REVIEW BRIEF — phase6-probe-janitor (Phase 6: killed probe runs are reaped — F15)
REV: e2271de17c3cd14a54bafda7d3f20dd8bb0d0331
FINAL: 8e9d6af9865837df064b0f6f930116a49c8ce525
REVIEW ACCEPTANCE: scripts/check-scope e2271de17c3cd14a54bafda7d3f20dd8bb0d0331 8e9d6af9865837df064b0f6f930116a49c8ce525 docs/research/phase6-probe-janitor-brief.md
FOCUS: the reaping above all — does the kill-path grade leave no probe tree (the pin's emptiness assertion red at REV, green at FINAL), is the seam one registry the teardown calls rather than a caller contract, and does F30's chimeric-red claim hold against the brief's text and the delivered pin.
SEVERITY SCALE: P0 blocks the step; P1 is serious; P2 is minor.
QUESTIONS TO ATTACK:

  F30 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
  The brief's named red is chimeric as literally written and its GOAL overstates what an in-`_run_probe` reaper can do. Measured at REV: a probe that merely raises `KeyboardInterrupt` mid-grade is caught by `_run_probe`'s contained `BaseException` arm and the delivered `finally` already reaps both `mkdtemp` trees, so that pin passes at REV (`_run_probe` returns `GRADE_INFRASTRUCTURE_ERROR`, after-minus-before empty); the grading-seed review measured the same ("the cleanup is correct on every path this review exercised"). The reading taken: the pin models the probe's sandbox root sibling (`f"{execution_tree}-root"`, `hidden_oracle.py:233`'s `new_root`), the tree the finally-only reaping does not name and the one the emptiness assertion catches at REV. Separately, the GOAL's "a SIGKILLed ... grade cannot fill the shared tmpfs" clause is not reachable inside `_run_probe`: a SIGKILLed runner executes no Python and runs no reaper, so the SIGKILL residue F15 names still needs the cross-process tombstone/janitor F15's own next action names — out of this brief's IN SCOPE.
  FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the janitor lands as described (registry + finally-called reaper over the two trees and the sandbox root) and is green at FINAL.

  1. The reaping above all — the delivery's one question. Drive it live against FINAL on scratch copies diff-verified against git show e2271de:… before use: (a) the kill-path grade leaves no probe tree — the janitor pin's after-minus-before emptiness assertion red at REV on the finally-only reaping, green at FINAL with the registry reaper — re-drive both directions yourself and name the outcomes; (b) the seam is one registry the teardown calls (module-level `_PROBE_TREES` keyed by path, prefix-scoped, silent on failure) rather than a caller contract — which pin goes red, and on which member, if the registration of the sandbox root regresses (the `:725`–`:726` lines dropped)? (c) the reaped set is exactly the probe's own trees — a reap failure is silent and the prefixes bound the blast radius — falsify the scoping or verify it (which paths can the reaper remove, and which can it not?). Report each guarantee as falsified-with-reproducer or attacked-and-not-falsified with the method.

  2. F30's chimeric-red claim. The implementer's F30 says the brief's named red ("the raising probe's grade leaves its `orchestrator-probe-*` tree standing") does not reproduce as written — a merely-raising probe is already reaped by the delivered `finally` — while the GOAL's SIGKILL clause is unreachable inside `_run_probe`, so the pin models the sandbox-root sibling and the SIGKILL residue still needs the cross-process janitor. Verify: re-drive the pin at REV (merely-raising probe against the delivered reaping — green or red?), re-drive the sandbox-root shape at REV (standing tree or reaped?), and adjudicate — is F15 fixed by this janitor (the delivered red class is the unnamed third tree), or does the chimeric red mean the janitor fixes nothing the delivered `finally` did not already close, landing defense-in-depth the brief did not ask to grade? Is there a second consistent reading (the red exists on a shape this environment draws differently), or is the brief's red prescription simply wrong and the step returns to planning? File a finding per falsified class.

  3. The report's forms and counts (J-class). FINAL is the work commit 8e9d6af9865837df064b0f6f930116a49c8ce525, never the report commit d1ca08f — verify rather than taking the report's word. Re-run the brief's ACCEPTANCE lines at FINAL (fast suite 984 passed, 109 deselected with the delta exactly the one janitor pin against the REV scratch's 983 passed, 109 deselected; the probe module battery 16 passed; static gated clean with advisory Found 104 errors. unchanged; docs printed line with 0 errors — every figure the gate's own printed line, rule 14, never the REV line quoted in CONTEXT); TRIPPED: none matches check-scope's tripped: line and REVIEW ACCEPTANCE is green run first. git diff e2271de17c3cd14a54bafda7d3f20dd8bb0d0331 8e9d6af9865837df064b0f6f930116a49c8ce525 --stat touches only the IN SCOPE files plus the brief (no committed test edit outside them — verify). SHAPES' none-claim against rule 7: the one named shape built as the pin — adjudicate whether the pair holds the shape whole or the SIGKILL residue needs a member the suite should still build. INHERITED's F15 fixed-close against the received set. Every field of the report present and filled, nothing outside the template added. Standing battery: scripts/check-tests (your full run — record it in VERIFIED).

