REPORT — phase5-run-entry-fixes-9-review
STEP: phase5-run-entry-fixes-9 (Phase 5: FFR-4's mutant prediction corrected; Phase 5 status current)
REV: 49e98407fbe3c96ecbc093afafd2ff8b8e5185aa   FINAL: ea9c3b000683d673dfd09336cb2f7b6a732c3a7a
CHANGED: docs/research/phase5-run-entry-fixes-9-review.md — this report (no new test file: every
  finding this review adjudicates is ORIGIN: BRIEF prose with no regression test, and every shape
  named below is a measurement of a committed test's face, so nothing rides the suite).
VERIFIED: scripts/check-scope 49e98407fbe3c96ecbc093afafd2ff8b8e5185aa ea9c3b000683d673dfd09336cb2f7b6a732c3a7a docs/research/phase5-run-entry-fixes-9-brief.md — exit 0, run first as REVIEW ACCEPTANCE. Prints `claimed: docs/self-host.md`, `tripped: none`, `scope: ok (2 path(s) in 49e98407fbe3c96ecbc093afafd2ff8b8e5185aa..ea9c3b000683d673dfd09336cb2f7b6a732c3a7a)` — the two paths are the brief itself (added at `40d102b` under `docs/research/`) and the in-scope `docs/self-host.md`. Its companion on the empty diff, `scripts/check-scope 49e9840… 49e9840… …`, exits 1 printing `scope: … is empty; is FINAL the right sha?` — the reading-2 test below.
VERIFIED: question 1, the counterfactual re-measured on a scratch copy (`git archive ea9c3b0 | tar -x -C /tmp/kilo/fix9-mutant`), diff-verified before every use: `diff <(git show ca9a25a:src/orchestrator/run_entry.py) src/orchestrator/run_entry.py` names exactly one line — `214c214`, the `-`/`+` pair `if not isinstance(kernel_id, str) or not kernel_id.strip():` → `if not isinstance(kernel_id, str):`. (1) `.venv/bin/python -m pytest tests/test_run_entry_kernel_id_refusal.py -v` on that copy: `1 failed, 1 passed in 0.29s`, RED at `tests/test_run_entry_kernel_id_refusal.py:86` with `Failed: FRK-1: a empty kernel_id was admitted by validate_run_spec` — the **empty** member first, exactly as the brief's correction states, and `:86` is the pin's `pytest.fail` line, so the named failure is the assertion, not a collection error. Member-by-member under the same mutant (`validate_run_spec` hand runner with the test module's own `_spec`): empty, whitespace `   `, tab `\t`, newline `\n` all ADMITTED (they pass the mutant's type test); dict, null, int REFUSED. (2) The matrix-reaching counterfactual the brief's clause describes: with the scratch matrix's members after `null` reordered so whitespace is the first admitted member, the same pin REDs at the same `:86` with `Failed: FRK-1: a whitespace-first kernel_id was admitted by validate_run_spec` — the whitespace member does red when the loop reaches it. (3) The matrix order in evidence is as cited: `tests/test_run_entry_kernel_id_refusal.py:60-68` holds `_WRONG_TYPED_KERNEL_IDS` with `("", "empty")` at :66 preceding `("   ", "whitespace")` at :67, so on the delivered order the loop stops at empty and never reaches whitespace. Green at FINAL: the kernel-id battery line below is `7 passed` with no code change. Attacked-and-not-falsified.
