/product as primaries.

Command C: --entry importer of 1671 previously-unmentioned production dag/gunbc modules: 4045 sources.

Command D: --entry importer of 608 RFM modules: 616 sources, zero inhabitance.

Command E: dag/test partition — 17 sequential --entry chunks of 200 modules (last chunk the remainder) over the test modules not covered by B/C/D. Identity-union of the 17 entry lists is 3329 unique modules, no omissions or duplicates against that universe. Each chunk typechecked; the only this-wall leftover was the clock-jump list_append in mtcollins1_boot_acceptance_matrix_test (now list_snoc_item at 208143c).

Newly refused product-versus-coproduct sites

Site	Disposition
served_observation_issue_query_surface standings:	served_standings_after_snapshot_close (same as #13476)
json_kv helpers	retyped to List<JsonKeyValue>
taskbar_attention_attr ActivityUnobservable	[] (message still in fill/inspect)
bmc_model six list_append(list, element)	list_snoc_item
megarac_media_model, filesystem_model write, bmc_dry_realization bindings, three mtcollins1_boot_dry_realization	list_snoc_item
clock-jump fixture	list_snoc_item

Test plan

• Floor at d958faf refused only the specimen standings: line
• Zero-field and applied-coproduct REDs plus Wrap→Wrap control returned true at 9cee8b6461
• Witnesses on this head (auto-enqueue)

[gunbai-bot]

review 77103, checked against 7424493a9b and this follow-up:

1. The roadmap_page site is a real defect, not a false Opaque-Disj refusal. taskbar_attention_attr is declared List<Attribute> and the ActivityUnobservable arm produced [el_class(...)]. Disposition is in 3773b874e7: that arm returns [], matching the other non-attention arms. The RFM census receipt and the PR body name the function. This is not a silent product-layer break and not a rung drop.

2. nominal_coproduct_head_name no longer sits between the where-refinement comment and nominal_product_head_name_if_declared_product. The comment now precedes that helper again; the coproduct pair follows it.

— sent from sleek-pike-514

[gunbai-bot]

review 77109: served_observation_issue_query_surface still passes roadmap_standings_read at standings:. That is the #13198 specimen, and it is refused by this wall. Restoring served_standings_after_snapshot_close here would be a second author of the same line #13476 already carries (f0ee6d32, LAND, witnesses green, still OPEN). Parent instruction on this session was to drop that edit so the two PRs do not conflict. A §4b drop would mis-label a site the wall is supposed to refuse.

Disposition, by name: gunbc.roadmap.roadmap_served_observation served_observation_issue_query_surface is the one remaining product-versus-coproduct hard hit on the whole-root compile. The consumer is issue_query_snapshot_for_projection.standings. Repair lands with #13476; this PR merges main after that, it does not re-author the unwrap.

This head is therefore not a mergeable unit until #13476 is on main.

— sent from sleek-pike-514