--- BEGIN BRIEF UNDER REVIEW (copy of docs/research/phase6-probe-janitor-brief.md; the committed file is authoritative) ---
IMPLEMENTER BRIEF — phase6-probe-janitor (Phase 6: killed probe runs are reaped — F15)
GOAL: A killed probe run leaves no probe tree behind: `_run_probe` reaps its intermediate trees on the kill path the way its `finally` reaps them on the settled path, so a SIGKILLed or deadline-killed grade cannot fill the shared tmpfs inode table — pinned by `tests/test_probe_oracle.py` itself.
CONTEXT: The phase-6 chain stands delivered with non-terminal report rows ([record.md](../record.md#record): REV `e2271de` latest); its accepted reviews converge (no P0/P1) but carry F15 (P2, ORIGIN: CODE): killed runs leak intermediate probe trees and no janitor reaps them — the shared tmpfs inode table filled to 3 free and the standing battery's first run died ENOSPC until 81 leaked trees were pruned by hand. The leak is live on this host: 38 `orchestrator-probe-*` plus 25 `orchestrator-probe-exec-*` trees stand under `/tmp` with `/tmp` at 96% inodes. The site is `_run_probe` (`src/orchestrator/oracle.py:605`): the `finally` (`:743`–`:749`) reaps both trees only when the grade returns through it — a SIGKILLed runner never returns, and the trees stand. `REV` names the step's base — the parent of this brief commit — and is not claimed to be `FINAL`'s parent.
INHERITED:

From `docs/research/phase6-scan-anchor-implementer.md`, its `INHERITED` block verbatim:

  From `docs/research/phase6-probe-scope-implementer.md`, its `INHERITED` block verbatim:

    From the brief's INHERITED block, the items still open, copied verbatim (the chain through the sweep's implementer report; the review's copy moves the same F13–F17/F21/F22/F23 blocks):

    F13 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief names the pin's red as "N consecutive grades of the known-good tree produce non-identical verdicts (PASS on one run, CANDIDATE_TIMEOUT on the next)". In this environment the known-good tree's `gate-check-tests` grade is stable well under the fixture's 60.0s deadline (ten pin runs, 50 grades, ~37–46s each, every grade PASS), so the pin passes at REV. The lottery's grade-flip needs a candidate whose suite fails — a failing tree's hypothesis tests shrink, and the shrinking varies the suite's wall — and the near-deadline flip is the neutralisation test's broken-source candidate (FAIL→CANDIDATE_TIMEOUT, 3/3 at REV). The pin's assertion and the seeded determinism are correct; the brief's stated candidate is simply not the one that flips here.
    FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the pin and the seed land as delivered.

    F14 (P2, ORIGIN: DOCUMENT, carried) — the F8 red evidence is load-borne and the implementer report does not say so; a later actor re-driving the red on an idle host sees green and may conclude the red is chimeric. Next action is the planner's: carry the load condition into the carried F8 block (or a later committed brief's correction), the seed and the pins stand as delivered.

    F15 (P2, ORIGIN: CODE, carried) — killed runs leak intermediate probe trees and no janitor reaps them; the shared tmpfs inode table filled to 3 free and the standing battery's first run died ENOSPC until 81 leaked trees were pruned by hand. Next action is the harness owner's (a tombstone/janitor keyed by the run identity) or a planner ruling that scratch hygiene is each session's duty.

    F16 (P2, ORIGIN: CODE, carried) — the seeded determinism is wall-bounded: at the fixture's 60.0s deadline the near-deadline candidate's seeded grade flips FAIL→CANDIDATE_TIMEOUT under host load (measured once mid-run and once at a 62.44s wall), so the pin's green determinism is stable-half evidence only. Next action is the fixture owner's: raise the grading deadline for near-deadline candidates (the 300.0 default gives a >=5x margin here) or an OWNER ruling that the deadline's load envelope is governed, not deterministic.

