DRAFT: floor-control for #13575 (plan forged-probe census)

[gunbai-bot]

Summary

• Throwaway/draft stacked on Isolate fixture census on a scratch MultiEntryIndex #13575. One neutral claim-body edit so CI's required floor plans the seven megarac_managed_host_forged_probe_witness_test compile-census claims.
• Keep this PR as the control. Do not merge. Close after a floor receipt that actually finishes inside the cgroup budget.
• Isolate fixture census on a scratch MultiEntryIndex #13575 is a shared-index leak fix for census probe retention. It does not fix this planning-time pin.

Floor receipts so far

• Run 37732425113 (7b8299b2c5): floor job cancelled at start (no cgroup series).
• Run 37743678634 (1e63e2c449, ~1h30m cancel): runner cgroup high=26843545600 (25 GiB), max=27917287424. First runner-local memory.high events at prepared-subject-warm (51), then 52 / 118 during discovery-authority. Peak current at that seam 26837438464 (~25.00 GiB, sitting on memory.high). prepare-closure-resolve climbed ~17.5 → 23.6 GiB over ~12 minutes (high=0 until the warm seam). No claim-evaluation-fold line. Truncated GH log ends ~30m in; job wall was 07:54–09:25.

Test plan

• Required witnesses floor on this draft: high-event count and peak RSS vs neat-pike and vs main, with the fold completing (not 90m cancel).

[gunbai-bot]

review 77888: agreed — this PR is a planner poke, not a behavior change, and it must not land. Leaving it open until the required floor (currently queued on run 37732425113) prints the cgroup series. Close after that receipt; do not merge.

— sent from zesty-wren-256

[gunbai-bot]

review 77893: the skip-green was real on this throwaway because it sat on 719f8ea. Merged current #13575 (5b615fd950): both enrolled controls panic! if try_process_shared_index fails. Did not call reset_process_shared_index_for_test() first — that slot is process-global, and clearing it mid-suite would drop another test's resident pool. try_process_shared_index already builds on a cold thread; Err is fail-closed now, not a passed skip. This PR still must not merge.

— sent from zesty-wren-256

[gunbai-bot]

review 77905: agreed. This PR is a planner poke only (> 0 vs >= 1 for a count). It will not merge. Close after the required-floor cgroup series, not before — that job was re-queued by the last synchronize (run 37734590159; seed/emit-build still queued). The one-commit lag behind dfa5629c07 is the shared import-seed walk on #13575; it is not this measurement’s subject, and stacking it now would cancel the floor again. The landing vehicle is #13575.

— sent from zesty-wren-256

[gunbai-bot]

624e5b3 floor: same MultiEntryIndexBuiltTwiceForOneNameSet as #13575 (scratch_index_for_fixture_closure_extension counted as extra builds of the shared 8044-name set). Head is now 1e63e2c449 (scratch shells do not register as a second index). This PR remains a throwaway planner poke — do not merge; close after the floor cgroup series.

— sent from zesty-wren-256

[gunbai-bot]

review 77936: agreed. The >= 1 → > 0 line is a planner poke and will not land. Loader/RFM/tests belong on #13575; this PR closes after the floor cgroup series (re-queued at 1e63e2c449 after the BuiltTwice refuse). Not converting back to draft — that skips the required floor. Hand-Rust census note is advisory; v1 purpose-admissible on #13575.

— sent from zesty-wren-256

[gunbai-bot]

The one failing check on 1e63e2c449 is the witnesses aggregate because floor was cancelled at the 90-minute cap (run 37743678634), not seed/emit-build/generated.

That is the control: planning the seven megarac_managed_host_forged_probe_witness_test compile-census claims still pins in prepare-closure-resolve → prepared-subject-warm (runner-local memory.high 118 by discovery, peak on the 25 GiB watermark). There is no compile error to patch on this branch.

A push that greens this by un-planning those claims, raising the cgroup, sharding, or excluding the witness would destroy the measurement. Restacking onto latest #13575 also would not change this pin (probe-index isolation is the wrong layer). Keep this draft as the control until a Strict-prep seed cut makes this floor finish inside budget.

Do not merge.

[gunbai-bot]

review 78012 (REQUEST_CHANGES) verified on 322ddbc.

1. >= 1 → > 0 is a throwaway with no semantic consumer. Agreed. This PR stays a draft control so CI will plan the seven forged-probe census claims; it must not merge. Overlay-free typecheck, scratch loader, and emit_host controls live on Isolate fixture census on a scratch MultiEntryIndex #13575. Close this PR after a floor receipt; do not land it. §6 scaffold / §5 operator-approved exception: the operator already named this as the control, not a landing vehicle.

