REPORT — phase6-closeout-status-review
STEP: phase6-closeout-status (Phase 6: the exit-gate evidence lands in the phase Status)
REV: 18d9f89   FINAL: 18e7122736daeec4157b03d167dc2c57f3269061
CHANGED: docs/research/phase6-closeout-status-review.md — this review report; no test file was added to the tree: every figure the Status names is re-derived from the committed reports' printed lines and the at-rest campaign stores (the ledger parse and the file census are session-side reads, not tree code, and the one module whose pin this review drove lives in the suite already), so the guarantees the review attacked were re-derived live in VERIFIED instead of riding new nodes.
VERIFIED: REVIEW ACCEPTANCE, run first — `scripts/check-scope 18d9f89 af08b8d89824b97664496e34dd5365b019399cb9 docs/research/phase6-closeout-status-brief.md` — exit 0, prints `claimed: docs/self-host.md`, `tripped: none`, `scope: ok (3 path(s) in 18d9f89..af08b8d89824b97664496e34dd5365b019399cb9)` — byte-identical to the final report's own pasted check-scope line (the @@FINAL@@ placeholder finalized at `18e7122`), and the three paths are exactly the brief (`docs/research/phase6-closeout-status-brief.md`, added at `2e2aad3`) plus the Status update plus the report, with `docs/record.md` untouched across the range (`git diff 18d9f89 af08b8d8… --stat`: three files, none named record.md).
VERIFIED: the embedded copy of the brief under review is the committed brief — the extraction between the BEGIN/END markers diffed against `docs/research/phase6-closeout-status-brief.md` at HEAD `18e7122` is empty (`diff` exit 0), so the brief's committed file is authoritative and matches.
VERIFIED: the `docs/self-host.md` diff at `git diff 18d9f89 af08b8d89824b97664496e34dd5365b019399cb9 -- docs/self-host.md` is the one Status paragraph: the stale "Status: partially done —" / "The phase's remaining work is the exit gate itself: … The folded Axis 3 runs …" lines are replaced by "Status: delivered for the exit gate — … The exit gate draws on the 16-trial real-provider campaign ([implementer] with the target `research/phase6-exit-gate-runs-implementer.md`, [review] with the target `research/phase6-exit-gate-runs-review.md`; Record row `18d9f89`): sixteen trials settled across the four arms and seeds 1001–1004 under one authorisation, ledger committed `0.8020469000000002` under the `10` cap, credential-bytes scan 0 hits over the 2326 stores present, scope `tripped: none`." (The two links are quoted with their targets backticked rather than re-emitted, because a relative link pasted from `docs/`-rooted self-host.md resolves wrongly from this report's own `docs/research/` directory.) The `Exit gate:` definition line and the `<!-- informative -->` line are content-identical pre/post (a `grep -n` over both revisions shows the only paired differences are line-number shifts from the paragraph's +2 lines); the supervisor-links sentence survives unchanged inside the paragraph.
VERIFIED: the focus figures, every one pasted against the committed exit-gate-runs reports' own lines and re-derived from the at-rest stores. The implementer report's `[15] spend ledger status` prints `committed = 0.8020469000000002` under `limit = 10.0` with `in_flight = {}` (line 25) and the review report re-derives the same from the on-disk ledger at sha256 `68d23325754f827b24d3fcfc6e054a154898dc704aedd4bc9eadb6c5a1588d76` (its VERIFIED focus 1(a)); this review re-parsed `/tmp/kilo/matrix/experiment-spend.json` at rest and its sha256 is the same `68d23325…8d76` (the workdir was left at rest), parsing to authorization `{limit: 10.0, owner: 'OWNER', method: 'owner-decision', authorized_at: 2026-10-14T23:40:46Z}` (ONE authorization record — the prose's "under one authorisation"), `committed: 0.8020469000000002`, `in_flight: {}`, and sixteen trial documents named `EXP-<ARM>-<seed>` over DIRECT, FIXED_ONE_LEVEL, RECURSIVE_LAZY, RECURSIVE_CONCURRENT and seeds 1001–1004 with every trial carrying a concrete numeric `actual` whose sum is `0.8020469` (`repr` `0.8020469`, the ledger's own `committed` float prints `0.8020469000000002`) — the prose's "sixteen trials settled across the four arms and seeds 1001–1004" and "committed `0.8020469000000002` under the `10` cap" hold on every surface. The workdir census re-counts exactly 2326 regular files with zero symlinks and zero special entries (`find /tmp/kilo/matrix -type f | wc -l` = 2326; the symlink/block/char/fifo/socket sweep = 0) — the reports' `files scanned: 2326` / `raw-credential-byte hits: 0` line's file count reproduces (the raw-credential re-scan itself needs the campaign session's env key, not re-driven — no spend, per OUT OF SCOPE; the figure stands on the reports' own printed lines plus the prior review's independent re-scan). `tripped: none` appears on both committed reports' check-scope VERIFIED lines and both of this step's own check-scope runs.