    F17 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief's boundary-fix instruction is internally inconsistent (FINDINGS, F17): the double-fork pin must leave no descendant alive after the exit while the F10 q1e boundary shape must survive the fix, and the two name one live shape (a session-escaped middleman spawning a session-escaped sleeper then exiting before the deadline, the sleeper adopted by init) that no container boundary can kill and spare at once. The GOAL reading was taken (F11/F12 fixed by the container boundary, both pins green at FINAL, no committed discriminating member changed verdict); the boundary this step retires is the named init-adopted orphan, and the new boundary is what the child never forked. Next action is the planner's: a later committed brief's text correction — the q1e-survival sentence, F12's "without also catching the documented boundary" clause, and the boundary sentence in the module docstring this step corrected in code.
    FIXED: no regression test — the reading is delivered in the work commit and its evidence is the two reds above and the FINAL battery above; the correction itself is a committed brief's prose, which has none.

    F21 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
    The implementer report's supervisor-module battery line records the module battery widened by the review module (`.venv/bin/python -m pytest tests/test_phase6_supervisor.py tests/test_review_phase6_supervisor.py --durations=5` — `15 passed in 17.72s`) and never the brief's documented form (`.venv/bin/python -m pytest tests/test_phase6_supervisor.py --durations=5`) with the documented form's own printed line. The wide form is a strict superset (the module's 13 pins plus the review module's 2 probes), green, and every figure it carries is its own printed line — nothing is evaded; but the brief's acceptance names the documented command, and the report's own template requires each claimed figure to name the command that actually ran while this acceptance line names both the documented form it did not run and the wide form it did. This review ran the documented form at FINAL — `13 passed in 12.46s` — so the figures stand; what carries is the report-text deviation from the brief's named battery.
    FIXED: no regression test — a document deviation in a committed report, which only a later step (or the planner's correction of the carried note) absorbs; the documented form's gate line is in this review's VERIFIED.

    F22 (P2, ORIGIN: BRIEF) — the brief's review-block copy of the carried F17 finding is truncated mid-sentence (the literal `(line truncated to 2000 chars)` marker), which rides as a carried text-correction to the planner.

    F23 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief's named red at REV is not reproducible — "at REV, with the delivered single-capture _kill, the racing forker's final tail forks plus the double-forked sleeper survive teardown adopted out-of-tree". In a NAMESPACE_V1 drive the workload is the PID-namespace init, _descendants always captures it (descendants=128 namespace_inits=[…]), and its SIGKILL makes the kernel reap the whole namespace (init death kills the namespace; measured: 124 live descendants after a weakened kill, 0 after the init dies), so the delivered _kill already leaves no descendant alive and the pin passes at REV. The sweep lands as the brief's requested defense-in-depth port and the pin as a regression pin; the named red is chimeric for the sandbox, and the load-borne leak F20 measured rode the nested-grading/oracle path, not a standing sandbox-drive race.
    FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the sweep and the pin land as delivered, and the pin's red class is the kill-surface mutant above.

    Then each finding marked carried by this step, with its ID:

    F26 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the repoint's assertion form — the precise inverse `assert carrier.pid not in found` taken over a literal whole-scan `assert not found`; see FINDINGS above; its next action is a later committed brief's text correction).

    Received-set dispositions kept: F19 fixed by the prior step's absorbed pin; F20 fixed as recorded in the sweep report; F11/F12/F18/F25 fixed as recorded in the reports that filed them; F24 fixed by this step (above); F13–F17, F21, F22, F23 carried with their named owners and text corrections.

  From `docs/research/phase6-probe-scope-review.md` — its `INHERITED` block, then each carried finding's block from its `FINDINGS` section, moved verbatim (rule 2 governs the planner's copy of this block; the review's received set is the probe-scope brief's carried chain plus its F24/F26, F24 fixed at FINAL, F26 carried):

    From the brief under review, the still-open items copied verbatim (the chain through the probe-scope step's own carried set):

    F13 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief names the pin's red as "N consecutive grades of the known-good tree produce non-identical verdicts (PASS on one run, CANDIDATE_TIMEOUT on the next)". In this environment the known-good tree's `gate-check-tests` grade is stable well under the fixture's 60.0s deadline (ten pin runs, 50 grades, ~37–46s each, every grade PASS), so the pin passes at REV. The lottery's grade-flip needs a candidate whose suite fails — a failing tree's hypothesis tests shrink, and the shrinking varies the suite's wall — and the near-deadline flip is the neutralisation test's broken-source candidate (FAIL→CANDIDATE_TIMEOUT, 3/3 at REV). The pin's assertion and the seeded determinism are correct; the brief's stated candidate is simply not the one that flips here.
    FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the pin and the seed land as delivered.

    F14 (P2, ORIGIN: DOCUMENT, carried) — the F8 red evidence is load-borne and the implementer report does not say so; a later actor re-driving the red on an idle host sees green and may conclude the red is chimeric. Next action is the planner's: carry the load condition into the carried F8 block (or a later committed brief's correction), the seed and the pins stand as delivered.

