Lane A: kill E0255 std_dup at the emitter root (local-decl exclusion + use-symbol ownership) + re-probe wave carriers

[briansrls]

Lane A (XL): kills the E0255 std_dup class at its emitter root and rides the resulting re-probe onto the self-host frontier (probe TSVs, flips where cargo-green, plan-row updates).

Summary

Seam probe (decides the fix location, by execution). The E0255 List defined multiple times resurfacence (v1_deletion_plan std_dup_assembly_fix, first-error on the curated 4-module baseline) is emitter-root: the raw pre-assembly emit of the 04_infer closure already carries the colliding lines (pub use crate::v2_std_collection::List; into std_types.rs, which declares type List<element> itself; same shape for Map and GroupCompletion); cssl assembly introduces none.

Fix A — local-decl exclusion (reference_derived_use_lines, src/v1/05_emit_rust.dag): a name the module itself declares is never a synthesized-import candidate. The pass previously consulted only the bare-name last-write-wins registry, so a dual-std-tree homonym stole the row and the pass emitted a self-colliding pub-use. By execution on 04_infer: E0255 zero (was first-error); the unmasked residue is recorded per-module in the TSV. Discriminating RED control: new local_homonym arm in tools.namespace_import_closure_behavioral_transport (consumer declares an alias colliding with a provider enum) — verified in both directions (pre-fix binary emits the colliding line; post-fix does not); receipt PASS wet.

Fix B — cross-line use-symbol ownership (strip_repeated_use_symbols ending dedupe_rust_import_lines): within a module header a symbol binds through exactly one crate-use line; later repeats via a second transitive provider path drop, emptied lines disappear. This moves the last live cssl sanitize rule (dedupe_pub_use_symbols, its recorded dissolution trigger) to the construction seam. Proven on the measured firing (emit_produced closure, duplicate Optional via v2_std_grammar + v2_std_collection — now gone; whole-closure raw census 0).

Both fixes hold regen_stage0 at a 2-generation byte fixed point (divergence 0) — a green seed contains no duplicate bindings, so the passes are byte-inert on it by construction.

Named emit class (parent-directed carrier): curated_cargo_probe_record_fold_accumulator in frontier_probe_types.dag — the v1 emitter cannot faithfully emit compound-record fold accumulators (dropped-receiver field access, Rc<Vec<()>> mistyping); hit by execution while writing fix B; list-accumulator workaround noted in the row.

Probe instrumentation (classifier v3): curated_cargo_probe_one.sh gains residual_histogram (every rustc E-code + count — rustc masks later layers behind first_error; operator hard requirement) and raw_dup_pub_use (duplicate pub-use census on the raw emit BEFORE assembly rewrites — the by-execution receipt for dissolving the sanitize rule); unshimmed greens are CARGO_GREEN, PHANTOM reserved for lane-shimmed greens.

Wave results (complete)

• Definitive 21-module sweep TSV landed (docs/probes/curated_cargo_frontier_probe_sweep.tsv, classifier rule1-first-error-plus-residual-histogram-v3): E0255 = zero occurrences in any first_error or any histogram; raw dup-pub-use census = 0 on all 20 measurable rows (program_assembly is emit_fail — no artifact exists for the rule to fire on).
• cssl sanitize scaffold DISSOLVED (its recorded trigger fired): dedupe_pub_use_symbols / sanitize_emitter_artifact_in_place + fixture tests deleted; assembly now leaves emit-retained artifacts byte-untouched; remaining 8 assembly unit tests green; a fold_lowering re-probe through the sanitize-free assembly reproduces its sweep row byte-identically.
• Flips: ZERO — no module reaches cargo-green; the negative result is the receipt (2026-07-21 precedent). The unmasked successor classes are measured per-module: six deep modules share ONE type-ref histogram shape (E0433:15 E0425:14 E0422:8 E0109:8 E0107:6 E0391:1 — few shared-closure roots, exactly one E0391 cycle everywhere); materialization_carriers is the repr outlier (E0308:87 E0277:80 E0599:66 E0369:61); the rest is Lane D's emit-shape/dotted family.
• Frontier refresh: 6 knowledge_attributed rows → execution_measured on the sweep (attributed triggers removed per the 2026-07-22 disentangle ruling); 03_name_resolve corrected off ProbeStageAssemble (its closure contamination is gone — it now emits+assembles and refuses at cargo); honest_frontier_refresh_2026_07_23_note; census tests 10/10 green.
• v1_deletion_plan re-based: std_dup_assembly_fix → Placed with the emitter root named; import-closure pub-use gap closed; emit_representation_mismatch re-sized from ~4400-corpus-wide to one-module-deep (materialization_carriers); deep_module_lanes/probe_flip_fanout re-pointed at the measured successor classes.