VERIFIED: the prose "Record row `18d9f89`" — `git show 18d9f89:docs/record.md` contains the exit-gate-runs report-row at line 318: `` `docs/research/phase6-exit-gate-runs-implementer.md`, `docs/research/phase6-exit-gate-runs-review.md` | c41834285ba0c54f98356d9b108347868759ba25e | 0 | 0 | 0 | 0 | 1 ``, the closeout commit's own subject "Close out phase6-exit-gate-runs: campaign delivered, R15 consumed, F1-F4 actionless" is `18d9f89` itself, so the row the Status points at (the Record at `18d9f89`) exists — but see FINDINGS F3 below: the row's REV cell is not a commit object.
VERIFIED: `scripts/check-tests --fast` (the brief's ACCEPTANCE form) — `984 passed, 110 deselected in 114.40s (0:01:54)`, exit 0 — the same counts as the report's own line (`984 passed, 110 deselected in 107.38s (0:01:47)`) and the brief's EXPECTED WALLS (`984 passed, 110 deselected in 92.25s`); the wall is this host's, a baseline not a bound.
VERIFIED: `scripts/check-tests --fast --parallel` (the commit hook's other form, the parallel-form red the report classifies) — driven twice by this review, BOTH re-drives reddened the same member with the same class as the report's named red: `tests/test_probe_oracle.py::test_killed_probe_grade_reaps_every_intermediate_tree` failed at its `:170` whole-scan assertion (`1 failed, 949 passed in 22.96s`, then `1 failed, 949 passed in 22.50s`, the serial phase after each green `34 passed in 15.20s` / `34 passed in 14.37s`), the extras being peer-created live trees caught in the pin's scan window (`orchestrator-probe-_fv7jdpt`, then `orchestrator-probe-t3v5qi6i` + `orchestrator-probe-exec-1e264ced`) — versus the report's one red then three named green re-drives. The report's classification's substance VERIFIES and is not the defect the classification rules out: pin in isolation `1 passed in 0.28s`; whole module serial `16 passed in 2.32s`; whole module at `-n 6 --dist worksteal` green three times (`16 passed in 1.95s` / `1.92s` / `1.87s`); and no `orchestrator-probe-*`/`orchestrator-probe-exec-*` name remains anywhere on /tmp after any run (`find /tmp -maxdepth 2 -name 'orchestrator-probe-*'` empty; the delivered janitor's reaping held). What the re-drives measure is the pin's whole-shared-`gettempdir()` scan coupling the committed member to concurrent unrelated state at a HIGH rate under the full battery's interleave — FINDINGS F2.
VERIFIED: `scripts/check-tests` (the full standing battery, the documented form, run as this review's falsification ground) — `1094 passed in 1221.86s (0:20:21)`, exit 0 — the same 1094-test population the exit-gate-runs review's standing-battery line displayed, green on this review's own drive; the slowest-five instrument is not passed (adding one would substitute a non-documented command form; the battery's wall is recorded instead).