    F15 (P2, ORIGIN: CODE, carried) — killed runs leak intermediate probe trees and no janitor reaps them; the shared tmpfs inode table filled to 3 free and the standing battery's first run died ENOSPC until 81 leaked trees were pruned by hand. Next action is the harness owner's (a tombstone/janitor keyed by the run identity) or a planner ruling that scratch hygiene is each session's duty.

    F16 (P2, ORIGIN: CODE, carried) — the seeded determinism is wall-bounded: at the fixture's 60.0s deadline the near-deadline candidate's seeded grade flips FAIL→CANDIDATE_TIMEOUT under host load (measured once mid-run and once at a 62.44s wall), so the pin's green determinism is stable-half evidence only. Next action is the fixture owner's: raise the grading deadline for near-deadline candidates (the 300.0 default gives a >=5x margin here) or an OWNER ruling that the deadline's load envelope is governed, not deterministic.

    F17 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief's boundary-fix instruction is internally inconsistent (FINDINGS, F17): the double-fork pin must leave no descendant alive after the exit while the F10 q1e boundary shape must survive the fix, and the two name one live shape (a session-escaped middleman spawning a session-escaped sleeper then exiting before the deadline, the sleeper adopted by init) that no container boundary can kill and spare at once. The GOAL reading was taken (F11/F12 fixed by the container boundary, both pins green at FINAL, no committed discriminating member changed verdict); the boundary this step retires is the named init-adopted orphan, and the new boundary is what the child never forked. Next action is the planner's: a later committed brief's text correction — the q1e-survival sentence, F12's "without also catching the documented boundary" clause, and the boundary sentence in the module docstring this step corrected in code.
    FIXED: no regression test — the reading is delivered in the work commit and its evidence is the two reds above and the FINAL battery above; the correction itself is a committed brief's prose, which has none.

    F21 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
    The implementer report's supervisor-module battery line records the module battery widened by the review module (`.venv/bin/python -m pytest tests/test_phase6_supervisor.py tests/test_review_phase6_supervisor.py --durations=5` — `15 passed in 17.72s`) and never the brief's documented form (`.venv/bin/python -m pytest tests/test_phase6_supervisor.py --durations=5`) with the documented form's own printed line. The wide form is a strict superset (the module's 13 pins plus the review module's 2 probes), green, and every figure it carries is its own printed line — nothing is evaded; but the brief's acceptance names the documented command, and the report's own template requires each claimed figure to name the command that actually ran while this acceptance line names both the documented form it did not run and the wide form it did. This review ran the documented form at FINAL — `13 passed in 12.46s` — so the figures stand; what carries is the report-text deviation from the brief's named battery.
    FIXED: no regression test — a document deviation in a committed report, which only a later step (or the planner's correction of the carried note) absorbs; the documented form's gate line is in this review's VERIFIED.

    F22 (P2, ORIGIN: BRIEF) — the brief's review-block copy of the carried F17 finding is truncated mid-sentence (the literal `(line truncated to 2000 chars)` marker), which rides as a carried text-correction to the planner.

    F23 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The brief's named red at REV is not reproducible — "at REV, with the delivered single-capture _kill, the racing forker's final tail forks plus the double-forked sleeper survive teardown adopted out-of-tree". In a NAMESPACE_V1 drive the workload is the PID-namespace init, _descendants always captures it (descendants=128 namespace_inits=[…]), and its SIGKILL makes the kernel reap the whole namespace (init death kills the namespace; measured: 124 live descendants after a weakened kill, 0 after the init dies), so the delivered _kill already leaves no descendant alive and the pin passes at REV. The sweep lands as the brief's requested defense-in-depth port and the pin as a regression pin; the named red is chimeric for the sandbox, and the load-borne leak F20 measured rode the nested-grading/oracle path, not a standing sandbox-drive race.
    FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the sweep and the pin land as delivered, and the pin's red class is the kill-surface mutant above.

    F26 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
    The repoint instruction names the assertion as "the assertion inverted (the carrier alive, the scoped scan empty)". The probe's old line is `assert carrier.pid in found`; the reading taken is its precise inverse, `assert carrier.pid not in found` (the scoped scan is empty *of the carrier*), with the carrier's liveness asserted beside it. The parenthetical also admits a literal whole-scan `assert not found` reading, which would red on any live workload survivor — the F1/F20 class the brief says stays with the pins — and would couple this host-side probe to unrelated host state, the over-match brittleness F24 is about. Next action is the planner's: a later committed brief's text correction naming the assertion form.
    FIXED: no regression test — an ORIGIN: BRIEF finding is fixed once a later committed brief corrects the text; the repoint lands as delivered (carrier-pid inversion) and is green at FINAL.

