REVIEWED-137 (PENDING-169 §5) — the verdicts sitting placed, with six corrections

Steward placement of the four judgments PENDING-169 §5 reserved, produced under
REVIEWED-136's measure/verdict split. Two answered, two recorded UNANSWERABLE
with their reasons — which is the sitting's result, not a failure of it.

Presence: occupied on a per-machine clock, occupancy uncorrelated with where the
work is. Fidelity retrospective: unanswerable — the 14-word pile-up and the
tick-to-terminal lag are one phenomenon on a 44/50 vs 1/52 partition, so the
evidence base for every cap argument is confounded with an unidentified
mechanism. Condition G: the instrument is designed and NOT authorized; no write
path is restored. Wrap: cost answered, cap unanswerable.

Six corrections applied after jurist review, by exact-match anchor:
  A  §7 duplicated line removed
  B  §6 recorded deviation, on steward release
  C  §8 rewritten — 168/180 merged onto one count-unit decision, the
     assertion-versus-file gap added, the non-event pairing §3 requires
  D  §1 coverage restated as 12 of 13 against 11.35, excluding the measuring
     session from both terms; the 13-of-14 pair is perishable and drifted from
     11.62 to 12.07 during the sitting itself
  E  §6 line numbers given as filed/defective/repaired, with the warning that a
     line number is not a durable anchor
  F  §6 condition G re-verified at 3d45e3a, untouched by 049ca57

Also joins the entry title, which the paste had broken across two lines — it
had truncated at "designed but not", inverting the meaning at the break.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BsN7nKHjKBsn5bfNRRCNmo

REVIEWED.md

@@ -3112,3 +3112,250 @@ which is condition 2's rationale defeated the moment it is honoured.
 what bound, with what expiry and whose act — including whether counting can be made
 structurally content-free at the point of write — is filed as open and is owed a
 ruling before any write path is restored.
+
+## REVIEWED-137 — PENDING-169 §5 — The verdicts sitting: presence ruled, fidelity recorded unanswerable, the condition-G instrument designed but not authorized
+
+**Date:** 2026-09-09
+**Decision:** The four reserved judgments are MADE. Two are answered; two are recorded
+UNANSWERABLE with their reasons, which is a finding and not a deferral. **Condition G is
+NOT lifted by this entry.** PENDING-167's cap question is recorded as unanswerable on
+present evidence.
+**Ruled by:** jurist (Claude.app), on measurements produced by the executor under
+REVIEWED-136's split and reported with their controls.
+**Authorized by:** steward, 2026-09-09.
+
+**Produced under REVIEWED-136's split**, which replaced PENDING-162 option 2: the
+executor produces the measurements and does not deliver the verdicts. The split held —
+every figure below was re-derived from the substrate by the executor and judged by the
+jurist, and three of the executor's findings were overturned in the process.
+
+---
+
+### 1 · PRESENCE — the position is occupied, and its occupancy is uncorrelated with the work
+
+**Finding, as one thing rather than two.** The position is occupied on a **per-machine
+clock**, and where it is occupied is unrelated to where the work is.
+
+**Coverage: 12 of 13 occasions** — the measuring session is excluded from both terms, so neither figure moves. Counting it in gives 13 of 14, but that pair is perishable: it drew three further ticks during the sitting, which is why the same quantity read 11.62 when judgment 1 was made and 12.07 at close. Three further sessions are excluded as non-occasions, each with zero ticks and zero Tarbuckle hook runs. The finding is unchanged either way: one silent occasion, gap runs [1]. Against a pre-registered expectation of 11.35. The old_str unchanged, and it verifies unique. The comparison is the routed one — 27% is hardcoded in `DRAW` and pre-registered in §8. The spoken comparison is NOT reported: its estimate is drawn from its own sample.
+
+**Occasion is a judgment, and its threshold is named:** a session in which at least one
+Tarbuckle surface executed. Three sessions fired only `SessionStart` with
+`wake-digest.py` and are not occasions. Deliberately not a size threshold — a 22 KB
+session with a three-second transcript took 43 draws and 8 lines.
+
+**Gap runs [1].** The position has never been vacant for two consecutive occasions.
+
+**Allocation is inverted.** `tarbuckle-last-tick` is one path with no session key, so the
+draw is a race won by render frequency:
+
+- 15% of draws and 30% of spoken lines fall within 30 minutes of recorded activity in
+  their own session; 72% of draws fall beyond two hours.
+- `acaabadf` — still listed by the wake digest as an ARMED stopped worker — took **175
+  draws and 11 lines at a median 43.1 hours** from its last transcript record.