VERIFIED: `scripts/check-static` — exit 0; the gated `ruff check --select BLE001,S110 src/orchestrator` prints `All checks passed!`; the advisory line prints `Found 104 errors.` (unchanged from REV's 104 and the report's own line).
VERIFIED: `scripts/check-docs` — exit 0, prints `🔍 410 Total (in 761ms) 🔗 183 Unique ✅ 410 OK 🚫 0 Errors` at this review's HEAD (`eddfdd0`) — the report's pasted line at FINAL reads `🔍 409 Total (in 753ms) 🔗 183 Unique ✅ 409 OK 🚫 0 Errors` (lychee's own summary line, rule 14), and both counts are accounted: the base 406/181 at REV grows by exactly the three links the range adds (the brief's one `[self-host.md](../self-host.md#phase-6-operational-envelope)` link — its only internal link; the Status paragraph's two report links — self-host.md's internal-link count alone goes 23→25), then one more link in the review-brief file this review's HEAD carries (`docs/research/phase6-closeout-status-review-brief.md` contains exactly one internal link, the same self-host anchor, and `eddfdd0` is the only commit after `af08b8d8`) — the report's "406→409 delta is exactly the three markdown links added across the range" holds, Unique +2 because the two report targets were linked for the first time anywhere (`git grep` over `18d9f89` returns no link to either report). No baseline claim is contradicted.
NOT VERIFIED: the campaign itself is not re-driven — no new provider run, no spend (OUT OF SCOPE); its figures stand on the reports' printed lines plus this review's at-rest re-derivations named above. Nothing else in the step's delivery was left unverified.
FINDINGS:
  F1 | SEVERITY: P2 | ORIGIN: CODE | STATE: carried
  The step's two commits under review (`af08b8d89824b97664496e34dd5365b019399cb9`, `18e7122736daeec4157b03d167dc2c57f3269061`) carry no `Actor: IMPLEMENTER` trailer (subject-only messages), against the method's actor-trailer mandate and against the review brief's gate sentence that asserts the commits "carry the trailer `Actor: IMPLEMENTER`" — the assertion is false here, as it was for the previous step (the exit-gate-runs review's F4, `67f5531`/`36e42a4`). Reviewer authorship was verified independently in the trailer's absence: this review's session began only after the planner's review-brief commit `eddfdd0` was HEAD, and both commits pre-date it (the author identity is the shared `gituser <a@b.c>`, which the absent trailer would have disambiguated). The lapse continues its step-wide drift: the last trailerred commit on the branch is `77a5195` (`Actor: REVIEWER`); since it, `18d9f89` (planner), `2e2aad3` (planner), `af08b8d8` and `18e7122` (implementer), and `eddfdd0` (planner) carry none.
  FIXED: no regression test — a commit-message form the delivery cannot amend post-review without moving FINAL ("FINAL must not change after the report names it"); next action the planner's: restore the trailer discipline in practice and let later steps carry it, absorbing with the carried F4-class note.
  F2 | SEVERITY: P2 | ORIGIN: CODE | STATE: carried
  The janitor pin's scan couples the committed member to unrelated concurrent state: `_probe_tree_names()` (`tests/test_probe_oracle.py:138`) snapshots the WHOLE shared `tempfile.gettempdir()` listing by prefix, and the pin's assertion (`:170`) set-differences that whole listing across the test's window — so any other process on the host creating an `orchestrator-probe-*`/`orchestrator-probe-exec-*` tree inside the window reddens the pin, and under the full battery's 950-test worksteal phase the collision rate is high (this review's two consecutive `--fast --parallel` re-drives both reddened at `:170` with peer trees in the window, against the report's three named green re-drives; the red is the drive's, not the machine's). Nothing is evaded and nothing is leaked — the delivered reaper is intact (isolation, module-serial, and module-at-`-n 6` all green; no tree residue after any run) — but the class is exactly the one `scripts/check-tests`' own serial-modules comment already files ("These modules scan the process table for leftover sandbox children … so another worker's processes make them fail", lines 49–52): a member whose window is exposed to other workers belongs in the serial set, and `tests/test_probe_oracle.py` is not among the three `SERIAL_MODULES`; this is the F24 "over-match brittleness" family surfacing on the janitor pin the delivered step's runner text itself names ("whose whole-tempdir `_probe_tree_names()` scan is exposed to xdist peers — contention (isolation), not a reproducible defect": the classification's substance is verified here, but as a battery member the red reproduces at a rate that will steward future steps' commit hooks randomly).
