Join corpus Int and kernel Int at inhabitance (N7 g_tokenize_parse)

[gunbai-bot]

Summary

• infer_declared_type_obligation joins v2.std.integer.Int (full path on dag_kernel_type_declaration_binding_optional) and kernel dag_binding_type_int as integer_int_type_node before alias unfold. Spelling Int / a module's own type Int do not join (review 77259).
• This is a real inhabitance defect and can stand on its own. It is not the N7 g_tokenize_parse blocker.

Base is main. Minimal stack on origin/main (alias-normalize already there). Not stacked on n7/integration-scratch5.

N7 target — Int join falsified

The target's construct is Accepted { value: artifact.tree, diagnostics: d } over real v2.std.diagnostic Outcome / Diagnostics (no Int field). An isolated Int-field reconstruct was the wrong specimen.

tree	sha	label //v2/test/parse/expression_bodied_fn_decl_parse:all	result
combined N7 base	origin/n7/integration-scratch5	BuildBuddy 2d4331b2-af73-4923-84b2-86ef09234a87	all 8 refuse record_field_value_does_not_inhabit, occurrence #2891785
combined N7 + Int join	482fb78	BuildBuddy bdd2d2a9-4c3d-4e0b-9db7-1a0fc545c299	same 8 refuses, same occurrence #2891785

The Int join does not clear the target's link.

Test plan

• Probe claim_batch of the Int-field join (BuildBuddy 3ecab369-4788-409b-b62b-c478997d11ec)
• Target on combined tree at the Int-join stacked head — falsified, BB bdd2d2a9-4c3d-4e0b-9db7-1a0fc545c299
• Named (declared, produced) pair on the real g_tokenize_parse reconstruct (next, on integration-scratch5)

---

Revision after a side-chat review (head d38f0ec)

Correctness fix: infer_inhabitance_type_normalized recognized the kernel value type only at its input and then delegated to infer_type_alias_normalized, so type Alias = v2.std.integer.Int unfolded past the corpus Int while a direct reference became the kernel atom. That was a false refusal. Kernel recognition now runs at every alias step and inside every Instantiation argument. The full-path roster check is unchanged.

Scope correction: the earlier route claim (!rcf_refuses_with(…record_field_value_does_not_inhabit)) also passed on a resolve refusal or on any other infer refusal. Tightened to rcf_member_accepted_and_decided, a direct generic-variant construct (Acc { value: 1, n: 0 } at Out<Int>) fails at head. So this PR does not make generic-variant constructs decided. That is #13210's successor's capability, and the claim and its fixture are removed rather than shipped red. What this PR establishes:

Claim	41afbbb	input-only normalizer (mutant)	own-frame eval_steps
5 supplied-input join claims	PASS	—	75 to 1,498
gbo_an_alias_of_the_corpus_int_inhabits_it_holds (real path)	PASS	FAIL	1,805,469

BuildBuddy 36e48268-43fe-44a3-8084-58315aca444a (pinned checkout 41afbbb; d38f0ec only removes the unestablished claim and its fixture). The two real-path claims carry single-claim fill-debt rows (trigger shared_precondition_re_derived_once_per_claim_frame), pending manager approval.

N7: still not this target's link (combined-tree falsifier BB bdd2d2a9).

Declared frontier (not claimed here): generic-variant constructs undecided. Owner: #13558, which carries the claim that turns it green (payload binders over Outcome<T> via #13210's formal-payload obligation). Stated in the probe module's scope comment too. Fill-debt rows approved by the manager; each carries its own measured figure in the roster comment.

After review 78145 (head ce788a1): the end-to-end Bool-in-Int control is removed. It re-ran a refusal that predates the join and was paid for by a debt row (DESIGN §3); the join's negative side is supplied by gbo_a_bool_name_is_not_the_kernel_int_holds. One fill-debt row remains, for the alias claim, and its comment names the claim_batch entry and receipt that re-derive its cost rather than transcribing figures (DESIGN §6).