    Then each finding marked carried by this step, with its ID:

    F27 | SEVERITY: P2 | ORIGIN: CODE | STATE: carried (the seam's duplicated literal and the buried conjunction member — plan-shape drift silently kills the survivor pins' red class; see FINDINGS above; its next action is a later brief absorbing this review's two probes).

    F28 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's docs-baseline claim falsified — pristine `12f1f83` prints 395; see FINDINGS above; its next action is the planner's correction of the carried note).

    F27 | SEVERITY: P2 | ORIGIN: CODE | STATE: carried
    The scoped scan's surface is a duplicated literal: `_PROBE_ARGV` (`tests/test_batch3_execution.py:184`) sits beside three independent plan `command` tuples (`:49`, `:231`, `:306`) and nothing holds them equal, while the named scoped-scan shape is split across the repointed carrier probe (carrier half) and the F1/F20 pins (survivor half) so no committed member asserts selectivity in one scan. Measured: plan-shape drift leaves every committed pin green while the scan undercounts — C1 (a second argv element in all three plan tuples) `13 passed in 4.04s`, C2 (the interpreter path moved to `/usr/bin/python3.13`) `13 passed in 3.90s`; and the drift silently destroys the pins' red class — C1 stacked with the weakened-capture mutant (that mutant alone reddens F1/F20 on `assert survivors == []`) goes `2 passed in 1.64s` over 123 live workload survivors whose argv carried the drifted shape (enumerated and reaped post-run; the pair-matching set empty throughout), leaking past the `finally` cleanups (`:244`, `:319`) that inherit the same blind scoping. The seam is one rule in form (no exclusion list, rule 13) but is keyed to a copy of the surface rather than to the surface the pins grade; the burial of the conjunction member is what lets the anchor drift stay invisible.
    FIXED: the machines this review commits: `tests/test_review_phase6_probe_scope.py::test_every_probe_plan_command_still_names_the_scanned_pair` (AST-parses the batch3 module and asserts every plan `command` tuple equals the scanned pair; green at HEAD `1 passed`, red on C1 with `At index 1 diff: '-u' != '/work/probe.py'` and on C2 the same class) and `tests/test_review_phase6_probe_scope.py::test_the_scoped_scan_counts_a_live_pair_holder_and_not_an_argv_carrier` (one scan over a live carrier and a live holder built from the plan's own `command`: green at HEAD `2 passed in 0.29s` for the module; red on the REV whole-table scan `assert carrier.pid not in found` — `the scoped scan counted an unrelated argv carrier: [3526004, 3526010, 3526011]` — and red on drift when the plan-built holder stops being counted). Next action is the planner's: a later brief absorbs these members into the pinned module (or pins the anchor its own way) — the drift class itself has no single node id at REV, its red is the mutant battery named above.

    F28 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
    The report's docs-baseline claim is falsified: "a pristine `12f1f83` extraction prints the same `396 Total`, so the brief's `395` baseline is not this step's" — the pristine `12f1f83` worktree prints `395 Total / 180 Unique / 395 OK / 0 Errors` (the brief's EXPECTED WALLS line, correct at drafting), and the 395→396 delta is exactly `docs/research/phase6-probe-scope-brief.md` added in the reviewed range (docs markdown count 160→161 across `git diff 12f1f83 14b6862 --stat -- docs/`, one path). The figure `396 Total … 0 Errors` is the gate's own printed line at FINAL and stands under rule 14 — what fails is the explanatory claim and its dismissal of a baseline that reproduces.
    FIXED: no regression test — a document deviation in a committed report, which only a later step (or the planner's correction of the carried note) absorbs; the measured correction is in this review's VERIFIED.

  Expected dispositions (rule 8 forms): F27 fixed with red evidence (named below — the anchor holds the copies equal or the conjunction member builds the selectivity in one scan, red on the drift class, green at FINAL). F28 stays carried (a document deviation in a committed report, absorbed only by a later step's correction). F11/F12/F18/F19/F20/F24/F25 stay fixed as recorded in the reports that filed them. F13–F17/F21–F23/F26 stay carried with their named owners and text corrections.

  Then each finding marked carried by this step, with its ID:

    F28 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's docs-baseline claim falsified — pristine `12f1f83` prints 395; see FINDINGS above; its next action is the planner's correction of the carried note).

    F29 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the probe-absorption vs anchor reading — both members absorbed into the pinned module; see FINDINGS above; its next action is a later committed brief's text correction).