+- On 2026-08-31 two abandoned panes took **48** attributed draws; the 16.7 MB working
+  session took **7**; a 50-minute human session took **1**.
+- `736ef3cb` rendered **380** times that day and won none. `62083f11` rendered 817 and
+  won them. **Render count, not attendance, decides.**
+
+**The wrap surface is a second mechanism, not a footnote.** `Stop` fires for `claude -p`
+subprocesses, so **3 of 10 wrap runs marked the end of the fool's own generator.** The
+wrap is a threshold occasion; 30% of those occasions were thresholds of nothing.
+
+**The one genuine miss is `736ef3cb`, and it is a different failure from every other.**
+Fifty minutes of attended human work, and he was never asked — no draw, no rejection, no
+record. **Tick starvation is invisible to every instrument built so far**, and it is the
+failure mode a positional presence is most exposed to.
+
+**Consequence:** frequency is the wrong dial. The clock is.
+
+⚠ **Bound.** The activity proxy is transcript records, which under-report attendance. The
+unbounded direction runs in the finding's favour: a pane can render while closed to view.
+n = 14; this is a description, not a test. **What would overturn it:** a second
+tick-starvation instance, making `736ef3cb` a class rather than an incident.
+
+---
+
+### 2 · FIDELITY, RETROSPECTIVE — UNANSWERABLE, and the reason is the finding
+
+**The pile-up and the lag are one phenomenon.** Terminal draws are written by a detached
+child, so a rejection's timestamp is its completion, not its tick. Pairing rejections to
+their tick succeeds on ordinary days and fails on exactly the days in question:
+
+| | pairable within 120 s | not |
+|---|---|---|
+| the five 14-word days (08-28/29/30, 09-01/02) | 6 | **44 — 88% unattributable** |
+| all nine other days | 51 | **1 — 2%** |
+
+**A partition with no exceptions in either direction, over fourteen days.** Whatever
+condition produced the lag produced the 14-word rejections, and nothing produced either
+without the other.
+
+**The attended/unattended split is VOID.** 88% of the rejections it needs are
+unattributable and the remaining corpus is selected against the thing being measured.
+**No rates are recorded here.** A figure with an interval that wide, drawn from such a
+corpus, would be cited in a month without the sentence that qualifies it — the same class
+as the seven-hour interval REVIEWED-136 corrected.
+
+**The temporal framing is dead.** 08-31 — 86 draws, 21 spoke, 6 rejected, zero 14-word,
+between two heavy days — has 6 of 6 rejections pairable. The condition was simply absent.
+**It is not W1 versus W2 and it did not improve after 09-02; it toggles.** Any cap
+argument resting on a date boundary rests on the wrong variable.
+
+**Mechanism NOT established, and not asserted.** `claude -p` carries `timeout=120`, so the
+child cannot be inside the generator for 90 minutes, and a timeout would log
+`generator-failed` (7 in the fortnight). Machine sleep is not supported. The render-volume
+association is strong and not monotone. ⚠ **Its attraction is its plausibility, which is
+PENDING-164's class.** It is the first hypothesis to test and it is untested.
+
+⚠ **Consequence for the banked per-day series**, which is the report's best evidence and
+was banked precisely because 101 entries cannot be re-split after deletion:
+`log_rejection` timestamps at write time in the same child. **The series is binned by
+child-completion time**, and on the five days that matter that is an unbounded interval
+after the tick. The pile-up is far too concentrated to be a binning artifact, but the
+scattered-versus-clustered judgment is being made on completion times and the corpus that
+could have checked it is gone.
+
+**The evidence base for every cap argument ever made about this system is confounded with
+a mechanism nobody has identified.** It was two days from being ruled on as a cap finding.
+
+---
+
+### 3 · THE CONDITION-G INSTRUMENT — designed, NOT authorized
+
+⚠ **This entry does not restore any write path.** Condition G stands. What follows is a
+design judgment; the act it describes requires its own ruling.
+
+**Established by the case rather than argued from principle.** The decisive finding above
+was derived entirely from timestamps, counts and surface labels. No line content was
+needed or used, and the one content-bearing field contributed nothing to it while being
+the reason the corpus had to be destroyed. **The content was net-negative: it bought
+nothing and it cost the evidence base.**
+
+**The instrument is an extension of `log_event`, not a successor to `log_rejection`.**
+That reframing is the substance: `log_event` already records occurrence-not-utterance
+under §8, and `tarbuckle-draws.jsonl` is explicitly outside REVIEWED-128's deletion order.
