product_introduction census: an identity join over named Conjs, not a count (red since #12766)

[gunbai-bot]

v2.test.execution.infer_product_introduction.product_introduction_derives_fully_evidenced_products_holds was a silent red on main, counted on #12860 under calm-hawk-793.

Bisected (BuildBuddy, path-restricted to src/v2/{compiler,std,extdeps} and dag/std, from 1375c80 to main): the first bad commit is d5dc100, #12766 ("named-fn parameter references as path-keyed references; infer grounds them through the index"). That change is deliberate. Each named-fn parameter reference became a path-keyed reference carrier, a Conj under <parameter-reference>, so the specimen body x + y now carries two more products. Both derive.

Not a literal bump (lane-manager direction; DESIGN §5: "completeness is an identity join, not a count"). The claim read conjs == 4. It now names every Conj infer is asked to derive: it walks node_inferred_subtree_nodes, the existing authority, so no second set of skip rules exists. Each Conj is named by its children, meaning each child edge's label plus the last atom that child's subtree carries, because a parameter reference's path ends at the parameter it names. The claim requires exactly this named list, in walk order, with each derived:

[grammar_production_identity_node_projection=dag_surface_module][grammar_production_captured_node_projection=y]
[add_probe=y]
[grammar_production_identity_node_projection=namespace_graft_module_body][add=y]
[x=dag_binding_type_int][y=dag_binding_type_int]
[<parameter-reference>=x]
[<parameter-reference>=y]

A non-derived Conj appears as !name, so it can't match. The single-Arrow assertion is unchanged.

Executed green on BuildBuddy: this claim, plus product_introduction_composed_evidence_carries_child_groundings_holds (control). The identity leaves the #12860 amendment, and the projection is regenerated.

🤖 Generated with Claude Code