Order-edge claims: assert A4's admission-wall refusal, not the pre-A4 order-absent reason - #12939

Fixes the three claims that #12514's merge-group differential found passing at base and failing at head, with #12770 (A4) between them. Measured red on main 224cdc4623f: each returned false.

Root cause: A4 moved where the refusal happens; no product behaviour broke

A4 (v2.std.node arrow_signature_order_conforms → arrow_domain_binds_a_name) makes an Arrow whose domain binds names but carries no order edge ill formed at Node admission. So v2.std.arrow_signature arrow_declared_parameter_order now answers ArrowParameterOrderMalformed for that shape, where before A4 it answered ArrowParameterOrderAbsent. Each claim read the old answer.

In each case the input is still refused with a typed reason; the refusal now comes earlier, at the wall.

The re-statements (the discrimination is kept)

• The two refusal claims first assert the route (the stripped Arrow reads ArrowParameterOrderMalformed at the admission wall), then the typed _malformed refusal. They still red if the renderer or closure realizer ever accepts the order-less form.
• The host-seam claim keeps its intent: the v1 host seam mints no second order constructor, and A4 relies on that, since export_signature_facts is order-less on purpose and consumed only through export_signature_declared_facts. It now asserts this structurally: no edge carries arrow_signature_order_label. It also asserts the raw shape reads Malformed on its own, which is why the only consumer is the lift, the one constructor that makes it well formed. Its neighbour exported_signature_carries_named_binders_and_declared_order still pins that the lifted signature carries the declared order.

All three modules are green on a remote run (19 claims), including the unchanged neighbours.

🤖 Generated with Claude Code