+Fields, ranked by what this sitting proved they would have bought:
+
+1. **`tick_id`** — an opaque id minted at the tick and carried to the detached child. It
+   alone delivers both of the sitting's blocking failures: session attribution and direct
+   lag measurement. **Named by two failures, not designed.**
+2. **`words`** — an integer; the cap question's only real input.
+3. **`reason_category`** — a closed enum. This replaces `why`, which is the actual leak:
+   `echoes_soul()` returns a literal 4-to-6-word run from the suppressed line.
+
+**Structurally content-free means the writer never holds the content** — `acceptable()`
+returns `(ok, category, n_words)` and the logger takes no string parameter, so the write
+path cannot leak because the content is never in its scope.
+
+⚠ **The claim is bounded and is NOT that condition G dissolves.** The hard form — safely
+retaining suppressed speech — was **not needed for this case**, and **no case requiring it
+has been constructed.** That is one instance, not a proof. A later question that genuinely
+needs content must not read this entry as having settled it.
+⚠ **Two limits stated rather than discovered later.** Whether the discarded lines were
+good is permanently out of reach and must not be levied on the instrument; sampled live
+surfacing is ruled out by `log_rejection`'s own docstring, which holds that reading
+rejections as they accumulate *"would shape the net toward lines the reader happens to
+like."* And a rejection counter cannot see tick starvation — **it must be paired with a
+non-event measure.**
+
+---
+
+### 4 · THE WRAP SURFACE — cost answered, cap UNANSWERABLE
+
+**The cap half is not attempted.** Its evidence base is the pile-up, which §2 shows is
+confounded.
+
+**Cost, from `durationMs` on transcript hook attachments, independent of the corpus:**
+
+| | n | median | max | total blocking |
+|---|---|---|---|---|
+| real occasions | 7 | **6.96 s** | 13.57 s | 61.9 s |
+| into mumble subprocesses | 3 | 5.77 s | 6.66 s | 18.2 s |
+
+**80 seconds of blocking `Stop` across a fortnight, of which 18 s — 30% of occasions,
+23% of the time — marked the end of the fool's own generator's run.** Seven real
+thresholds at ~7 s each is proportionate. **The waste is the 30%, not the duration.**
+
+**Remedy, `[FIX]`, not applied in the sitting:** gate the wrap on `TARBUCKLE_CHILD`, which
+`tarbuckle-body.py` already sets and already passes to the child. ⚠ This gives that
+variable a **second meaning** — *and do not speak here*. One comment must name both, or
+the next reader removes the coupling as arbitrary.
+
+⚠ **Floor, not census:** hook runs with empty output are never persisted. The seven are
+the occasions on which he spoke.
+
+---
+
+### 5 · WHAT THIS ENTRY DOES NOT DECIDE
+
+- **Condition G is not lifted.** No write path is restored. §3 is a design, owed a ruling.
+- **PENDING-168's count unit is not settled**, and the executor has declined to increment
+  it — see §7. PENDING-180's deferral binds that ruling to state its unit before its number.
+- **The item list is deliberately held** for a rested sitting. Judgment 2 turned up a
+  partition nobody predicted, and scoping its consequences at the end of a long day is how
+  a cap item was aimed at the wrong file for two weeks.
+- **PENDING-167 is not re-scoped here.** Its cap question is unanswerable on present
+  evidence; the re-scope is an output of §4 and gets its own pass.
+
+---
+
+### 6 · EXECUTED IN THE SITTING — `[FIX]`, under PENDING-180
+
+PENDING-180's own recommendation, in its order.
+
+⚠ **Recorded deviation.** PENDING-180's fix was on the sitting brief's held-out list and was executed in the sitting nonetheless on steward release. Recorded explicitly rather than passed over,
+per the practice REVIEWED-136 established one day earlier.
+
+**(c) Census.** 15 positive-form source assertions across 8 governed scripts. **Exactly
+one self-planted needle:** `tarbuckle-seam.py:165`, `line[:` + `200]`. The polarity argument predicted siblings; there are none. ⚠ Bounds from below — a needle assembled from parts inside a defective control is invisible to this method.
+
+⚠ **The first census carried the defect it was auditing.** Its discriminator (`count == 1`)
+could not separate a self-planted needle from a correct control whose assertion escapes its quotes, and it misflagged a correct control. Corrected to *every occurrence lies on the assertion's own line*, with must-find and two must-not-flag controls.
