REVIEW BRIEF — phase6-exit-gate-runs (Phase 6: the seed-matrix campaign settles under the cap — R15)
REV: c418342
FINAL: 36e42a4dbf350586567fab37e620ab4d025c2c57
REVIEW ACCEPTANCE: scripts/check-scope c418342 36e42a4dbf350586567fab37e620ab4d025c2c57 docs/research/phase6-exit-gate-runs-brief.md
FOCUS: the campaign evidence and its weakest seams — the ledger paste against the on-disk ledger at `/tmp/kilo/matrix/experiment-spend.json` and `--inspect` (committed 0.8020469000000002, 16 trials, in_flight {}), the seven `budget_exhaustion=True` trials that still settled (per-trial flag vs the run's settled verdict), the credential-bytes scan's method against the stores that exist (the debug transcript is absent by contract), and the report's claim that no project file moved (`git diff c418342 67f5531408f8c9d57de6f38630e1f6ba6d410ecc -- scripts/smoke_run.py` empty, range diff limited to the brief plus this report).
SEVERITY SCALE: P0 blocks the step; P1 is serious; P2 is minor.
QUESTIONS TO ATTACK:
  F1 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
  The brief's CONTEXT and RESIDUALS make the report commit the step's content commit and say `FINAL` names it, but the template also requires the report to paste the `check-scope` line, whose range carries `FINAL`'s sha — a sha that cannot be known before the commit that carries the report exists, and cannot be written into that commit's own tree (a fixed point). A one-commit realization is impossible.
  FIXED: no regression test — a committed brief's prose has none. Reading taken: the commit that first carries the report is `FINAL`; the report's two self-referential fields (`FINAL`'s value in the header and the `check-scope` VERIFIED line's range) are finalized in the immediately following commit, so `scripts/check-scope c418342 <FINAL> docs/research/phase6-exit-gate-runs-brief.md` reproduces the report's pasted line exactly against `FINAL`. Next action is the planner's: a later committed brief's text correction (rule 8's form) — this step delivers the reading and cannot correct the brief itself.
  F2 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
  The brief's step-3 store list names "the debug transcript" among the stores to scan, and `docs/record.md`'s campaign paragraph says the scan includes it, but the contract's "export exactly these" block omits `ORCHESTRATOR_DEBUG_TRANSCRIPT`, so the transcript (opt-in, default off; `resolve_transcript_path` returns `""` and nothing is written) does not exist for this run.
  FIXED: no regression test — a committed brief's text has none. Reading taken: the contract's exact export list governs, and the scan covers every store present (the transcript's absence is the contract's own omission, not a scan gap — the secret is proven absent from every file that exists). Next action is the planner's: align step 3's store list with the contract (or add the export if the transcript is wanted scanned).

  1. The ledger above all — the delivery's one question. Drive it live against the FINAL work commit 67f5531408f8c9d57de6f38630e1f6ba6d410ecc on scratch diff-verified against git show c418342:scripts/smoke_run.py before use: (a) the `--json` ledger status the report pastes (`committed = 0.8020469000000002`, `reserved = 0`, `available = 9.1979531`, `in_flight = {}` under `limit = 10.0`, exit `code: 0`) matches the on-disk ledger at `/tmp/kilo/matrix/experiment-spend.json` and the `--inspect` trial lines — re-run `--inspect` against the workdir and name any mismatch; (b) the seven `budget_exhaustion=True` trials (RECURSIVE_LAZY 1001/1002/1004, RECURSIVE_CONCURRENT 1001/1002/1004) still settled with a concrete `actual` — which ledger field carries each trial's settled actual, and does any trial lack one? (c) the scan's 2326-file, 0-hit claim covers every store present — which stores exist under the workdir, and is any store the scanner could have missed? Report each guarantee as falsified-with-reproducer or attacked-and-not-falsified with the method.