  FIXED: no regression test — the red needs the full 950-test interleave (the module's own `-n 6` drives do not reproduce its window), and a dedicated reproducer node would re-choreograph cross-process scheduling rather than pin a delivered behaviour; next action the planner's: a later brief re-scopes the pin's window scan (an own-subtree capture instead of the whole-`gettempdir()` set-difference, at minimum capturing the `before`/`after` window inside the run's own pytest tmp root) or moves the module into the check-tests serial set alongside the three modules its comment already names.
  F3 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
  The Record's exit-gate-runs row's REV cell is not a commit object: `docs/record.md:318` names `c41834285ba0c54f98356d9b108347868759ba25e`, and `git rev-parse --verify -q c41834285ba0c54f98356d9b108347868759ba25e^{commit}` fails (`unknown revision or path not in the working tree`); the true full hash of the commit the reports name as REV (`c418342`, the brief's parent — the "R15 re-ruled: the worksheet's 16-trial shape, kilo-auto/free, env-pinned" commit) is `c418342625ca65ea3f5637b03ada4acde5641d0c`, diverging from the cell at the ninth hex digit. Every sibling row's 40-char hash resolves exactly (five checked: `d9c660e0…`, `12f1f839…`, `cc2ac822…`, `e2271de1…`, `a1d4e07b…`, each `git rev-parse --verify` true), so the column's convention is the full resolvable hash and the table's header defines the cell as "the same `REV` the reports name". The row was written by the exit-gate-runs closeout at `18d9f89` — before this step's base — and the row's existence is what the updated Status's "Record row `18d9f89`" pointer leans on (the pointer holds; the cell inside it does not).
  FIXED: no regression test — a hash cell in a committed record the step under review did not touch; next action the planner's: correct the cell at the next closeout's record write (or by the planner's carried-note correction), preventing procedure step 2's reader from following an unresolvable pointer.
  F4 | SEVERITY: P2 | ORIGIN: BRIEF | STATE: carried
  The Status sentence misifies the scan's population: "credential-bytes scan 0 hits over the 2326 stores present" — the committed reports' own line is `files scanned: 2326` (2326 regular files over the whole workdir, the `.git` internals and calibration/target trees included), and the reports' own store census enumerates the actual stores as the named artifacts (sixteen `trials/<ARM>-<seed>/events.db`, `probe/events.db`, `experiment-spend.json`, `identity.json`, the materialised trees) — by no report's line is "the stores present" 2326. The figure is exact on every surface; the noun is the brief's own mandated wording — the ACCEPTANCE line dictates "the 2326 stores present" while the same brief's CONTEXT writes "0 hits over 2326 files" and this review-brief's FOCUS writes "2326 files / 0 hits" (an internal inconsistency in the brief's own text, the F13/F17/F23 class). The implementer had no latitude: the sentence is the brief's verbatim text.
  FIXED: no regression test — ORIGIN: BRIEF; next action the planner's: a later committed brief's text correction (the sentence's noun), absorbed into the carried F28/F31/F33-class note, and correcting the self-host sentence the next time a step touches it.
SHAPES: two probed, unbuilt.

- The whole-battery parallel janitor window (the F2 red class): the pin's `:170` whole-`gettempdir()` set-difference window colliding with a peer's live `orchestrator-probe-*` tree, probed by re-driving `scripts/check-tests --fast --parallel` twice and by the three graded exposures (isolation, module-serial, module-at-`-n 6`) named in VERIFIED. Not built as a test: the red needs the full 950-test cross-worker schedule; a committed reproducer would pin the battery's scheduling, not the delivered code.