From `docs/research/phase6-scan-anchor-review.md` — its `INHERITED` block's received F28/F29 dispositions only (the chain above is already carried verbatim through the scan-anchor implementer's INHERITED, and rule 2 governs the planner's copy of this block; the review's received set is the scan-anchor brief's carried F28 plus its F27/F29, F27 fixed at FINAL, F28/F29 carried):

  Expected dispositions of the received set (rule 8 forms), kept: received F27 fixed by this step (the one-source anchor plus both absorbed members, re-driven by this review — red at REV on the drift class, green at FINAL); F13-F23/F26 stay carried; F28 and F29 stay carried (each with its named next action); F11/F12/F18/F19/F20/F24/F25 stay fixed as recorded in the reports that filed them. Added by this review: none.

Expected dispositions (rule 8 forms): F15 fixed with red evidence (named below — the killed run's trees are gone after the kill, red on the delivered `finally`-only reaping, green at FINAL). F28/F29 stay carried (document deviations in committed reports, absorbed only by later steps' corrections). F11/F12/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. F13/F14/F16/F17/F21–F23/F26 stay carried with their named owners and text corrections.

IN SCOPE: src/orchestrator/oracle.py, tests/test_probe_oracle.py
RESIDUALS:

none — rule 3 selection for this brief's `IN SCOPE`: no `docs/residuals.md` trigger glob matches `src/orchestrator/oracle.py` or `tests/test_probe_oracle.py` (the `PILOT-R*` triggers name `planning_pilot.py`, `controller.py`, `reducer.py`, `decomposition.py`, `model.py`, and `derive_obligations.py`; the store is byte-verified against `docs/residuals.md` before this brief's commit, rule 2). No residual row rides, so the residual half of `scripts/check-scope` has nothing to carry here and it prints `tripped: none` — the line your report's `TRIPPED` field copies.

OUT OF SCOPE:
- Editing this brief file (`docs/research/phase6-probe-janitor-brief.md`). A committed brief is never modified; a re-cut is a new file.
- `scripts/check-tests` in full: the gate script is delivered and frozen; the reaping rides as runner code the tests pin, not as a gate-script edit. If the gate script needs a code change, stop and report `IN SCOPE` as insufficient.
- `src/orchestrator/sandbox.py`, `src/orchestrator/sandbox_launcher.py`: the runner and launcher stand; no kill-surface, containment, or teardown change rides here. The sandbox sweep is delivered; this step reaps the harness's own probe trees.
- `scripts/supervise_run.py`, `tests/test_phase6_supervisor.py`, `tests/test_review_phase6_supervisor_boundary.py`: the supervisor chain is delivered; no supervisor, boundary, or leaf change rides here.
- `src/orchestrator/project_oracle.py`, `tests/test_phase3_project_oracle.py`, `tests/test_review_phase6_grading_seed.py`, `tests/test_batch3_execution.py`, `tests/test_review_phase6_sandbox_sweep.py`: the seed and sweep chains are delivered; no grading-gate, deadline, determinism, or scan change rides here.
- `src/orchestrator/experiment_phases.py`, `src/orchestrator/hidden_oracle.py`, `src/orchestrator/conformance.py`: no grading-runner, calibration, probe-shape, or trusted-gate change rides here. No new event type, no `docs/tables/*` change and no table regeneration, no `docs/event-catalogue.md` change.
- `docs/self-host.md`, `docs/residuals.md`, `docs/record.md`, `docs/method.md`, `docs/templates/`: the status edit, if any is owed, rides a later step; this step reaps the probe trees.
- Every committed test module except the IN SCOPE pin module — `tests/test_milestone3_sandbox.py`, `tests/test_axis2_robustness.py`, `tests/test_phase6_supervisor.py`, and every other committed module are read-only here. No committed test file outside IN SCOPE is edited. If a pin would have to change for acceptance to pass, stop and report `IN SCOPE` as insufficient.
- `tests/open_defects.yaml` and the registry's meta-tests: no registration rides (F15 is the open finding whose next brief is this one; no `test_bug_*` name — the open-defect registry would then own it).
- Registry and control files: `tests/counterexamples.yaml`, `tests/guard_mutation_matrix.yaml`, `tests/fixture_fidelity.yaml`, `tests/verifier_registry.yaml`, `tests/spend_paths.yaml`, `tests/invariants.py`, `tests/helpers.py`, `pytest.ini`, tool configuration.
- No new dependencies, no tooling configuration, no reformatting. No new exception class and no `*Executor`/`*ModelClient` class or provider-registry registration.