  2. The report's forms and counts (J-class). FINAL names the report commit 36e42a4dbf350586567fab37e620ab4d025c2c57, never the work commit 67f5531408f8c9d57de6f38630e1f6ba6d410ecc — verify rather than taking the report's word. Re-run the brief's ACCEPTANCE lines that need no provider key (fast suite 984 passed, 110 deselected; 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 c418342 67f5531408f8c9d57de6f38630e1f6ba6d410ecc --stat touches no project file (the report commit is the step's only content commit — verify). SHAPES' one named shape (Sustained-settle-under-cap) against rule 7: no committed member asserts it (a suite member cannot spend the owner's USD) — adjudicate whether the report's VERIFIED lines carry it whole (sixteen settled trials, ledger under the cap, scan clean). INHERITED's received set: F16 fixed stays fixed, F13/F14/F17/F21/F22/F23/F26/F28/F29/F30/F31 stay carried with their named owners — name any drift. 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-exit-gate-runs-brief.md; the committed file is authoritative) ---
IMPLEMENTER BRIEF — phase6-exit-gate-runs (Phase 6: the seed-matrix campaign settles under the cap — R15)
GOAL: One invocation of the tracked runner drives all four arms across seeds 1001–1004 in real mode (16 trials), every trial settles, the spend ledger's committed total stays under the 10 USD cap, no credential bytes reach any store, and the evidence lands in a committed report the phase's exit gate draws from.
CONTEXT: Phase 6 stands delivered except its exit gate ([self-host.md](../self-host.md#phase-6-operational-envelope): sustained real-provider runs complete with every trial settled, no credential bytes in any store, and the cap enforced before spend). The Record's last row is the deadline-margin report row (REV `a1d4e07`); every committed brief through it has its report, R14 closed the chain's text defects as actionless, and R15 (amended after the referral `d1997b0`, re-ruled `c418342`, pricing confirmed this session) scopes this campaign and names this brief. `REV` below names the step's base — the parent of this brief commit — and is not claimed to be `FINAL`'s parent. The report the step writes is the evidence artifact; it records per-trial classifications, the ledger status, and the scan, and (per its template) it is committed separately after the work commit — but here the campaign writes no code and the work commit is empty of project files, so the report commit is the step's only content commit and `FINAL` names it.
INHERITED:
From `docs/research/phase6-probe-janitor-implementer.md`, its `INHERITED` block verbatim:

  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.

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

    F30 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the named red is chimeric as literally written and the GOAL's SIGKILL clause is not reachable inside `_run_probe` — the pin models the probe's sandbox root sibling, and the SIGKILL residue needs a cross-process janitor; see FINDINGS above; its next action is a later committed brief's text correction and the cross-process janitor F15's own next action names).

From `docs/research/phase6-probe-janitor-review.md` — its `INHERITED` block's received F30/F31 dispositions only (the chain above is already carried verbatim through the probe-janitor implementer's INHERITED, and rule 2 governs the planner's copy of this block; the review's received set is the janitor brief's carried chain plus its F15/F30, F15 fixed at FINAL, F30 carried):

  Expected dispositions of the received set (rule 8 forms), kept: received F15 fixed by this step (the delivered registry plus finally-called reaper over the two trees and the sandbox root; the red the step names re-driven by this review on the REV scratch — `1 failed, 15 passed in 2.33s`, standing tree `orchestrator-probe-exec-…-root`, green at FINAL `16 passed in 2.03s`); F13/F14/F16/F17/F21–F23/F26/F28/F29 stay carried with their named owners and text corrections; F11/F12/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. Added by this review: F31, carried.

  F31 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
  The implementer report's SHAPES field claims "none" while its own F30 block records a probe the suite does not build: "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" — a probed shape (re-driven and confirmed by this review, `1 passed in 0.26s` on the REV scratch) with no committed member and no SHAPES entry, against the template's "SHAPES: none claims nothing probed is left unbuilt — never that unbuilt probes were left unnamed." Substantively harmless: the delivered pin's guard subsumes the merely-raising shape's red class — dropping the reap call or either registration line re-reds the delivered pin (mutants M1/M2 above), so the merely-raising shape catches nothing the committed pin does not.
  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 shape's drive is named in this review's VERIFIED and SHAPES.

Expected dispositions (rule 8 forms): F16 fixed with red evidence (named below — the near-deadline grade is FAIL idle and FAIL under load, red on the 60.0s fixture deadline, green at FINAL). F28/F29/F30/F31 stay carried (document deviations in committed reports, absorbed only by later steps' corrections). F11/F12/F15/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. F13/F14/F17/F21–F23/F26 stay carried with their named owners and text corrections.

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

  F13 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the pin's stated candidate is not the one that flips; its next action is a later committed brief's text correction).

  F14 (P2, ORIGIN: DOCUMENT, carried) — the F8 red evidence is load-borne and the implementer report does not say so; its next action is the planner's correction of the carried note.