VERIFIED: question 1, the correction's honesty (no new prediction). The operative correcting text — the brief's FFR-4 disposition clause and the `Mutant-(b) correction (FFR-4's close)` paragraph — carries no `would red` form: it states the measured member (`the pin reds on the empty member first`, at `tests/test_run_entry_kernel_id_refusal.py:86` with the quoted `Failed: FRK-1: a empty kernel_id was admitted by validate_run_spec`), the matrix order (`empty precedes whitespace in _WRONG_TYPED_KERNEL_IDS` at `:60-68`), and the measured passes (`the whitespace-only member does pass`; the strip test covers empty/whitespace). The `would red only in a matrix that reached it` form appears in the brief at `GOAL` only, and is the fixes-8 review's own `FFR-4` `FIXED:` clause in the rule-2 moved block, copied verbatim — it is not this brief's own new claim. Its content is entailed by two measured facts (whitespace ADMITTED under the mutant; the pin fails on any admitted member at `:86`) and it names the loop-order consequence, not a predicted outcome of a new run: measurement (2) above makes it true, not merely plausible. Nothing here is a restated prediction of the FFR-2/FFR-3 class (both were *falsified* predictions); no finding per prediction rides. On the same question, the correction names the pin's file and line plus its named translation rather than spelling the full node id `tests/test_run_entry_kernel_id_refusal.py::test_kernel_id_load_refusal` — the form the brief's own disposition prescribes and the fixes-8 review used; the two-case module and the quoted message leave no ambiguity about which test, and an ORIGIN: BRIEF text-close owes no red evidence, so nothing is filed for it either. Attacked-and-not-falsified.
VERIFIED: question 2, F43's one-reading claim (guarantee attacked-and-not-falsified). The implementer's reading — the implementer applies the OWNER's `b68be80` status-current form to the status lines the OUT OF SCOPE bullet names and touches the status-current convention itself never — is the single reading consistent with the rest of the brief: GOAL ("`docs/self-host.md`'s Phase 5 status names the fixes-8 round so the exit gate's input is current") assigns the edit; IN SCOPE names the file; OUT OF SCOPE specifies the shape and adds "If the status needs any other change, stop and report `IN SCOPE` as insufficient"; ACCEPTANCE's `scripts/check-scope` needs a non-empty `REV..FINAL` diff; and the commit instructions make the edited tree's commit FINAL. The competing reading — the implementer touches nothing and an OWNER commit amends the document, as `b68be80` did outside every step's scope — cannot satisfy the brief's own mechanics: with nothing changed there is no work commit, and `scripts/check-scope 49e9840… 49e9840… …` exits 1 (`scope: … is empty; is FINAL the right sha?`) rather than green. No third reading fits: `docs/self-host.md`'s Phase 5 status is the only IN SCOPE path, so "the implementer does not touch it" can only attach to the convention invoked in the same clause, which is what the implementer took. The clause is therefore misworded on the actor while describing this step's work, so F43's one-reading close form holds. The trailer is separately confirmed: the brief's commit instructions prescribe `Actor: IMPLEMENTER` for the actor's own work and `ea9c3b0` carries exactly that trailer; rule 5 governs who may amend the convention and the status lines' form, never the actor's trailer. `git show ea9c3b0 --format=%b` names `Actor: IMPLEMENTER`; `git show 40d102b --format=%b` names `Actor: PLANNER` and `git show 230c61ab --format=%b` likewise — this session wrote none of them.
VERIFIED: question 2, F43's `STATE: fixed` adjudicated against rule 8's close forms (guarantee attacked-and-not-falsified on the form it used). The available close is the one the implementer used — `fixed` with a `FIXED:` line reading `no regression test` and naming the reading taken — and not the closeout-only form. Ground: the F35 class, `docs/research/phase5-run-entry-fixes-5-implementer.md` — a one-reading ORIGIN: BRIEF finding whose `FIXED:` reads "`no regression test` — the reading taken is the one consistent with the brief's GOAL… No brief text correction is needed because no other reading keeps the GOAL's sentence true", which the fixes-5 review adjudicated honest on record.md#observations' lesson "An `ORIGIN: BRIEF` finding's close form follows where the text lives", "the fix-order words are subordinate to the GOAL and live in the same brief; the GOAL's sentence is the text whose correction rule 8 would govern, and it needs none". F43 has that shape: its defective words sit in `DIVERGENCES`, subordinate to the GOAL/IN SCOPE/OUT OF SCOPE/ACCEPTANCE sentences that are correct as written and that no other reading keeps true. The `fixed: recorded in <the review that filed it>` alternative question 2 offers is the F41 class — `docs/research/phase5-run-entry-fixes-7-review.md`, a defect in an already-committed **report**, which was `carried` there and closed at that step's closeout; F43 is a defect in a committed brief, which rule 8 routes to record.md's Observations at closeout, not to that FIXED form, so neither closeout route supplies a `STATE` the implementer could have used instead. Nothing is left unclosed by this: rule 8 forbids carrying what no one can act on, a later brief's text correction is not owed (F35's ground), and the loop's exit condition stays paid. One imprecision is recorded rather than filed, because it changes no actor's action, no count, and no close: F43's last words tag the close "the closeout form for a finding with no possible next action", while rule 8's closeout sentence names two forms — the `recorded in` line for report defects and Observations for an authoring lesson from a finished brief — and the close the implementer actually applied is the one-reading form above. Whether the closeout owes an Observations line for F43's own authoring lesson is the closeout's `record.md` write, outside this brief's scope (its OUT OF SCOPE forbids a `docs/record.md` edit) and outside review — an implementer report cannot carry it either.