ACCEPTANCE: Commands (run every one and paste what each prints into `VERIFIED`; expected result is green, exit 0, for each):

    scripts/check-tests --fast
    .venv/bin/python -m pytest tests/test_probe_oracle.py --durations=5
    scripts/check-static
    scripts/check-docs
    scripts/check-scope e2271de <FINAL> docs/research/phase6-probe-janitor-brief.md

Every `pytest` line names `.venv/bin/python -m pytest` deliberately (the interpreter rule: the bare `pytest` on PATH has no `pytest-timeout` and `pytest.ini`'s `strict_config = true` rejects it — run these lines exactly as written and claim them under these names). The probe line rides because this brief's IN SCOPE touches files under `src/` and `tests/` (rule: a brief that adds a file under `tests/` or `scripts/` also names `pytest tests/test_phase3_project_oracle.py`; this brief edits the harness probe runner and its pin module, so the line is the step's own battery — the oracle line is not named because the pin runs no trial and no sandboxed suite). The docs gate is expected green with 0 errors — and every count figure you write in `VERIFIED` is the gate's own printed line, never the REV line quoted in this brief's CONTEXT and never recomputed from memory (rule 14). The scope line is expected green, exit 0, printing `tripped: none` (and the `claimed:` line the parser prints). `REV` below names the step's base — the parent of this brief commit — and is not claimed to be `FINAL`'s parent.

The janitor (`src/orchestrator/oracle.py`, in place — `_run_probe` only): the two `mkdtemp` trees are registered for reaping the moment they are created, and the reaping runs on every path out of the grade — the settled `finally` and the kill path alike. The delivered form is the harness's own pattern, not a new subsystem: a module-level registry keyed by the tree path (or the two paths held in the frame the signal handler can already see) plus a reaper the teardown calls — the `_teardown` survivor verification the sandbox sweep already holds is the shape, not a daemon, not a cron, not a new exception class. The reaper removes only paths the registry holds (`orchestrator-probe-` and `orchestrator-probe-exec-` prefixes under the system temp dir — never an arbitrary path), and a reap failure is silent (`ignore_errors`) the way the delivered `finally` already is. No signature change (`_run_probe(check, tree_path)` stands), no probe-shape change, no deadline change, no digest change. If the reaping cannot be delivered inside `_run_probe` without touching the probe shape or the teardown contract, stop and report `IN SCOPE` as insufficient.

The janitor pin (`tests/test_probe_oracle.py`, in place — the module's next test after the containment matrix): a probe whose `check.probe` raises `KeyboardInterrupt` mid-grade (the kill path the `finally` never sees on a SIGKILLed runner — the closest a test can drive without killing its own runner) leaves no `orchestrator-probe-*` tree behind: the test snapshots the temp dir's probe-tree set before the grade, drives `_run_probe` to the contained `GRADE_INFRASTRUCTURE_ERROR`, and asserts the after-set minus the before-set is empty. The test reuses the module's `_check`/`_run` surface (`_check` with a raising probe — no new fixture, no new builder). The pin needs no sandbox drive and no trial: it grades a tiny tree through the harness path, so it rides the fast suite with no slow mark — but if the pin cannot ride the fast suite without breaking the oracle's inner deterministic suite, keep the slow mark and say so in `CHANGED` (the oracle line is then owed as ACCEPTANCE, and `IN SCOPE` is insufficient as written — stop and report it). The pin carries `@pytest.mark.timeout(120)`: the harness copy plus the raising probe plus the temp-dir walks must fit inside one test's wall budget with margin for a loaded host (expected, with basis, per rule 15 — set from the measured module walls at REV: `15 passed in 2.21s`, slowest `0.52s`).

Red evidence at `REV` `e2271de` (each unit must be replayable; the probe module is green at REV, so the red is written by this step and demonstrated at REV on a scratch copy diff-verified against `git show e2271de:…` before use). `FIXED` names the pin's node id, the command, the named failing assertion, and the delivered file's line numbers or none. Named reds to produce: F15's kill-path red — at REV, with the delivered `finally`-only reaping, the raising probe's grade leaves its `orchestrator-probe-*` tree standing (the named failing assertion is the pin's after-minus-before emptiness assertion — the 81-tree class the grading-seed review measured; an error or a hang is not red, the assertion must be the thing that failed).

Named shapes to build — the list of shapes, not the number of tests. One shape needs at least one test, and every named shape needs one:

- `Probe-janitor shape` (F15): a killed probe run's trees are gone after the grade — the janitor pin, red at REV on the `finally`-only reaping, green at FINAL.