  F17 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the boundary-fix instruction's internal inconsistency; its next action is a later committed brief's text correction).

  F21 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's battery line deviates from the brief's named command; its next action is a later step's or the planner's correction).

  F22 (P2, ORIGIN: BRIEF) — the truncated review-block copy of F17; a carried text-correction to the planner.

  F23 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the named red at REV is not reproducible; its next action is a later committed brief's text correction).

  F26 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the repoint's assertion form; its next action is a later committed brief's text correction).

  F28 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's docs-baseline claim falsified; 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; its next action is a later committed brief's text correction).

  F30 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the named red is chimeric as literally written; its next action is a later committed brief's text correction and the cross-process janitor F15's own next action names).

  F31 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's SHAPES field claims "none" while its own F30 block records a probed shape; its next action is a later step's or the planner's correction).

Received-set dispositions kept: F16 fixed by this step (above); F11/F12/F15/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them; F13/F14/F17/F21/F22/F23/F26/F28/F29/F30/F31 stay carried with their named owners and text corrections.

  From `docs/research/phase6-deadline-margin-review.md`, its `INHERITED` block verbatim:

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

    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.

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

      F30 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the named red is chimeric as literally written and the GOAL's SIGKILL clause is not reachable inside `_run_probe` — the pin models the probe's sandbox root sibling, and the SIGKILL residue needs a cross-process janitor; see FINDINGS above; its next action is a later committed brief's text correction and the cross-process janitor F15's own next action names).

  From `docs/research/phase6-probe-janitor-review.md` — its `INHERITED` block's received F30/F31 dispositions only (the chain above is already carried verbatim through the probe-janitor implementer's INHERITED, and rule 2 governs the planner's copy of this block; the review's received set is the janitor brief's carried chain plus its F15/F30, F15 fixed at FINAL, F30 carried):

    Expected dispositions of the received set (rule 8 forms), kept: received F15 fixed by this step (the delivered registry plus finally-called reaper over the two trees and the sandbox root; the red the step names re-driven by this review on the REV scratch — `1 failed, 15 passed in 2.33s`, standing tree `orchestrator-probe-exec-…-root`, green at FINAL `16 passed in 2.03s`); F13/F14/F16/F17/F21–F23/F26/F28/F29 stay carried with their named owners and text corrections; F11/F12/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. Added by this review: F31, carried.

    F31 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
    The implementer report's SHAPES field claims "none" while its own F30 block records a probe the suite does not build: "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" — a probed shape (re-driven and confirmed by this review, `1 passed in 0.26s` on the REV scratch) with no committed member and no SHAPES entry, against the template's "SHAPES: none claims nothing probed is left unbuilt — never that unbuilt probes were left unnamed." Substantively harmless: the delivered pin's guard subsumes the merely-raising shape's red class — dropping the reap call or either registration line re-reds the delivered pin (mutants M1/M2 above), so the merely-raising shape catches nothing the committed pin does not.
    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 shape's drive is named in this review's VERIFIED and SHAPES.

  Expected dispositions (rule 8 forms): F16 fixed with red evidence (named below — the near-deadline grade is FAIL idle and FAIL under load, red on the 60.0s fixture deadline, green at FINAL). F28/F29/F30/F31 stay carried (document deviations in committed reports, absorbed only by later steps' corrections). F11/F12/F15/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. F13/F14/F17/F21–F23/F26 stay carried with their named owners and text corrections.

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

    F32 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried (the red clause's load regime; see FINDINGS above; its next action is a later committed brief's text correction, absorbed by F14's pending load-condition carry).

    F33 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried (the report's docs-line file-count narrative; see FINDINGS above; its next action is the planner's correction of the carried note).

  Received-set dispositions kept: F16 fixed by this step (the delivered margin re-driven by this review: the fixture's own deadline is the seam — the `:215`/`:216` `60.0` literal removed, the delivered `300.0` default driving the manifest, the corpus/seed/N untouched — and the determinism-under-load pin red at REV on its FAIL-verdict assertion under the loaded regime, green at FINAL); F13/F14/F17/F21/F22/F23/F26/F28/F29/F30/F31 stay carried with their named owners and text corrections; F11/F12/F15/F18/F19/F20/F24/F25/F27 stay fixed as recorded in the reports that filed them. Added by this review: F32, F33, carried.
IN SCOPE: scripts/smoke_run.py
RESIDUALS:

none — rule 3 selection: no `docs/residuals.md` trigger glob matches `scripts/smoke_run.py` (the accepted triggers name `planning_pilot.py`, `controller.py`, `reducer.py`, `decomposition.py`, `model.py`, and `derive_obligations.py`; byte-verified against `docs/residuals.md` at this brief's commit). The campaign writes no code and the work commit is expected to touch no project file; the step's content commit is the report file, which `check-scope` admits by its added-file exemption. `tripped: none` is the expected line the report's `TRIPPED` field copies.