VERIFIED: question 3, the delivered shape (guarantee attacked-and-not-falsified). `git diff 49e9840… ea9c3b0… --stat` names exactly two paths (the brief's 133 added lines and `docs/self-host.md`'s 5/5), and `git diff 49e9840… ea9c3b0… -- docs/self-host.md` is one hunk in the Phase 5 status paragraph: "eight recorded rounds" → "nine recorded rounds", the `[fixes-8](research/phase5-run-entry-fixes-8-implementer.md)` report added to the list (nine reports named, nine rounds — a count from the artifact's own list), the disposal range rewritten to "F26–FRK-1/FFR-1–FFR-4 disposed", and FRK-1 restated as "refused at load (R11)… and FFR-1–FFR-4 close by the fixes-8 and fixes-9 briefs' text" — the four elements OUT OF SCOPE names (list, range, R11 consumed, the `b68be80` status-current form, whose own diff at `b68be80` is the same paragraph shape). The Phase 5 goal and exit gate lines, Phase 6/7/8, and every other phase's status are byte-identical (the single hunk covers no other line); all nine linked reports exist and `scripts/check-docs` resolves them with 0 errors. `git diff ca9a25a ea9c3b0… -- src/orchestrator/run_entry.py` is empty — the six guard lines are byte-identical to the delivered fix — and `git diff 49e9840… ea9c3b0… -- tests/` is empty, so all seven committed battery modules are byte-identical to REV with no test edit. Every command above names the interpreter or script the brief's ACCEPTANCE names; no substituted form was run.
VERIFIED: scripts/check-tests — `1069 passed in 622.21s (0:10:22)`, exit 0 (this review's standing battery and falsification ground; 1069 = 963 fast + 106 slow, the fixes-8 review's shape unchanged).
VERIFIED: scripts/check-tests --fast — `963 passed, 106 deselected in 95.76s (0:01:35)`, exit 0 (the report's figure re-measured: same counts, wall-clock drifts run to run).
VERIFIED: .venv/bin/python -m pytest tests/test_run_entry_kernel_id_refusal.py tests/test_run_entry_loader_envelope.py tests/test_phase5_run_entry_fixes_6_review.py — `7 passed in 7.52s`, exit 0 (2 kernel-id cases + 2 envelope cases + 3 absorbed probes).
VERIFIED: .venv/bin/python -m pytest tests/test_phase5_run_entry_fixes_5_review.py tests/test_phase5_run_entry_fixes_4_review.py tests/test_run_entry.py tests/test_phase5_run_entry_fixes_2_review.py — `54 passed in 28.22s`, exit 0 (the loader's whole committed battery; the held sets of the run-entry stream — the `kernel_id load refusal` and `kernel_id fallback` pins, the envelope property, the `Read-refusal pin`, and the F38/F39/F40 absorbed probes — all green at FINAL with no test edit).
VERIFIED: scripts/check-static — exit 0. The gated `ruff check --select BLE001,S110 src/orchestrator` is clean and the advisory line prints `Found 104 errors.` (unchanged from REV's 104, and from every figure the report carries).
