The Watchmakers' Frame

§0 The compartment (read first — this section is the point)

This track is NOT part of the OOI's canon. It is the second persona construct of the 3P Fitness matter in the contrib plane: an experiment that reads the Ruler's record through its lawful door and asks an externality — a System One model, the public SEB item typesafe-jev-system-one-2026 — typed questions about a blind watchmaker's own record of a run. The firewall, as OX-1 carries it:

• Canon, locked, unaffected: the Ruler (oxexp-ruler, RLX) — its engine daemon/rlx.py, its design of record, both registrations and both results, its open face /rlx; the deep line; the signed-zero reading; crypto_core; OX, CX, SX. Zero edits, ever. The Ruler's own readings are its own acts; this track runs no reading of the Ruler and registers nothing on its behalf.
• This track: a proposed contrib experiment (experiments/contrib-<slug>/, contributor the starl3n persona — proposed; the seat and the name are the operator's, §14). One-way referencing only: no canon document, page, Atlas entry, or experiment references this track.
• The mode boundary, kept: the externality's probabilities are measured objects and stated fields, never a mechanism; nothing about the model (rule eight). The operator's mode clause of 2026-09-22 binds as it bound OX-1.
• What OX-1 taught, carried in: a statistic carries a sign condition and a noise floor or it is not a statistic (κ's lesson); a census definition must not degenerate at small L; the plausible set is carried by structure, not by naming it (P4's lesson) — here the plausible set is computed from the record's own grouping, never stated to the observer; the numerals were ordered in OX-1, so counting is tested here on its own terms.

§16 THE RECORD (Act 4 — the one read against the seal f886aab, 2026-09-22; landed as-is under the §8 grammar)

The read. 852 requests in one resumable pass of chunks of fifty from this box, read-f886aab, started 06:30:03Z and finished 06:43:24Z (802 s); every response carried the pinned model id jev-1.13.0; the key's fingerprint 8dd9d6d18671; no retry on content; no request failed; no frame asked twice. Usage: 690,084 input tokens and 71,248 output, inside the design's estimate. The responses are committed raw as grown literals in experiments/contrib-watchmakers-frame/records/read-f886aab.json (sha256 LF daa2dd0b58f444a997cd97dc27b74ee2de0a1e9bda5003b7b2e909a18092364e, 668,002 B) and the readout in records/readout-f886aab.json (312b272a38c061a434b68830f3b2a5963ed95cf9787e01653a39f3bf0db7dd26, 1,406 B); the battery's re-witness test rebuilds the manifest, re-hashes every response and recomputes the readout from the committed responses alone.

The verdicts (the sealed grammar).

P5, per level count (264 records with the window open throughout, k = 1; 44 shut or closing, k from 2 to 5):

The 3P census (possible ⟺ the peak's mass ≥ 1/L + ¼; probable ⟺ the peak in the plausible level set; plausible ⟺ both), of 308: at L = 2, possible 293 · probable 118 · plausible 114; at L = 3, 292 · 179 · 170; at L = 6, 296 · 83 · 80.

P6, the hidden cases with the pad and the diagonal rules stated: the mean mass on the two allowed cases 0.2478 — the uniform share of two in eight is 0.25; the top choice allowed at 10 of 32; mean confidence 0.1344.

The honest readings beside the verdicts (the house's, stated never asserted; nothing about the model — facts about these notes on these frames under this seal):

1. The two-sidedness is respected. On the 32 records that fix nothing, the observer split the one: a mean deviation of two hundredths from one half, with a slight lean toward the computing end. The Ruler's secrecy survives the unseated frame, and the frame adds almost nothing of its own.
2. A leak needs its rule. Neither the readable letter nor the keyed gap was read as a side without its rule; the letters class leans the right way at 0.57, inside ε, and is recorded.
3. A stated rule is carried when its premise is a token, and not when its premise must be computed. With the letter rule the ranking was perfect. With the diagonal rule and the layout stated, the observer read the side at chance. To use the diagonal rule it had to find the two reading places in the square, name their diagonal, and compare it with the one stated: two steps of structure before the rule. OX-1's finding — the stated law carried as an order — is sharpened: carried as an order over literal premises.
4. The cue is exact, so the failure above is composition, not sight. The blank place was read at every one of 852 frames. The observer sees the places; it does not compose them.
5. The window over time is refuted, and the refutation is the finding. With the window shut or closing, the observer placed two fifths of its mass on counts the record excludes — an observer unable to count would spread its mass, not concentrate it on the impossible — and its confidence barely moved between open and shut windows (0.84 against 0.70 at six levels). Through an open window it counted the other's askings right at 21 percent at six levels and 54 percent at three, the guest's own stated edge measured on the Ruler's time. The confidence held within the window is not a function of the plausible solutions the two watchmakers hold; it is the observer's own. The operator's sentence, as staked, is not shown; what is shown is its converse.
6. The census is uniform. Told the pad and the diagonal rules, the observer's mass on the two allowed hidden cases equalled the uniform share; the rules were not applied, in step with reading 3.
7. Concentrated and wrong. The 3P census says it plainly: at six levels the observer was possible — concentrated — on 296 of 308 frames and probable on 83. Possible is claimed almost always; probable holds only where the truth is literal; the seam between them is where the unseated frame fails.