OUT OF SCOPE:
- Editing this brief file (`docs/research/phase6-exit-gate-runs-brief.md`). A committed brief is never modified; a re-cut is a new file.
- `run.sh`: untracked and gitignored (`.gitignore:1`); no base commit pins it and rule 1 cannot make it an input. The contract below is pinned by environment, exported explicitly; the cited command is `.venv/bin/python scripts/smoke_run.py` at the pinned base. Do not read, source, or edit `run.sh`.
- Every committed file except `scripts/smoke_run.py` and the report: `src/orchestrator/*`, every other `scripts/*` file, `tests/*`, `docs/tables/*`, `docs/event-catalogue.md`, `docs/self-host.md`, `docs/residuals.md`, `docs/record.md`, `docs/method.md`, `docs/templates/`. No event, table, gate-script, oracle, or registry change rides here. If any is needed for acceptance to pass, stop and report `IN SCOPE` as insufficient.
- The credential: it enters the process environment from the session that runs the campaign (`ORCHESTRATOR_PROVIDER_API_KEY`), never a file in the tree, never the brief or the report. If the key is absent from the process environment, stop and report — do not source it from anywhere else.
- The phase's other work: this step drives the campaign and records its evidence; it does not close the phase (that is the loop's exit input at cycle step 5).

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

    .venv/bin/python scripts/smoke_run.py --probe
    .venv/bin/python scripts/smoke_run.py --json
    .venv/bin/python scripts/smoke_run.py --inspect
    scripts/check-tests --fast
    scripts/check-static
    scripts/check-docs
    scripts/check-scope <REV> <FINAL> docs/research/phase6-exit-gate-runs-brief.md

The contract (export exactly these; the values bind the run — R15 as confirmed this session):

    export ORCHESTRATOR_MODE=real
    export ORCHESTRATOR_CONFIRM_REAL=yes
    export ORCHESTRATOR_OWNER="<the authorising owner — the implementer session's owner identity>"
    export ORCHESTRATOR_CAP=10
    export ORCHESTRATOR_MODEL_COST=0.5
    export ORCHESTRATOR_PER_ATTEMPT_COMMITMENT=0.1
    export ORCHESTRATOR_SEEDS=1001,1002,1003,1004
    export ORCHESTRATOR_PROVIDER_BASE_URL="https://api.kilo.ai/api/gateway"
    export ORCHESTRATOR_PROVIDER_PATH=/chat/completions
    export ORCHESTRATOR_PROVIDER_MODEL=kilo-auto/free
    export ORCHESTRATOR_PROVIDER_MODEL_VERSION=1
    export ORCHESTRATOR_PRICE_INPUT_PER_MTOK=0.3
    export ORCHESTRATOR_PRICE_OUTPUT_PER_MTOK=1.2
    export ORCHESTRATOR_PROVIDER_UNITS=USD
    export ORCHESTRATOR_MAX_INPUT_TOKENS=4000
    export ORCHESTRATOR_MAX_OUTPUT_TOKENS=1024
    export ORCHESTRATOR_RUN_MAX_INPUT_TOKENS=65536
    export ORCHESTRATOR_RUN_MAX_OUTPUT_TOKENS=16384
    export ORCHESTRATOR_MAX_MODEL_CALLS=8
    export ORCHESTRATOR_CREDENTIAL_TTL_SECONDS=600
    export ORCHESTRATOR_WORKDIR=/tmp/kilo/matrix
    unset ORCHESTRATOR_SEED

