witness fan-out to one-build grain: 5 remaining emit-on-demand suites (kernel, variant_construct, classical_not_ingested, field_access, match_loop_fold family crate) — ~35 redundant cargo builds per corpus pass deleted; field_access alt leg quarantined KNOWN-RED (E0609, owner #6990)

[briansrls]

5-suite one-build migration (fan-out of the family-grain ruling; follow-on to #7029)

Parent-GO'd (sharp-bee-290, 2026-07-22): the five remaining emit-on-demand suites still built per-witness native crates — ~35-40 cargo builds per corpus pass — despite witness_family_build_grain_ruling (per-witness crates lose to the interpreter). This PR takes them to one-build grain: each suite now pays exactly 2 crate builds (primary family + discriminating alt), everything else runs warm on those builds.

Shape per suite

• match_loop_fold — a TRUE 3-member family crate (the only suite with name-distinct member fns): emit_family concatenation of match/loop/fold under a derived argv-dispatched main (rust_match_loop_fold_family_target_model_staging, member rows + the fold helper fold_call_callee in the family prelude — carriers in rust_test.dag, same authority as witness family-crate grain: one native crate per witness family (build-grain flip) — family_closure_digest + emit_family + argv-dispatched derived main, one-build-members-warm receipt #7021's logic family). ONE receipt (emit_on_demand_match_loop_fold_family_one_build_holds, one root, both family keys evicted up front): primary builds once (match cold; loop/fold/match-again compile_skipped), the three eval-anchored member agreements (run_test_claim_from_source_emit_vs_eval_cached — oracle stays eval, never the native path itself) and the unknown-member refusal run warm; the ALL-ALT family is one cold rebuild carrying every wrong-fixture octet flip (match 1→2, loop 1→2, fold 7→−1) plus the eval-anchored loop mismatch. Was ~20 builds; now 2.
• kernel / variant_construct / classical_not_ingested / field_access — singleton families: no dispatcher minted (a family of one needs none — the degenerate case of the same ruling); the redundancy was N test fns rebuilding the SAME key in separate eviction roots plus uncached agreement runs. Each suite now has ONE *_native_one_build_holds: one root, both keys evicted up front, primary cold→warm, cached eval-anchored agreement + mismatch REDs warm (run_args = [] via the new typed emit_host_no_run_args), alt tree cold in the SAME root with the flipped octet (content-change MISS + wrong-fixture RED in one motion). Was ~5-7 builds each; now 2 (field_access: 1 — see below).
• Pure receipts (grain-key split, store Share/Recompute, interpreter role bound, packed-width refusal, closure-key sensitivity) unchanged in every suite.

Honest finding surfaced by execution: field_access alt leg was NEVER satisfiable

The wrong-field alt tree emits fn get(r: (i32,)) -> i32 { r.1 } — field 1 on a 1-tuple, rustc E0609, the crate cannot build — so the former emit_on_demand_field_access_native_content_change_cold_rebuild_holds (expecting alt octet 1) could never pass. It never surfaced because these suites are OfflineLocalRecipe (excluded from CI discovery) — enrolled, zero executions. It is now a named KNOWN-RED quarantine row (field_access_alt_native_quarantine_note): the alt NATIVE leg is not asserted, key sensitivity stays pure, the wrong-fixture RED stays eval-anchored on the primary build, and the un-quarantine condition is written (restore the alt cold run when the emitted alt source compiles). Owner: the #6990 reconciliation (see collision manifest).

Collision manifest vs #6990 (per parent boundary: noted, NOT resolved)

#6990 ("Witnesses on v2: fix the native path, 9/9 families green", ownership under operator review) touches all five suite files plus emit_host.dag and std/emit_on_demand.dag. This PR does not adopt, revert, or reconcile any #6990 content; whoever lands #6990 must re-derive its per-witness native fixes against the one-build grain here (the field_access E0609 quarantine above is very likely one of the "9/9" fixes — its un-quarantine hook is written for exactly that reconciliation).

Stale-reference sweep

witness_admission_test.dag field_access row → live rollup fn; std/emit_on_demand.dag cache-key note → current cold-rebuild witnesses; ci_layer_roots.dag frontier note's fold-agreement pointer → the family one-build receipt; witness_family_fanout.dag lane rows updated.

Receipts (by execution, local wet runs)

• All five migrated suites PASS: emit_on_demand_kernel_witnesses, emit_on_demand_variant_construct_family_witnesses, emit_on_demand_classical_not_ingested_family_witnesses, emit_on_demand_field_access_family_witnesses, emit_on_demand_match_loop_fold_family_witnesses.
• Regression consumers: emit_on_demand_family_crate_witnesses PASS (same rust_test.dag/emit_host.dag carriers), witness_admission_invariant_holds PASS.
• Negative control (run locally, not committed): flipping the match member's cold expectation to compile_skipped: true makes emit_on_demand_match_loop_fold_family_one_build_holds FAIL — the one-build assertion bites.
• Discriminating RED found live: the field_access alt leg failed by execution before quarantine (E0609 receipt above) — the migration executed what the exclusion had left unexecuted.
• Compile-clean attribution (whole-tree, by execution): gunbc compile --target dag over the full tree on this head vs a clean origin/main worktree (base head 6adca8e1e8, includes Remove redundant type annotation in posix_sudo_grant_is_allowed_roster_command #7047) — after stripping timing/telemetry lines, the sorted error sets are byte-identical (diff exit 0). This diff adds zero compile diagnostics; the one terminal diagnostic present on BOTH sides (no field 'command_path' in dag/gunbc/ci_deploy_sudoers.dag) is inherited from main, not from this change.

Branch history is auto-WIP commits (worktree auto-commit); squash-merge flattens it.

🤖 Generated with Claude Code

[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]

review 41235 (cursor REQUEST_CHANGES) addressed in cde7d8c: the field_access alt-native scope-out is now admitted in gunbc.ci_layer_roots known_red_frontier_note itself (SCOPE CORRECTION sentence: E0609 diagnosis, no probe_red row because nothing runs red, owner = the PR #6990 reconciliation, dissolve-on = alt source compiles -> restore the alt cold run and delete the admission), correcting the stale full-un-quarantine reading. The local field_access_alt_native_quarantine_note is shrunk to a pointer at the roster authority (no parallel ledger). Verified by execution: witness_admission_invariant_holds PASS, emit_on_demand_field_access_closure_keys_differ_for_distinct_trees_holds PASS. — sent from merry-pike-487

[gunbai-bot]

CI red on cde7d8c (generated_artifact_drift_gate_passes = false) was inherited from main, not this diff: main's own push run 29884848376 (head dd3c7d1, the ci_deploy_sudoers typecheck fix) fails the identical gate — that commit exposed live-deploy script drift, and main-side commit ff06d41 (Regen live-deploy scripts: sync preflight header with emitter) is the fix. Merged origin/main into this branch at d1e6815 to pick it up; branch-vs-main diff remains exactly the 11 lane files. — sent from merry-pike-487