Namespace flip Dispatch 1 — turn on namespace-only resolution (global flip) + inline-qualify the import-dependent references. Residue=0 banked on main by #7165. Plan: docs/plans/namespace-flip-last-28-root-a-two-std-defork.md

[gunbai-bot]

Summary

What: Dispatch 1 — flip NAME_RESOLUTION_POLICY_NAMESPACE_ONLY default false → true (global NamespaceOnlyY; runtime_rust.dag + regen pair v1_rt.rs / v1_compiler_runtime_rust.rs) and inline-qualify bare references the witness corpus exposes as ambiguous under the flip (v2.std.nat.*, v2.std.cardinality.*). Net diff: 92 files, no dag/std/* edits — pure qualify pass.

Why: Residue=0 on the compile-clean histogram (#7165) does not cover witness-scope closures (whole_tree_strict_resolve_exclusion_substrings skips /test/ and several lens/*). The floor / CI witness corpus under the flipped default is the completeness oracle. Operator direction (2026-07-24): (A) inline-qualify — do not consolidate forks in this PR.

Declared scaffold (section 5 — de-fork deferred)

Two genuine section-3 forks; inline qualification unblocks the flip; consolidation is a separate lane (operator: re-ground these concepts on cited extdeps foundations, not std-to-std fold):

1. Numeric-tower fork — Nat / Zero / Succ: dag/std/nat vs src/v2/std/nat.
2. Cardinality fork — TerminationProof / RankingDimension: dag/std/termination vs src/v2/std/cardinality.

Mechanism extras (not qualify churn)

• namespace_unique_on_chain_policy_test.rs: ResolutionPolicyGuard save/restore; ImportScoped cases bracket set(false) explicitly.
• Carrier notes in 04_env.dag / 04_sigs.dag (+ seed strings): default ON = NamespaceOnlyY.

Discovery oracle

gh run view 30122932237 --repo gunb-ai/gunbc --log-failed

First AmbiguousReference rows surfaced via CI floor batch 3 (discovery-corpus resolve), not histogram or quick-bee census.

Test plan

• cargo run -p v1-compiler --bin regen_stage0 --release -- --verify → regen_divergence_count=0
• cargo run -p v1-compiler --bin compile_clean_diagnostic_histogram --release → HISTOGRAM_TOTAL_HARD 0 (partial oracle)
• cargo test -p v1-compiler-tests namespace_unique_on_chain --release → policy discriminating witness green
• CI floor green under flipped default (completeness gate — in flight)

[gunbai-bot]

Addressed review 42283 (REQUEST_CHANGES):

1. namespace_unique_on_chain_policy_test.rs — ResolutionPolicyGuard now save/restores the thread-local gate; ImportScoped cases bracket set(false) explicitly; NamespaceOnlyY cases bracket set(true). cargo test -p v1-compiler-tests namespace_unique_on_chain --release green locally.

2. Carrier notes — updated 04_env.dag, 04_sigs.dag, and matching seed strings in v1_compiler_infer_env.rs / v1_compiler_infer_sigs.rs: default ON = NamespaceOnlyY (step 4); host bracket false = ImportScoped.

— sent from snappy-tern-366

[gunbai-bot]

Flip floor red — root-caused to a pre-existing interpreter bug, NOT the qualification pass. Fix is a standalone PR.

The floor red at batch 3: re-eval generated_nat_mul_annihilator_claim in generated_conformance_floor_test.dag: non-exhaustive pattern match on: 0.

This is not a resolution ambiguity and not a missed qualification — resolution succeeded. Root cause: the interpreter's native-coproduct reconciliation in match_pattern (v1_interpreter.rs, the Value::Int(n) if name == "Zero" || name == "Succ" arm) compares the full pattern name against the short literal "Zero". Under NamespaceOnlyY the resolver stamps a bare Zero/Succ pattern arm with its qualified containment path (v2.std.nat.Zero), so name == "Zero" is false, the grounded Int(0) (Zero→Int(0), #5428) matches no arm, and the match falls through non-exhaustive.

Pre-existing, not flip-caused — reproduced on origin/main (default ImportScoped) with an explicitly-qualified pattern:

FAIL qualified_pattern_matches_grounded_zero (runtime error: non-exhaustive pattern match on: 0)

The flip merely makes qualified pattern arms universal, so it surfaces corpus-wide. The Value::Variant arm already handles this via a pat_last short-name fallback (with a comment naming exactly this hazard); the native-coproduct arms (Nat Int, List Str/List, Optional/Witness Null) never got that treatment.

Fix — PR #7186 (standalone, off main): normalize the pattern arm name to its last segment (name_last) at every native-reconciliation guard — mirrors the existing pat_last invariant; a no-op for bare patterns, correct for qualified ones. Ships with a discriminating regression witness (qualified_pattern_reconcile_test.dag) that reds without the fix and greens with it. Verified by execution under the actual flip. This PR (#7178) should merge main once the fix lands, then re-run the floor — the qualification work here is correct and complete; it was blocked on this interpreter bug.

— sent from clever-pike-49