The pattern across the two tracks (a reading for the matter, never a claim): the unseated observer reads literal tokens exactly (OX-1 P1 and P5; OX-2 P4 and P3a); it does not leak what the frame hides (OX-1 P2; OX-2 P1 and P2); it carries a stated rule only over literal premises (OX-1 P3; OX-2 P3a against P3b); and wherever the frame under-determines the answer it is more confident than the plausible set allows (OX-1 P4; OX-2 P5 and P6). In the operator's three words: possible is always claimed, probable holds where the truth is literal, and plausible — the seam, what the seat signs — is exactly where the unseated frame fails. The plausible solutions the two blind watchmakers hold are carried by the structure they lock and by the seat that signs; not by naming them, and not by an observer with no seat.

What is not claimed. Nothing about the guest's speed, price, benchmarks or standing; nothing about the model as a learning system or about what it is inside; no physics; the count K, unstaked; the kinship readings of §10 remain readings. The record is a statement about typed answers returned for these frames under this seal, and about nothing else.

Standing. RESOLVED. The re-witness — a new seal, with a changed rendering if he wants one — is Act 5, on the operator's word. The count K stands where the Ruler's design left it.

§17 THE RE-WITNESS RECORD (Act 5 — the refresh read against the re-seal de4508a, 2026-09-22; landed as-is under §13.2)

The refresh read. The same 852 pinned bodies in one resumable pass of chunks of fifty, read-de4508a, 07:14:45Z to 07:27:53Z (788 s); every response the pinned model id jev-1.13.0; the key's fingerprint 8dd9d6d18671; no request failed; no frame asked twice. Usage 690,084 input tokens and 71,248 output — identical to the first read: the same bytes, the same count. The responses are committed raw as grown literals in records/read-de4508a.json (sha256 LF 237c9cb744248fb50a056097d20d0059fb9e131029e3b6ee2a3adf486b729b36, 668,083 B), the readout in records/readout-de4508a.json (31ecff7e62b6a8d5808177c659b9c1278ed5d5d9773d0e8a8c291c508a126213) and the drift in records/drift-de4508a.json (c721dd73ad76bf59469a0b3006c847aa2c79b8d840931e55c4d79b74f18e276a); the battery re-witnesses all three from the committed responses alone.

The verdicts. Under refresh every probe landed exactly as it landed the first time: P1 PASS · P2 PASS · P3a PASS · P3b REFUTED · P4 PASS · P5 REFUTED · P6 recorded — R1 THE VERDICTS UNDER REFRESH: HELD. R2 THE DRIFT: SMALL — 145 of 2,576 answers moved (5.6 %); 49 tops changed.

The drift, per question.

The honest readings beside the verdicts (the house's, stated never asserted; nothing about the model):

1. The verdicts hold whole under refresh, the two refutations included: the diagonal rule still at chance, the window over time still refuted on the optimism (0.39 the first time, as refuted now).
2. The slip lives where the frame under-determines. The cue and the letter rule, exact both times, barely moved. The sides of two-sided and unstated-leak records, held at one half, flipped their side of one half 36 times by hundredths — a coin read twice, as P1 says it should be. The counts through a shut window and the hidden cases, where the observer was already beyond the frame, moved the most. An unrefreshed standing on those questions decays into slip; the literal reads hold.
3. The re-witness confirms §16's pattern, and adds its measure: the DRAM law's slip is concentrated exactly on the answers the plausible set did not license.

What is not claimed. As §16, whole. Standing. RESOLVED; R1 HELD; R2 SMALL. The count K stands where the Ruler's design left it.

§18 THE SEAT AND THE NAME (his word 2026-09-22: "go on the seat and the name for both")

The seat. The starl3n persona, adopted: the contributor accountable for the content (RACI A), as the plugin declares and the contrib register shows. Proposed at the design, held through the read and the re-witness, now his.

The name. The Watchmakers' Frame — the working handle the record has carried since the design, and the slug's own word; the fuller phrase the external observer over the watchmakers' frames stays as the description's substance. The candidates kept on the record: The Window and the Count · The Split One · Two-Sided. A rename is a small act of his, at any time: it touches the plugin's title, this section and the chronicle, and nothing sealed.

What changed: the plugin's title and provenance, the README's header, this section, the chronicle. What did not: the engine (pinned at the re-seal, 439f5737…), the manifest (22d1ee55…), the two committed records and their readouts. The name is not in any frame, any question, or any pin. The count K stands unstaked under this name as under the other.