VERIFIED: scripts/check-docs — exit 0. This tree prints `375 Total (in 999ms) 173 Unique 375 OK 0 Errors`; a throwaway `git worktree` at FINAL (`git worktree add --detach /tmp/kilo/fix9-final ea9c3b000683d673dfd09336cb2f7b6a732c3a7a`, removed afterwards, `git worktree prune` clean) prints `374 Total (in 310ms) 173 Unique 374 OK 0 Errors` — the report's own figure, so it is the gate's printed line at FINAL and neither CONTEXT's REV quote (372) nor a recomputed total. The +1 between the two is this review report's one counted link, the same drift class FFR-2 recorded.
VERIFIED: the Phase 5 exit gate through the documented head, `python3 scripts/orchestrate.py start --spec … --workdir … --store …`, with a dict `kernel_id` spec, run from a tmp dir: exit 1, stdout 0 bytes, terminal exception `orchestrator.run_entry.RunSpecError: project_kernel.kernel_id must be a non-empty string is invalid: {'a': 1}` raised from `validate_run_spec` (`src/orchestrator/run_entry.py:215`) through `load_run_spec` (`:295`) — a direct validation refusal, no chained cause, no raw escape — with the store file ABSENT and the workdir NOT created, so the refusal precedes any store open (guarantee attacked-and-not-falsified with this reproducer; the operator-observable face only, which is where this step's validation stage touches the gate).
VERIFIED: the report's forms and counts (J-class; guarantee attacked-and-not-falsified). Every figure is a gate's printed line re-measured above. `FINAL` names the work commit `ea9c3b000683d673dfd09336cb2f7b6a732c3a7a`, never the report commit `22f6ad4`. `TRIPPED: none` matches check-scope's `tripped:` line. The embedded brief copy is byte-exact against the committed `docs/research/phase5-run-entry-fixes-9-brief.md` (`diff` of the embedded copy against `git show 230c61ab:docs/research/phase5-run-entry-fixes-9-brief.md`, 133 lines, exit 0), and the brief itself is unchanged between FINAL and this tree, so nothing to review against another file. `Every field of the template present and filled, nothing outside it added`: REPORT/STEP/REV+FINAL/CHANGED/VERIFIED/NOT VERIFIED/FINDINGS/SHAPES/TRIPPED/PROPOSED ROWS/METHOD FRICTION/INHERITED all present and filled, and no extra field rides (no `DECISION`, no `CLOSEOUT` — the closeout stanza is the planner's append to the step's last report).
VERIFIED: `SHAPES: none` read against rule 7 (guarantee attacked-and-not-falsified). The two shapes the brief names each have their pins in the battery lines above (`kernel_id load refusal` and `kernel_id fallback` are the kernel-id module's two cases), the maintained surfaces (envelope property, `Read-refusal pin`, F38/F39/F40 probes) are named green rather than hidden, and the one shape probed that the suite does not build — the no-`strip()` mutant variant — is named by the report in both FINDINGS (FFR-4's red evidence) and SHAPES, unbuilt by design because the delivered `strip()` is the boundary the pin guards, so nothing probed is left unnamed. The `none —` lead with its stated reason followed by named unbuilt faces is the fixes-7 implementer's, fixes-8 implementer's and fixes-7 review's own form.
VERIFIED: `INHERITED: none` holds against the received set (rule 2/8; guarantee attacked-and-not-falsified). The brief received two copied blocks — the fixes-8 implementer's (`none — the received INHERITED carried exactly four open items… all four are disposed by this step`) and the fixes-8 review's (`Nothing from the received set is still open: FRK-1 is fixed… the three FFRs close… and F41 was closed at the fixes-7 closeout… One new item is still open and carried forward by this review: FFR-4`) — plus the moved FFR-4 entry. Exactly one item came in open (FFR-4) and the step closes it by the brief's own text under rule 8 (re-measured above); F43 is filed `fixed`, not carried, so the block's lead `none` is not the contradiction record.md#observations records (a block leading `none` while marking items `carried`). Nothing else is open: no residual row rides (`tripped: none`), no `tests/open_defects.yaml` registration is owed (every finding is ORIGIN: BRIEF with no regression test, the FFR-3 class), and no shape is left unbuilt unnamed.