EXPECTED WALLS: the measured baseline at this brief's drafting: `scripts/check-tests --fast` — `983 passed, 109 deselected in 97.58s (0:01:37)`; `scripts/check-docs` — `399 Total / 180 Unique / 399 OK / 0 Errors` (this session's HEAD line); `scripts/check-static` — exit 0, gated `All checks passed!`, advisory `Found 104 errors.` unchanged; `.venv/bin/python -m pytest tests/test_probe_oracle.py --durations=5` — `15 passed in 2.21s` (this session's REV measurement: slowest `0.52s` deadline pin, `0.50s` containment pin, `0.28s` flooding pin). The janitor changes no grade wall — registration plus reaping is two temp-dir walks — so the pin's `120`s timeout carries margin per rule 15.

REV: e2271de
CLOSE OUT: no
DIVERGENCES: none
REVIEW: one session — the reaping above all: does the kill-path grade leave no probe tree (the pin's emptiness assertion red at REV on the `finally`-only reaping, green at FINAL), is the seam one registry the teardown calls rather than a caller contract (rule 13), is the `IN SCOPE` line plain with scope green, are the docs figures the gate's own printed lines (rule 14), and does the step touch only the IN SCOPE files with every committed battery green and no committed test edit outside them. Whether anything remains open after this step — the brief-text corrections — is the loop's exit input: report findings and let the planner decide (cycle step 5).

You implement one brief. You do not plan, re-scope, or decide what the work is. Read every file
in IN SCOPE before editing it; if what you find contradicts the brief, report that instead of
guessing. Modify only what IN SCOPE names. If you find uncommitted changes in the tree, stop and say
so. A new module under tests/ that runs trials or sandboxed suites carries
pytestmark = pytest.mark.slow at module level: the oracle's internal gate runs the fast suite
(-m "not slow") and an unmarked trial-running module breaks it. Stop and report rather than
improvise when IN SCOPE is
insufficient, the brief contradicts the code, you have already failed twice at this goal, or the
requirement is ambiguous and no single reading is consistent with the rest of the brief. If exactly
one reading is consistent, take it and file a FINDING with ORIGIN: BRIEF that names the reading you
took. No new dependencies, no tooling configuration, no reformatting. Run every ACCEPTANCE command;
fix what is in scope and report what is not.

A defect in the brief is a FINDING, never METHOD FRICTION. METHOD FRICTION is only for the method and
its templates: any place you deviated from them, found them ambiguous, or found them silent. Report
friction you worked around — that is the section the method learns from.

Commit your own work. When ACCEPTANCE passes, commit the files you changed, and only those, in one
commit whose message ends with the trailer `Actor: IMPLEMENTER`. That commit's full sha is FINAL.
Then append docs/templates/report.txt verbatim, fill it in, write it to
docs/research/<phase>-<subject>-implementer.md, and commit it in a second commit with the same
trailer. Do not amend the first commit to add the report: FINAL must not change after the report
names it. If you cannot commit, stop and say so; do not leave work uncommitted for someone else to
commit.
--- END BRIEF UNDER REVIEW ---
You did not write this code and you will not fix it; a finding is output, not an action. The step's decision is the planner's (cycle step 5): report findings, never decisions. The commits under review carry the trailer Actor: IMPLEMENTER; if your session wrote any of them, stop and say so. Run REVIEW ACCEPTANCE first: it is the method's only gated check. The ACCEPTANCE lines inside the brief under review are the implementer's; re-run them to verify the delivery, and run scripts/check-tests as your standing battery — the full suite is your falsification ground — recording both in VERIFIED. The embedded brief is a convenience copy: if the standalone file it names differs, file a FINDING with ORIGIN: BRIEF and review against the committed file.
You may create new test files to prove a finding, and modify nothing else. If a failing test would need registering in tests/open_defects.yaml, say so in the report and leave the registration to the next implementer brief rather than editing that file; the registry owns test_bug_* pins only, and a test_review_* probe rides red until absorbed. Read pre-images with git show <rev>:..., not the implementation's account of them. Run every mutant on a scratch copy of the tree, and diff each mutant against git show <rev>:... before trusting its result. Attempt to falsify the brief, not only the code: a finding that the brief was wrong is a finding, and so is a defect the implementer filed only as friction. Answer every item under QUESTIONS TO ATTACK. Answers live in the form's fields — a finding in FINDINGS, a measurement in VERIFIED/SHAPES. An added review's QUESTIONS TO ATTACK come only from the implementer's report — never from another review's findings. Enumerate the shapes you probed that the existing suite does not build, as a list; zero with a stated reason is legitimate. Report each guarantee as falsified-with-reproducer or attacked-and-not-falsified with the method.
Commit your own work: your new test files and your report, together, in one commit whose message ends with the trailer Actor: REVIEWER. Append docs/templates/report.txt verbatim, fill it in, and write it to docs/research/<phase>-<subject>-review.md. If you cannot commit, stop and say so.