+
+**(b) `source_has()`** built beside `source_lacks()`, needle assembled from parts, with the must-fail arm the negative form has always had. **(a) repaired** — filed as `:164`, defective at `:165`, repaired at `:176`; aimed at the
+writer's file, split into two arms, with a fail arm. ⚠ The line number is not a durable anchor: it has moved twice in two days.
+
+**Verified by mutation, not by a green selftest.** The repaired control FAILS on both mutations; the old form PASSES the mutation that matters. 136 controls across five files green. Files: `scripts/tarbuckle-mumble.py`, `scripts/tarbuckle-seam.py`.
+
+⚠ **Condition G re-verified after this sitting's edits.** `tarbuckle-mumble.py:36` carries
+`REJECT_LOGGING_ENABLED = False`, confirmed unchanged at `3d45e3a` and untouched by `049ca57`, by diffing the symbol across `3d45e3a..HEAD`. The `= True` at `:472` is inside the selftest's guarded block, restored in the `finally` — the A8/A8n positive control kept rather than adjusted to pass. **No production write path was restored.**
+
+⚠ **A NEW GAP IN THE MECHANISM.** `source_lacks`/`source_has` assemble the **assertion's**
+needle from parts; **nothing stops a comment elsewhere in the file planting a contiguous copy.** Demonstrated: the executor re-created the bug inside the comment explaining the bug, and only the mutation test caught it. **The mechanism covers the assertion, not the file.**
+
+---
+
+### 7 · PARTY DATA — recorded because the configuration is falsifiable
+
+**Jurist, and it is the item that matters.** Three closures in the direction requiring no
+further work, across two days — the third **inside the sitting where the executor had
+flagged the first two**: (i) accepting `0 of 204` as structural in the same message that
+re-labelled it; (ii) offering the W1/W2 surface mix as the cheapest hypothesis, which
+assumes the date boundary §2 killed; (iii) naming `321769ae` as the clean attended
+instance, when it falls 0.03 h on the other side of the threshold the jurist itself
+demanded. ⚠ **Carried unfolded rather than folded into a symmetric tally.** This is the
+seat whose function is resisting premature closure. **Live for PENDING-89 and for
+Constraint 6's falsifiability clause.**
+
+**Executor.** (i) Reported `0 of 89 lines into mumble sessions` as structural on a control
+scoped to the invocation log — an instrument demonstrated on one surface and claimed
+across the set; overturned by transcript hook records. (ii) Arithmetic error in the
+sentence establishing judgment 1's denominator. (iii) A positive control asserted rather
+than established, which failed and falsified the assumption instead of the instrument.
+(iv) Re-planted the needle in the comment explaining the needle. **(i) and (iv) were
+caught by a second instrument, not by review.**
+
+⚠ **PENDING-168's count is NOT incremented here.** Item (iv) is a further occurrence of the self-planting shape and the third written by a party actively watching for it. The executor has declined to count it: choosing the unit that suits one's own number is the CLASS E move PENDING-180's deferral exists to block. **The unit is the steward's.**
+
+---
+
+### 8 · WHAT IS OWED
+
+1. **The item list**, at a rested sitting — the carry list is in the 2026-09-09 daily note and the session record, so it is not reconstructed.
+2. **A ruling on the condition-G instrument** (§3) — three fields on the occurrence log, one of them an id. ⚠ **It must be paired with a non-event measure:** a rejection counter cannot see tick starvation, and `736ef3cb` left no record of any kind.
+3. **The count unit — one steward decision with two dependents.** Instances or occurrences. **PENDING-168's ruling** must state it before its number and carries two corrections (the 2 h 33 min interval; PENDING-180's occurrence). **PENDING-180 is dischargeable** — (c), (b) and (a) executed, addendum drafted and unplaced — on the same decision. Neither moves until the unit is chosen.
+4. **The `TARBUCKLE_CHILD` gate** (§4), `[FIX]`, with its dual-meaning comment.
+5. ⚠ **The assertion-versus-file gap** (§6). `source_lacks`/`source_has` assemble the **assertion's** needle from parts; nothing stops a comment elsewhere in the file planting a contiguous copy. Demonstrated in this sitting, caught only by mutation. **The repaired mechanism is not yet whole**, and §6 declares it verified.
+6. **The render-volume hypothesis**, first to test, plausibility flagged as the hazard.
+
+**For a reader in a month, two things.** **The pile-up was never evidence about the cap** — it was evidence about a condition nobody has named. **And the field that would have survived deletion is one opaque integer**, content-free by construction, which did not exist because nobody had asked the question this sitting's instrument just asked.