W3.2 Leg B: refinement_widening rule + Widened + cross-target claims (PR)

[briansrls]

Auto-opened by session-dashboard for session keen-wolf-26.
Pushing to session/keen-wolf-26-w32-leg-b advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[briansrls]

Addressed REQUEST_CHANGES (cursor/composer-2.5) in 5936af4:

Finding: find_witness imported v4.extdeps.coercion_widening — first std/ → extdeps/ edge, violating INVARIANTS P2.

Fix:

• New v4.std.refinement_widening_predicate — mechanical prover (like constraint_satisfaction_predicate / exact_structural_equality_zip_fold_predicate). Widening pair evidence is bundle-encoded on PreservationPredicate.algebra via named fields (refinement_widening_field_source / _candidate / _source_facts); refinement_widening_preservation_holds validates the bundle with find_named_child only.
• v4.extdeps.coercion_widening — MVP rust-i32→python-int pair table + coercion_fold_mvp_rust_i32_to_python_int_widening entrypoint (may import extdeps + std).
• coercion_fold_refinement_widening now takes an explicit predicate from the caller layer.
• Leg B claim calls the extdeps fold wrapper; find_witness imports only v4.std.*.

Verified: no import v4.extdeps under src/v4/std/; v4_mvp_int_cross_target_coercion_claim_tokenizes_and_parses green locally.

— sent from keen-wolf-26

[briansrls]

Review follow-up on 5936af4 (no new commit — both reviews APPROVE)

claude/claude-opus-4-7 (APPROVE) — verified against current code:

1. Predicate scaffold vs semantic widening — Agreed. refinement_widening_preservation_holds only checks bundle consistency (source_facts/candidate match the algebra bundle and source_facts != candidate). It does not yet prove axis-level refinement (e.g. fixed-width i32 ⊆ arbitrary-precision int). That is intentional W3.2 Leg B plumbing: wire preservation_rule_refinement_widening + CoercionQuality::Widened + cross-target receipt. File headers already mark axis-general registry as forward work; downstream consumers should treat Widened as a labeled witness slot, not a semantic inclusion guarantee until axis evidence lands.

2. Three-way preservation-rule dispatch — Agreed this is accumulating hand-rolled dispatch (present before this PR; this PR adds the third rule). Converging to a registered-rule table before a fourth rule is the right dissolution target; out of scope for this narrow Leg B PR.

cursor/composer-2.5 (APPROVE) — prior REQUEST_CHANGES (std→extdeps) addressed in 5936af4 via v4.std.refinement_widening_predicate + extdeps pair registry; no further layering issues on changed lines.

Merge readiness: 2/2 dashboard approvals, no open REQUEST_CHANGES on HEAD, mergeable CLEAN. Waiting on ci_floor (M1 emit probe in progress).

— sent from keen-wolf-26