affected set lens

[briansrls]

PR-B of the affected-set lane (design: docs/plans/interface-summary-declared-use-arity.md, merged #6244): witness enrollment rows gain their two missing facts — kind (what medium runs this witness) and span (which seams it certifies).

What

• std/realization_schedule.dag: WitnessKind = CorpusWitnessKind | ExecutionWitnessKind, WitnessSeam {producer, consumer}, WitnessSpan = SpanUndeclared | SpanSeams{seams}, witness_kind_eq (exhaustive match, no wildcard); ScheduleWitnessEntry carries kind.
• gunbc/commit_workflow.dag: CommitWitnessClaim carries kind + span; live roster rows stamped (CorpusWitnessKind, SpanUndeclared); span well-formedness (self-seam, duplicate seams, empty seam list all rejected); projection wildcards _ => acc replaced with explicit arms (fail-open fold → total match).
• workflow/ci_floor_plan.dag: corpus/execution partition now decided by kind via exhaustive match — the path-prefix heuristic is demoted to an agreement cross-check (witness_entries_kind_path_all_agree), dissolve-on noted.
• gunbc/merge_admission.dag: roster content-hash folds kind+span (hash flips once on landing — expected, it's a real roster fact change).
• Witness tests: 7 new span/kind witnesses incl. RED controls + kind-wins-over-path discriminator.

Receipts (by execution, ctrl-build remote, this head)

• commit_workflow_span_kind_witnesses — exit 0
• ci_floor_optin_roster_witnesses — exit 0
• ci_floor_plan_witnesses — exit 0

Notes for reviewers

• Seed mirror std_realization_schedule.rs intentionally NOT regenerated: regen_stage0 --emit-fresh exits 1 on clean main (pre-existing, differential-verified; residual 02_parse.dag diagnostics owned by open Register builtin get in the resolver registry (unbreaks regen self-compile) #6255). Rust never constructs the changed types on a live path — claim_executor decodes entry/function by field name; extra fields are inert.
• The ci check is red structurally: every ref incl. main is killed by the 10-min job timeout (fix in flight on fix/floor-timeout-verdict — timeout-as-failure semantics per operator ruling 2026-07-05).
• Span facts are declarative v0 (SpanUndeclared everywhere live) — selection consumption lands with the seam-factorization lane, per the design note's FLAG C.

🤖 Generated with Claude Code

[gunbai-bot]

CI red @ ebf1373 is structural, not this diff: the ci job hits the 10-minute timeout inside batch-1 compile-clean (whole-tree compile ~7min+ when the build is warm), and every main push run for the past day is 'cancelled' the same way (gh run list --branch main: 8/8 cancelled). The floor currently cannot fit the 10-min budget on any ref; batch-2 gates never execute. Not fixable per-PR — needs an operator call on budget vs batch-1 scope (ROADMAP §2's 'affected-set-scope the compile-clean' lever). PR-B verification is running out-of-band on the ctrl-build remote (compile check + witness runs); receipts will be posted here. — sent from gentle-owl-459

[gunbai-bot]

Re: both cursor reviews.

Findings 2 & 3 (second review) — valid, fixed by commit. The origin/main merge had brought in the enforcement-intent lane's three construction sites, all pre-dating the kind/span fields; all three are now stamped (commit_workflow.dag roster row, lens/enforcement/contract.dag, enforcement_live_witness_test.dag), and a post-merge exhaustive sweep over every ScheduleWitnessEntry/CommitWitnessClaim construction shows zero unstamped sites (the one residual grep hit is a destructuring pattern, not a construction). One correction to the finding's mechanics, which makes it more important, not less: v1 does not enforce record-field completeness — the old-shape rows would have decoded with kind as silent Null, not failed typecheck. Receipts by execution on this head: commit_workflow_span_kind_witnesses, commit_workflow_witnesses, ci_floor_optin_roster_witnesses, ci_floor_plan_witnesses, enforcement_consistency_gate_holds, roster_read_liveness_discriminates — all exit 0.

Finding 1 (first review, seed mirror) — deferral receipt, four layers deep, all by execution. The regen transport (regen_stage0 --emit-fresh, which is also SelfHostRealizedComparisonGate's declared transport) exits 1 on clean main and has all week; the blocker chain as of tonight: (1) 04_resolve.dag if-branch types — fixed by #6250; (2) 02_parse.dag builtin get — fixed by #6255; (3) regen_stage0 registry double-park from #6242's hand-maintained window — their in-tree comment declares the model-lag deliberately, with a named dissolution trigger ("until … std.content_hash memo rows emit"); (4) probing behind that guard surfaces json_quote/collect_dag_nodes scope failures in the self-compile, from the same #6242-changed module. So RegenVerifyGate cannot run green on ANY head today — this PR cannot regress a gate whose transport is unrunnable, and no live Rust path constructs the changed types (claim_executor decodes entry/function by field name; unknown fields inert, verified at claim_executor.rs:167-183). The mirror regen lands as an immediate follow-up when #6242's dissolution trigger fires; it cannot land before it by construction.

— sent from gentle-owl-459