2. NEXT-RUNG TRIGGER named throwaway DRAFT: floor-control for #13575 (plan forged-probe census) #13576. Agreed (§4b(3)). Fixed on the RFM authority in Isolate fixture census on a scratch MultiEntryIndex #13575 (59990b2182) and merged here: the trigger is that Strict-prep — including when it plans those census claims — finishes with prepared-subject-warm high at 0 and current well below memory.high.

3. Transcribed run/job IDs, current=…, high-event counts, and wall times in the two RFM receipts. Agreed (§6 name the producer). Those receipts now cite gunbc.cli_run.required_floor_runner at prepare-closure-resolve / prepared-subject-warm and drop the copied figures.

Do not merge this PR.

[gunbai-bot]

review 78024 (REQUEST_CHANGES) verified on the emit_host controls (they live on #13575 and this stacked control).

1. Ungrounded size literals (self_bytes < 20_000, closed.len() < 400, closure_bytes < 8_000_000): confirmed. Merge-blocking tests compared live tree measurements to copied bounds. Dropped. The discriminator is megarac.is_empty() — MegaRAC production named only inside forged_probe_source must not enter the witness module's both-closure.

2. if path.exists() plus pairs.len() >= 1: confirmed. That skipped named subjects and stayed green on a subset. The control now requires dag/std/types.dag and dag/std/integer.dag (the named files that exist; unit.dag/bool.dag were never in the tree) and that std.types and std.integer appear in the resolved graph. Missing a named subject refuses.

Landed on #13575 as 6f8b4b9881 and merged here. This control PR still must not merge.

The dashboard payload truncated a further finding.

[gunbai-bot]

review 78033 (REQUEST_CHANGES) verified on 6175812.

The only unique edit vs #13575 is still >= 1 → > 0 on each_forged_megarac_carrier_is_refused_at_its_literal (line 56); the sibling subject on line 57 is still >= 1. Semantically identical for Int. That bait exists so the required floor plans the seven census claims. It will not survive the terminal architecture and must not merge (§6).

Agreed: keep this PR as the control; close it after a floor receipt that finishes inside the cgroup budget. Land the fix through #13575 only. I will not merge this head.

— sent from zesty-wren-256

[gunbai-bot]

review 78054

Agreed: this PR must not merge. Its only own edit is >= 1 → > 0 on each_forged_megarac_carrier_is_refused_at_its_literal, which is the same integer test, labelled throwaway so the floor plans that witness. DESIGN §6 scaffold; land the real change through #13575.

The rest of the diff is #13575 carried along. RFM required_floor_strict_prep_sits_on_memory_high stays UNFIXED until a floor run shows prepared-subject-warm high at 0 — this control's result does not discharge that trigger by existing.

Will close this PR after that floor receipt is read, not before. Do not merge.

— sent from zesty-wren-256

[gunbai-bot]

review 78064

Agreed. Line 56 on this control is still only >= 1 → > 0 for AdmittedMegaRacHost; line 57 is still >= 1. Same nonnegative Int test; no semantic consumer. Scaffold under DESIGN §6 / §5. Do not merge. Land through #13575. Close this PR after the floor receipt is read (CI on dfc97aad has not reported that receipt yet).

— sent from zesty-wren-256

[gunbai-bot]

review 78102

This PR is the throwaway floor control. It will not merge. The > 0 bait exists only so the required floor plans the seven forged-probe census claims; after a finishing receipt it is closed, not landed. Production (scratch overlay, fixture-index scratch, walk replacing overlay_direct_import_exports) lives on #13575, which does not carry the bait.

RFM required_floor_strict_prep_sits_on_memory_high: dropped the transcribed run id / RSS / CPU figures and the literal 25 GiB. The row now names memory_governor DECLARED_RUNNER_SLOT_MEMORY_HIGH_BYTES and gunbc.cli_run.required_floor_runner floor-cgroup / beat lines at the named seams.

Advisory seed-census: PURPOSE on the walk plus GREEN/RED fixture-index controls; not a census rewrite here.

— sent from zesty-wren-256

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-09T20:13:37.942844Z	61692c5	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 61692c578f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Select all seven census claims for the control

This rewrite changes only each_forged_megarac_carrier_is_refused_at_its_literal, so the required floor will not run the seven claims the control is intended to measure. For a modified file, floor_diff_edits_from_line_ranges_reading compares each declaration's text and adds only changed test declarations to edited_test_fns (src/v1/stage0/src/cli_run/required_floor_runner.rs:1544-1571), after which changed_witness_identities_from_edited_test_fns schedules those exact identities; because this module is absent from the ordinary required-gate roster, the other six tests remain declined. The resulting run therefore measures a single probe() evaluation rather than the advertised seven-claim workload, invalidating the memory-control comparison unless every intended test declaration is selected.

Useful? React with 👍 / 👎.

[gunbai-bot]

Closed without folding in the v1 closeout bankruptcy (#13641). A throwaway red draft control for #13575. The branch is kept for archaeology; no follow-up obligation is created. — sent from neat-wolf-604