The 03_normalize wet-receipt red (stale hand shims vs post-#7057 closure) is routed to stern-badger-729 — deliberately not absorbed here.

Test plan

• regen_stage0 --verify → divergence 0 (both fixes, 2-gen)
• 04_infer cssl probe (CSSL_STD_SEED_LINK=1): E0255 first-error → gone; residual histogram recorded
• emit_produced raw-emit census: duplicate pub-use symbols 1 → 0
• claim_batch --entry dag/test/claim/namespace_import_closure_witness_test.dag --function namespace_import_closure_receipt_holds --wet → PASS (incl. new homonym arm; red arm still refuses)
• A/B fixture vs pre-fix binary (origin/main control worktree): colliding line present pre-fix, absent post-fix

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[gunbai-bot]

Re review 41673 (both advisories addressed):

(a) Variant names in the local-decl filter — confirmed unreachable by execution, no code change needed. Fixture: a provider declaring a top-level type PilotRed homonym + a consumer declaring type PilotColor = PilotRed | PilotGreen | PilotBlue locally and returning bare PilotRed from a fn body, with a genuine cross-module ref alongside. The emitted consumer header synthesizes only the genuine refs (PilotWidget, pilot_provider_widget_default) and binds the local variant through use self::PilotColor::*; — no PilotRed use-line appears. Variant references don't enter the candidate walk as importable names (they ride is_known_variant routing plus the local enum's self-glob), so the exclusion set is complete for the class as framed.

(b) dissolve-on for strip_repeated_use_symbols — landed in strip_repeated_use_symbols_note (commit 3c9f8ed): the pass is named a bounded post-hoc scaffold and dissolves when the import surface derives from one symbol→provider ownership map built before any line renders (duplicates unwritable by construction), or when the resolver step-2 typed refusal replaces reference-derived synthesis (the existing nic_scaffold_dissolution_trigger event); the raw_dup_pub_use sweep column is the census that must read zero first. Seed regen included, divergence 0.

— sent from smart-crane-730

[gunbai-bot]

Addressed in 6aac97e (three items, one push):

1. Seed/authority drift on strip_repeated_use_symbols (review 41751 blocker, also the minor in reviews 41737/41750): the .dag rewrite to the linear symbol→owning-line owner map is now regenerated into the seed — v1_compiler_emit_rust.rs carries the owner-map shape (single-pass build, per-line projection), two-generation fixed point, regen_divergence_count=0. The quadratic per-line seen recomputation is gone from both carriers.

2. The CI regen-gate red on d559e92 (field 'alias' not found in type 'Present') was a merge-ref collision with v1 seed alias-decl parse arm (§13): syntax spec row + v1.compiler.parse arm + typecheck-env binding + regen fixed-point + corpus probe (dissolves the walk_target_alias_apply corpus --write gate) #7094, which made alias an item-declaration keyword in the v1 parser: my Present { value: alias } binding in rust_use_bound_symbol misparsed under the merged grammar (local checkout without v1 seed alias-decl parse arm (§13): syntax spec row + v1.compiler.parse arm + typecheck-env binding + regen fixed-point + corpus probe (dissolves the walk_target_alias_apply corpus --write gate) #7094 passed, merge ref failed). Renamed the binding to bound_name. regen_stage0 --emit-fresh --verify green locally on the merged tree.

3. Roster dedupe: merged origin/main and dropped this PR's nfr backfill block in favor of main's Main-latent nfr red: roster the two #7088 cheap-gate wildcard sites (unblocks PR floors) #7114 block (one authority for the backfill provenance); nfr witness trio PASS locally with the deduped roster.

— sent from smart-crane-730