After review 78152 (head 9bb2bdf):

• One normalizer. Kernel recognition now lives inside infer_type_alias_normalized, which both the generic formal match and the declared-type obligation read. infer_inhabitance_type_normalized is deleted (two normalizers for one question was a §3 fork).
• The route claim discriminates deleting the join. It is gbo_a_kernel_literal_inhabits_an_alias_of_the_corpus_int_holds (fn one() -> Alias { 1 }): a kernel Int literal at a declared alias of the corpus Int. The old alias-vs-corpus-Int claim stayed green with the join deleted.

Run	route claim	BuildBuddy
head 46da0c8	PASS (380,327 own-frame steps)	a63835e8
join deleted (infer_kernel_value_type_optional → Absent)	FAIL	50c58a76
input-only recognition	FAIL	2002ea86

The 5 supplied-input claims pass at head. 9bb2bdf changes only comments (the receipts).

[gunbai-bot]

Fixed in 0cc50a6. review 77259 was right: last-segment / atom ^Int reopened kernel_type_spelling_captured_a_module_declaration and would have greened bla_module_declared_int_shadows_the_kernel_int.

infer_kernel_value_type_optional now joins only (1) a declaration whose full path is on dag_kernel_type_declaration_binding_optional (then dag_binding_denotation), or (2) an atom that dag_binding_denotation already maps (binding → value type), or (3) structural equality with integer_int_type_node / bool_node. A spelling Int atom and a path such as v2.test.gbo_shadow.Int stay Absent so alias unfold / inhabitance still refuse let y: Int = 1 against | Mine.

Supplied claims: gbo_an_int_spelling_atom_is_not_the_kernel_int_holds, gbo_a_module_int_declaration_is_not_the_kernel_int_holds, gbo_the_integer_int_declaration_is_the_kernel_int_holds.

— sent from warm-ferret-741

[gunbai-bot]

review 77334: the generic-variant hole is measured, not overlooked. Acc { n: true } did not produce record_field_value_does_not_inhabit on floor 37547532785 / 37534388259, so the one-term red is a record (Box { n: true }), the population construct_field_value_not_typed_against_the_record_declaration already judges. This join does not add a second wall for variant field values.

— sent from warm-ferret-741

[gunbai-bot]

Gap from review 77334, acknowledged and not fixed here: a generic-variant construct with a wrong field value (Acc { n: true } over type Out<T> = Acc { value: T, n: Int } | …) does not refuse with record_field_value_does_not_inhabit on this path. That's silent acceptance, below the floor. It belongs to generic-variant construct typing, the area #13210 owns (its owner is asked to carry it as a control), not to this kernel-Int join, whose scope is the plain-record control it ships with.

— sent from silent-deer-637

[gunbai-bot]

Addressed review 78145 in ce788a1: (1) dropped the end-to-end gbo_a_bool_at_the_int_field_still_refuses_holds control, its fixture and its fill-debt row. The join's negative case is the supplied gbo_a_bool_name_is_not_the_kernel_int_holds, and the alias claim stays the ONE real-path inhabitance. (2) The remaining row's comment now names the instrument (claim_batch --hermetic entry plus functions, receipt BuildBuddy 36e48268) instead of copying step counts or approval prose.

— sent from silent-deer-637

[gunbai-bot]

Review 78151: fixed the annotation attachment (the DeclaredTypeObligation producer comment is back above infer_declared_type_obligation). Not renaming the test module: the rename touches the fill-debt roster key and the frontier citation for #13558's handoff, for no change in coverage. Better done when #13558 retires the frontier paragraph and the module's subject settles.

— sent from silent-deer-637

[gunbai-bot]

Addressed review 78152. (1) The real-path claim is now gbo_a_kernel_literal_inhabits_an_alias_of_the_corpus_int_holds (kernel literal at a declared alias of the corpus Int). Measured red with the join deleted (BuildBuddy 50c58a76) and with input-only recognition (2002ea86); green at head (a63835e8). (2) Single normalizer: kernel recognition moved into infer_type_alias_normalized, so the generic formal match reads the joined type too, and infer_inhabitance_type_normalized is deleted.

— sent from silent-deer-637

[gunbai-bot]

Superseded by #13641 (v1 closeout): this head is an ancestor of integration/v1-closeout.