`ORCHESTRATOR_PROVIDER_API_KEY` enters from the process environment (present in the session that runs the campaign); `ORCHESTRATOR_REVOKE_URL` is exported too if the provider publishes one, otherwise left unset and the report says so. `ORCHESTRATOR_SEED` (singular) is unset so the four-seed matrix is unambiguous (`parse_seeds`, `scripts/smoke_run.py:596`: the plural wins when set, but an accidental singular export beside it is a mis-set matrix — unset it). `ORCHESTRATOR_REQUIRE_CONFORMANCE` keeps its default `1`: the conformance preflight runs. `ORCHESTRATOR_AUTHORIZED_AT`/`ORCHESTRATOR_AUTH_METHOD`/`ORCHESTRATOR_APPROVAL_REF` default from the session unless the authorising owner directs otherwise — the report records the values the ledger bound. `ORCHESTRATOR_WORKDIR` is fresh per authorisation (`/tmp/kilo/matrix` is absent at dispatch — verified this session: `ls: cannot access '/tmp/kilo/matrix'`; an existing ledger binds the directory to its recorded authorisation and trials are never overwritten, so if the directory exists at dispatch, stop and report rather than reuse it). The base commit is `REV` below: clone it fresh (or `git archive REV`) — the run executes the tracked `scripts/smoke_run.py` at `REV`, never the live working tree.

Why these numbers (each measured this session at REV `c418342`, basis named — expected, per rule 15):

- The worksheet passes: `cap_worksheet` (`scripts/smoke_run.py:734`) prints `required_cap 9.6000 / authorized_cap 10.0000 / margin 0.4000` over `trials 16.0` (4 arms × 4 seeds). The first ACCEPTANCE command sequence runs the worksheet before any trial — a negative margin refuses the run (`scripts/smoke_run.py:749`), which is the gate that caught the previous figures.
- The declared worst case fits the commitment: at the frozen `0.3/1.2` USD accounting prices, one call's declared ceiling (`4000` in / `1024` out) prices at `0.0024288`, under the `0.1` per-attempt commitment (`scripts/smoke_run.py:694`) — expected PASS. The prices are USD accounting rates (confirmed this session although the tier bills no USD — expected, per rule 15, basis: the runner's frozen-price gate refuses zero prices in real mode at `scripts/smoke_run.py:689`, and the gateway catalog prices the `kilo-auto/free` tier at zero in every dimension, fetched this session); the report records both the accounting rates bound and the gateway-published zero at dispatch. `ORCHESTRATOR_MODEL_COST=0.5` is the per-trial ledger reservation (`trial_driver.py:807`); `ORCHESTRATOR_PER_ATTEMPT_COMMITMENT=0.1` is the per-call ceiling the controller enforces before each billable dispatch and the model profile's `cost_per_call` in real mode (`scripts/smoke_run.py:1081`, `:1322`).
- `MAX_MODEL_CALLS=8` carries margin over the measured scripted attempt shape (expected, per rule 15 — basis: the scripted shape probe at REV, one invocation with all four seeds at generous budgets, exit `code: 0`, per-trial model attempts DIRECT 1, FIXED_ONE_LEVEL 1–3, RECURSIVE_LAZY 1–4, RECURSIVE_CONCURRENT 2–4). Real-model attempts fail and retry where scripted attempts pass, so the ceiling doubles the measured max rather than matching it; the controller refuses past the ceiling (`controller.py` model-attempt cap) instead of spending unbounded.
- One invocation, not four: the manifest covers every arm and `arm_assignment` pairs each arm with the full seed tuple (`scripts/smoke_run.py:1071`), so one invocation runs the 16-trial matrix under one ledger and one authorisation — which is what "cumulative spend under the cap" means here. The reservation is per trial (`0.5` × 16 = `8.0` against the `10` cap) and settlement releases each reservation for its actual; the report records the ledger's `committed`, `reserved`, and `available` at the end.

The campaign (in order; stop-and-report, never improvise, on any refusal — a refusal is evidence, not a prompt to re-tune):

