test umbrella dissolution: corpus sweep B (src/v2 + orphan wire)

[gunbai-bot]

Summary

Re-opened after #7274 was squash-merged into sweep-A, which pushed #7273 over GitHub’s 300-file gh pr diff cap and made MODELING-COHERENCE fail-closed UNAVAILABLE.

Same content as #7274: dissolve src/v2/** umbrellas and wire check_orphan_helpers_or_err. Bases on restored sweep-A (f07e96d, 198 files vs main). Combined tip backup: session/proud-wren-892-sweep-ab-combined.

Stack

1. test umbrella dissolution: corpus sweep A (dag/test) #7273 — sweep A (dag/test) — land first
2. this PR — sweep B (src/v2 + orphan wire)

Test plan

• gh pr diff succeeds vs A (<300 files)
• CI floor green on affected set
• Orphan naming walk refuses dark helpers

Made with Cursor

[gunbai-bot]

Verified review 43264: confirmed. commit_workflow still enrolled dissolved emit_host_native_only_verdict_mock_all_holds / ingested_classical_not_emit_mock_shape_all_holds.

Retargeted explicit check_fns to the surviving mock leaves (native_only match/mismatch pair; classical_not resolves / arrow_has_match / octet_bridge / source_arm). Did not switch to file-grain — that would pull the heavier wet equals_eval leaves into per-PR scope.

— sent from proud-wren-892

[gunbai-bot]

Verified review 43320: confirmed. claim_witness_corpus_ci_runner.dag still enrolled dissolved *_keystone_holds names.

• Switched those rows to file-grain (function: "")
• Fixed round-trip entries manual/ → execution/
• Fixed gunbhub_react_browse_witness_test.dag path
• Dropped duplicate pick2 generality row (covered by file-grain pick entry)
• Deleted leftover pick_ingested_equals_eval_keystone_holds umbrella; promoted loop_eval_divergent_body_step_loop_holds and dropped its thin keystone wrapper

— sent from proud-wren-892

[gunbai-bot]

Addressed review 43330 on tip 69174d2444:

1. Fail-open zero-test fn skip — verified at orphans_in_module. Removed the silent “assume fixture library” arm. A *_test.dag with plain helpers and no enrolled test fn/test data now refuses those plains (demotion failure mode). The only exemption is structural path role: /extdeps/languages/ or /examples/ (cross-module export libraries; dissolve-on rename off _test.dag). test data-only modules still run reachability and orphan unreachable plains. Unit REDs: zero_enrolled_plain_helpers_refuse_demotion, test_data_only_module_still_orphans_unreachable_plains, cross_module_fixture_library_path_exempts_zero_enrolled_plains. Live dag+src/v2 census under the new rule: 0 orphans.

2. Hand-Rust receipt — added §7 scaffold marker TEST_MODULE_HYGIENE_ORPHAN_GATE_SCAFFOLD_MARKER with declaration-grain rg -F pin (== 1), named DELETE WHEN / ROADMAP 5-dissolve-patches lane deferral, and dag/gunbc/test_module_hygiene_scaffold.dag (Disposition = Scaffold + dissolve trigger + discriminator names).

— sent from proud-wren-892

[gunbai-bot]

Addressed review 43340 on tip 800be0663a:

Verified in pick_ingested_structural_lowering_test.dag: after the keystone umbrella drop, five pick1 nullary *_holds (normalized_swapped_arms_structural_red, pick_true_executes, pick_false_executes, swapped_arms_red, pipeline_resolves) were still plain fn while only pick2 leaves were test fn — unreachable under the module-local orphan gate. Promoted those five to test fn. Parameterized helper pick_ingested_normalized_branch_arms_magnitude_holds stays plain (reached from the structural_red test fn). File-grain corpus enrollment already covers the entry, so the promoted leaves execute.

— sent from proud-wren-892

[gunbai-bot]

Addressed review 43356 on tip after this push:

Verified pbf_tree_contains_atom_identity in parse_binding_fidelity_support.dag — whole-tree search found only its own recursive call; long-lane and body_lowering consumers import other pbf_* helpers, not this one. Deleted the dead helper (and the Atom/TypeNode imports it alone needed). Support-module note updated.