NOT VERIFIED: none. Wall-clock durations drift run to run and the counts are re-measured above; no pre-image or historical claim was taken from a report's account — every one was read with `git show`/`git diff`/`git archive`, and the tree was clean at dispatch (`git status --porcelain` empty).
FINDINGS: none
SHAPES: probed by this review's hand runners and scratch copies; the committed suite builds none of these, and each is a measurement of a committed test's face or of a deliberately unbuilt counterfactual (the brief's two named shapes and the maintained surfaces have their own pins and are not repeated here):
  - the no-`strip()` mutant variant (`isinstance(kernel_id, str)` alone), diff-verified against `git show ca9a25a:src/orchestrator/run_entry.py` as exactly the `strip()` removal at `src/orchestrator/run_entry.py:214` — the pin reds at `tests/test_run_entry_kernel_id_refusal.py:86` on the **empty** member with `Failed: FRK-1: a empty kernel_id was admitted by validate_run_spec` (1 failed, 1 passed). Unbuilt by design: the delivered `strip()` is the boundary the pin guards, so the suite must not build its absence.
  - the same mutant's member-by-member face, hand-driven through `validate_run_spec` with the test module's own `_spec` — empty, whitespace `   `, tab `\t` and newline `\n` ADMITTED; dict, null, int REFUSED. No committed probe drives an individual admitted member outside the loop form.
  - the matrix-reaching counterfactual: on the same diff-verified copy with the admitted members reordered so whitespace is reached first, the pin reds at `:86` naming that member (`Failed: FRK-1: a whitespace-first kernel_id was admitted by validate_run_spec`) — the reach-contigent half of the correction, which no committed matrix reaches.
  - `python3 scripts/orchestrate.py start` against a wrong-typed `kernel_id` spec from a clean tmp dir: exit 1, empty stdout, named `RunSpecError` with `__cause__ is None`, store file absent and workdir not created — the operator-observable refusal at the documented head; no test drives the `start` head against a refused spec in this tree.
  - `scripts/check-scope` on the empty diff (reading-2's falsifier): exit 1 with `scope: … is empty; is FINAL the right sha?` — the suite's own `tests/test_check_scope.py` covers the tripped and claimed faces, not this one.
  - `scripts/check-docs` at a throwaway `git worktree` of FINAL — `374 Total … 0 Errors` against this tree's `375`; not a test, and not automatable without leaving a worktree behind. No facet is claimed unchanged on a surface the step touched: the step touched no code, so every code facet is undisturbed by construction and re-driven green in the battery and full-suite lines above.
TRIPPED: none
PROPOSED ROWS: none
METHOD FRICTION: none — the review brief's QUESTIONS TO ATTACK and the review form were followed verbatim; every answer lives in VERIFIED or SHAPES, no template field was added, and nothing outside the form was required to answer. Nothing else departed from `docs/templates/report.txt` or the review brief's own instructions.
INHERITED:
Nothing received is still open, and nothing new is open after this review. The received INHERITED carried exactly one open item — FFR-4 (STATE: carried) — moved by this step's brief into its own body and closed by that brief's own text under rule 8 (its correction re-measured and its honesty adjudicated above); F43 is filed `fixed` by the implementer's report and its close form is adjudicated above, so it is not carried; the two copied blocks' other items (FRK-1 fixed at `ca9a25a`, FFR-1/FFR-2/FFR-3 closed by the fixes-8 brief's text, F41 closed at the fixes-7 closeout) were already disposed before this step. No residual row rides (rule 3: `tripped: none`).

CLOSEOUT
DECLINED: none
DIVERGENCES CLOSED: document amended (b68be80aa9762eb964bd27190331617e404014ae) — the brief's DIVERGENCES named the stale Phase 5 status paragraph for the OWNER's status-current convention; the OWNER's status-current form at b68be80 was the amendment, applied by this step's work to the status lines the brief's OUT OF SCOPE names and re-verified at ea9c3b0 (delivered shape adjudicated in this review, question 3).