1. `--probe`: exactly one model call against `kilo-auto/free` through the real client (bounded 429 retry inside the one per-call deadline, `ORCHESTRATOR_MAX_RATE_LIMIT_RETRIES=2` default). Expected: the probe settles and its cost report prices in USD. If the probe fails (auth, routing, model id), stop and report — do not change the model id, the prices, or the budgets.
2. `--json` (the full matrix): the 16-trial run under the cap. Expected: exit `code: 0`, every trial settled (no `BudgetExhausted`, no `SpendCapExceeded`, no ledger-integrity refusal), `committed` under the `10` cap. If the cap refuses mid-run, that refusal is the finding — record the ledger status and stop; do not raise the cap.
3. The credential-bytes scan: after the last trial, with the key's bytes in hand only in the session (never written to the tree), scan every store under the workdir — the trial `events.db` files, the ledger, and the debug transcript — for those bytes, and record the method and the result. The client redacts secrets from stored observations (`redacted:sha256:` digests) and the transcript is redacted and `0600`; the scan proves it for this run's stores. Precedent for the method: the axis-2 robustness pin asserting `SECRET.encode() not in blob` over the store bytes (`tests/test_axis2_robustness.py:228`, `:451`).
4. `--inspect`: the recorded ledger and trial state; the report pastes its printed lines.
5. The three standing gates plus scope: `scripts/check-tests --fast`, `scripts/check-static`, `scripts/check-docs` (all expected green, exit 0; every count figure in `VERIFIED` is the gate's own printed line — rule 14), and `scripts/check-scope <REV> <FINAL> docs/research/phase6-exit-gate-runs-brief.md` (expected green, exit 0, `tripped: none`).

The report (the step's evidence artifact — this is what the phase's exit gate draws from): per-trial arm/seed, classification, controller outcome, cost, and budget-exhaustion flag (the runner's own results lines); the ledger's final `committed`/`reserved`/`available` against the `10` cap; the gateway-published `kilo-auto/free` rates observed at dispatch beside the USD accounting rates bound; the credential-TTL value bound and whether `ORCHESTRATOR_REVOKE_URL` was set (if unset, the report says the provider publishes none and revocation is TTL-expiry); the scan's method (which stores, what byte string, what tool) and its result; the base commit cloned and the manifest hash; and every ACCEPTANCE command's printed lines in `VERIFIED`. No code change is expected: if the campaign passes without touching `scripts/smoke_run.py`, the work commit is empty of project files and `FINAL` names the report commit — say so in the report. If any gate, refusal, or verdict contradicts this brief, that contradiction is a FINDING (with `ORIGIN: BRIEF` where the brief mis-stated the runner), never a silent re-tune: record the exact refusal line and stop.

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:

- `Sustained-settle-under-cap shape` (R15): sixteen real-provider trials settle back-to-back under one authorisation with cumulative committed spend under the cap — the campaign itself, red at REV in the sense that no such evidence exists at REV (the runner's worksheet, ledger, and scan machinery stand delivered but undriven against the real provider), green at FINAL with the ledger status and the per-trial lines above. No committed test member asserts it (a suite member cannot spend the owner's USD); the evidence is the report's pasted ledger and trial lines, and the shape is recorded here, not filed as a node id.

EXPECTED WALLS: the measured baseline at this brief's drafting (HEAD `c418342`): `scripts/check-tests --fast` — `984 passed, 110 deselected in 85.94s (0:01:25)`; `scripts/check-docs` — `404 Total / 181 Unique / 404 OK / 0 Errors`; `scripts/check-static` — exit 0, gated `All checks passed!`, advisory lines unchanged. The real-provider campaign walls are not yet measured — the first VERIFIED lines this step writes set the budgets (expected, with basis, per rule 15 — basis: the scripted 16-trial matrix at REV exits `code: 0` in under ten minutes of runner wall; the real campaign adds provider latency, the bounded 429 backoff, and sixteen sandbox grading runs, so a single generous step timeout with the runner's own per-call `60`s deadline bounding each dispatch covers it; the report records the campaign's own wall).

REV: c418342
CLOSE OUT: no
DIVERGENCES: none — R15's "four consecutive real runs" reads as the four-seed matrix in one invocation (confirmed this session: one invocation with `ORCHESTRATOR_SEEDS=1001,1002,1003,1004`); "sustained ... back-to-back under one authorisation" is the single ledger the invocation opens. The USD accounting prices for the free tier are the confirmed owner direction recorded above, not a divergence: the runner refuses zero prices in real mode.
REVIEW: one session — the campaign and its evidence above all: did all 16 trials settle with committed spend under the 10 cap on one ledger, does the report's ledger paste match `--inspect`, was every store (trial DBs, ledger, transcript) scanned for credential bytes with the method named, are the gateway-published zero rates recorded beside the accounting rates bound, is the scope line green with `tripped: none` and the diff limited to the brief plus the report, are the docs figures the gates' own printed lines (rule 14), and is any refusal recorded verbatim rather than re-tuned. Whether the phase's exit gate is met after this step is the loop's exit input: report findings and let the planner decide (cycle step 5).
--- 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/phase6-exit-gate-runs-review.md. If you cannot commit, stop and say so.