— sent from proud-wren-892

[gunbai-bot]

Addressed review 43367 (dashboard artifact /api/reviews/43367/artifacts/stdout.log):

1. Unreadable/unparsable *_test.dag no longer green — collect_orphan_helpers accumulates read/parse failures and returns Err (DESIGN §5). Same for umbrella roster parse failures (any failure refuses, not only when the roster ends empty). collect_dag_files now refuses on read_dir errors instead of silently continuing.

2. Directory-wide fixture bypass removed — is_cross_module_fixture_library_path is no longer /extdeps/languages/ or /examples/ prefix matching. Exemption is an explicit allowlist of the two zero-enrollment fixture libraries that remain on purpose (rust_test.dag, interp_test.dag), with RED coverage that a sibling under those directories is not exempt.

— sent from proud-wren-892

[gunbai-bot]

review 43601 — verified and fixed on tip.

1. ROADMAP row — 5-dissolve-patches does exist as gunbc.roadmap_authority ticket id, but ROADMAP.md renders only the headline. Rewrote the scaffold/deferral to name the concrete visible row: ROADMAP §1 drain the HAND_MAINTAINED queue, with authority ticket 5-dissolve-patches and plan anchor dag/gunbc/v1_deletion_plan.dag ^hand_queue_drain. test_module_hygiene.rs is already on HAND_MAINTAINED_STAGE0_FILES, so dissolution is that queue entry’s drain.

2. Receipt form — dropped marker-count rg as the checkable receipt. HAND-RUST GATE receipt is now the named unit RED plants (zero_enrolled_plain_helpers_refuse_demotion, allowlist/directory/unparsable/test-data/plain-data discriminators), matching the peer cli_run_workspace_root_scaffold pattern.

— sent from proud-wren-892

[gunbai-bot]

review 43604 — verified and fixed.

The whole-file return Vec::new() for allowlisted paths was a fail-open (§5). Replaced with declaration-grain export seeding:

• CROSS_MODULE_EXPORTED_PLAINS: path → named plains that seed BFS (long/ consumers for no_dual_representation_test / edit_locus_resolver_test)
• Non-exported plains in those files still orphan
• Unit REDs: cross_module_export_roster_seeds_named_plain_not_whole_file, cross_module_export_roster_alone_does_not_green_empty_file (11/11 hygiene unit tests green)
• Fixture libraries rust_test.dag / interp_test.dag renamed off *_test.dag (rust_test_fixtures.dag, interp_example.dag) so they exit the orphan scan without any bypass; module import paths unchanged

— sent from proud-wren-892

[gunbai-bot]

Addressed review 43610 (REQUEST_CHANGES — plain data as enrollment).

Verified: the reported fail-open was real — has_enrollment_surface treated any data decl as enrollment, and every data initializer seeded reachability, so data fixture = formerly_test_holds() with zero test fn/test data certified clean.

Fix on a71cd427b5:

• Executable enrollment = test fn / test data only (scan_test_decl_names); plain data is not enrollment.
• Plain data bodies seed reachability only when executable enrollment already exists.
• RED: plain_data_alone_does_not_enroll_demoted_holds (replaces the old “plain data counts as enrollment” unit).
• Honest enrollment for the four manuals that would otherwise go red: promote CompilesClaim/unified_claim/run_* rows to test data, and rust_wire witness_*_holds to test fn (already CI-executed by name).

Corpus orphan probe over src/v2+dag greens on this tip.

[gunbai-bot]

Addressed review 43617 (REQUEST_CHANGES — dead plain data beside an unrelated test).

Verified: with any test fn/test data present, every plain data body was still a reachability root, so test fn unrelated() + data fixture = demoted_holds() greened the demotion.

Fix: only test fn / test data are roots; plain data bodies are walked only when that data name is reached from an executable path (test → data → helper chains still cover). RED unrelated_test_plus_dead_plain_data_does_not_enroll + chain green control. Corpus orphan probe over src/v2+dag still greens.