- The at-rest campaign store re-derivation (the ledger parse, the sha256, the 2326 census): probed by session-side reads over `/tmp/kilo/matrix` named in VERIFIED. Not built as a test: it audits session-stored evidence, not tree code — no committed node owns a contract over `/tmp/kilo` stores, the same justification the exit-gate-runs review's SHAPES entry gives.

TRIPPED: none
PROPOSED ROWS: none
METHOD FRICTION: none — both documented forms (`--fast`, full `check-tests`) ran as documented; the parallel form's red is recorded with both its red and green draws; no interpreter, wrapper, or narrowed form was substituted anywhere.
INHERITED:
From `docs/research/phase6-deadline-margin-review.md` (the exit-gate-runs chain's carried set, the review brief's INHERITED as received — the chain's carried blocks verbatim), the still-open findings, each with its block as the brief moves it:

  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).

  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).

From `docs/research/phase6-exit-gate-runs-review.md`, its find-blocks' received-carried items verbatim (the chain's F1/F2 are the exit-gate-runs implementer report's filings; F3/F4 its review's filings — all carried, none with an action a step of this chain has taken yet):
  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).
  F3 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: carried
  The implementer report's figure "The seven `budget_exhaustion=True` trials" miscounts its own enumeration (which names six trials: RECURSIVE_LAZY 1001/1002/1004, RECURSIVE_CONCURRENT 1001/1002/1004) and every countable surface of the run agrees on six: the narration census (`6 True / 10 False`), the runner's structured `trials` array (6), and exactly six `BUDGET_EXHAUSTION_RECORDED` rows across the trial stores (all 54 attempt-level `budget_exhausted` values are False — the "seventh" exists on no surface). The review brief's FOCUS carries the same "seven" twice.
  FIXED: no regression test — the figure lives in a committed report; the census is re-derived in this review's VERIFIED; next action the planner's (or a later step's) correction of the carried note, in the F28/F31/F33 form.
  F4 | SEVERITY: P2 | ORIGIN: CODE | STATE: carried
  The step's two commits under review (`67f5531`, `36e42a4`) carry no `Actor: IMPLEMENTER` trailer, against the method's mandate ("Every actor commits its own work with an `Actor: …` trailer, because … the trailer is the only thing that makes 'you must not be the author of the commits under review' checkable") and against the templates' commit instructions; the review brief's gate sentence asserts the trailer "carry" and its assertion is false here. The lapse is step-wide: since `c418342` five consecutive commits lack their actor trailer (`d1997b0`, `bc56b7c` — planner; `67f5531`, `36e42a4` — implementer; `715f425` — planner); the last trailerred commits are `c418342` (`Actor: OWNER`) and earlier. Reviewer authorship of the two commits was still verified independently (both were delivered from the campaign session's own workspace before this reviewer's session began; the commit author is the shared `gituser` identity, which the absent trailer would have disambiguated).
  FIXED: no regression test — a commit-message form the delivery cannot amend post-review without moving FINAL ("FINAL must not change after the report names it"); next action the planner's: restore the trailer discipline in practice (its corrections travel as rulings; a re-cut is not possible for a past commit) and let later steps carry it.

Then each finding marked carried by this review, with its ID: F13, F14, F17, F21, F22, F23, F26, F28, F29, F30, F31, F32, F33 (received carried, kept carried — this review touched neither the brief-text defects they name nor any code path they cite); the exit-gate-runs chain's F1, F2 (received carried, kept carried — re-verified in the exit-gate-runs review's own VERIFIED and their premises unchanged here); its F3, F4 (received carried, kept carried — the trailer lapse its F4 filed recurs in this step's two commits as this report's own F1, and the six-vs-seven figure stands as re-derived); and this report's F1, F2, F3, F4 (filed by this review, carried — a commit-form lapse in the step's two commits, a battery-member scan-coupling in the janitor pin, a falsified hash cell in the exit-gate-runs Record row, and the brief's mandated "stores" noun — each with its named next action).
