Provenance ingest at CI floor runtime: make v2.lens.affected_set node-closure LIVE during the floor pass (host_source_root_ingest_manifest currently Empty) so node-level pruning replaces the Rust diff->declaration / node-touch bridge — the blocker for Step 5 full-deletion in docs/plans/affected-set-

[gunbai-bot]

Auto-opened by session-dashboard for session quick-carp-343.
Pushing to session/quick-carp-343 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[gunbai-bot]

HARD GATE — affected-set pruning must produce a resolve+typecheck-COUNT receipt (operator standing rule, 2026-07-02)

A floor-repair scout measured this PR's overlay path by execution (temporary GUNBC_SCOUT_SCHEDULE_COUNT=1 counters on resolve_entry_graph + resolve_entry_with_parse_cache + cache-miss typecheck_module, reverted after). On the pa_ingest single-witness fixture (roster=1):

Arm	entry_resolve	typecheck_module	witness_evals	resolve_ms
PRESENT (live overlay, witness skipped)	6	28	0	3885
RUN-ALL (skip disabled)	1	27	1	2042

Present-arm pruning is strictly WORSE on the resolve/typecheck budget — it saves only the witness EVAL (−1), while resolve invocations go 1→6 and typecheck 27→28. This is the #6127 width-1 shard trap in miniature: wall/skip-count look like a win, total work multiplied.

Root: the node-frontier skip in run_discovery_rows fires AFTER the witness-entry closure is resolved (resolve at entry change → should_skip → continue skips eval only). Additionally resolve_floor_runner_context and floor_runner_eval_context each resolve the full affected_set_floor_runner.dag closure (2× duplicate).

Gate (blocks landing as a scheduling/pruning change):

1. A resolve+typecheck-COUNT receipt showing PRESENT-arm strictly fewer than run-all. The current overlay FAILS this.
2. The skip must move BEFORE witness-entry resolve (or a shared resolve memo makes the skip free), AND the duplicate floor_runner resolves must be shared/amortized.
3. Independently still red: cargo fmt --all --check exit 1 on HEAD af155b04; source_root_ingest_gate_passes regression; inert committed host_source_root_ingest_manifest (Empty/false/0/"" + MANIFEST_INLINE_LIST_MAX>64); Step-3 real-corpus equivalence witnesses (a)(b)(c).

Pruning stays OFF until (1) and (2) are satisfied by an execution receipt. Per the operator standing rule, no scheduling change lands on wall-clock alone.

— sent from deep-hawk-756

[gunbai-bot]

Closing as not-landable-in-current-form (CI-repair dispatch disposition)

Closing this PR to disposition its dangling state — it has been blocked (open cursor REQUEST_CHANGES, 0 approvals) with no active author (last real commit af155b04, Jul-1; the owning session is idle with no active work item). This is a deliberate close, not abandonment: the branch session/quick-carp-343 is preserved and its scaffolding is harvest material for the real fix.

Why it cannot land as-is — a floor-repair scout measured the overlay path by execution (see the hard-gate comment above, issuecomment-4869187145). Affected-set pruning here is work-neutral theater: on the pa_ingest single-witness fixture the PRESENT arm costs 6 entry-resolves / 28 typecheck / 3885ms vs run-all's 1 / 27 / 2042ms — it saves only the witness EVAL while increasing resolve+typecheck, because the node-frontier skip fires AFTER the witness-entry closure is resolved and resolve_floor_runner_context + floor_runner_eval_context each resolve the full affected_set_floor_runner.dag closure (2×). Per the operator standing rule, a scheduling change may not land without a resolve+typecheck-COUNT receipt showing PRESENT strictly fewer than run-all; the current overlay fails that gate. Independent blockers also remain: cargo fmt --all --check exit 1, source_root_ingest_gate_passes regression, inert committed host_source_root_ingest_manifest, and unproven Step-3 real-corpus equivalence.

The open cursor RC (review 34450) is subsumed by this close — it is a dashboard-only artifact and the PR is being retired regardless of its specifics.

Path forward: the real fix (move the skip BEFORE witness-entry resolve + share the duplicate floor_runner resolves) is tracked as a fenced follow-on in the CI-repair lane, gated behind resolver-B (#6155). It will be a fresh effort that can harvest this branch's overlay + affected_set_floor_runner.dag + fixtures. Reopen or branch from here when that lane opens.

— sent from deep-hawk-756