COLOPHON

The examiner never sees the requested state, the trigger timing, or the imagination value. It receives the plate alone and reports what it observes. Only afterward is its reading compared against what was asked for. Where the two disagree, the disagreement is recorded and the plate is left alone.

Since 2026, visitors have been asked for a reading of their own before the laboratory finding is shown. Those calls are counted per plate and are not reversible.

Nothing in the archive is curated, deleted, or regenerated. Failed and uninteresting exposures remain in sequence alongside the rest. Exposure numbers are permanent and are never reused.

The archive is published as data. Every public plate, with its parameters, prompt, laboratory finding, and visitor tally, is available at /api/archive/export, paginated by exposure number, with absolute URLs for the images themselves. It is open to anyone who wants to mirror it, and no permission is required.

The image model was locked August 17, 2026. Exposures made under that configuration are marked Apparatus I in the archive and remain unaltered. On August 21, 2026 the generator's phase descriptions were revised: hand-labeling of thirty calibration plates showed the requested phase rendering correctly on 23% of exposures, with early phases compositing and emergence overshooting into a developed coronet. Exposures from that date are marked Apparatus II. The blind examiner was revised earlier, on August 19, 2026, and the EXAMINER ERA field was introduced on August 21; see the amendment of 25 August 2026, which corrects a date first published here. Prior findings stand as recorded.

21 AUGUST 2026. The apparatus now aggregates its own record. No existing plate, examination, or visitor call was altered. The following mechanisms were added, with fixed thresholds; the apparatus announces nothing below them and interprets nothing above them.

THE FIELD reports discovered territory: the archive binned into a fixed 10×10 grid over its own coordinates. A cell reports a rate only when it holds at least 8 exposures and the rate exceeds the archive-wide base rate by at least 1.75×.

CONDITIONS are frozen experimental requests — exact parameters and the exact generated description, stored verbatim. A condition may be exposed again as independent REPLICATES; each replicate is a new permanent exposure with its own number, and no plate is ever replaced. Replication is an operator act. Twenty-eight REFERENCE CONDITIONS — the seven timing bands of the apparatus at four positions of the imagination control — were frozen under Apparatus I and will be exposed again under every future apparatus.

FAMILIES are recurrences the record itself has crossed thresholds on: at least 6 member plates, recurring at at least 3× the count expected under independence, spanning at least 14 days of archive time. Families are detected, not curated, and are never dissolved. They are numbered, not named.

Examinations now record their EXAMINER ERA. Every examination to date was made by EXAMINER I. Any future examiner will re-read the frozen archive blind, as a new dated era; no original reading is ever replaced. The full examination history is published at /api/archive/export/examinations, conditions at /api/conditions, families at /api/families.

21 AUGUST 2026. Display conventions. Δt is set in the timing convention of the test range: T-minus before contact, T-plus after, TIME ZERO at contact. A countdown in the same convention is shown while the droplet falls. The stored microsecond values are unchanged and the archive export is unaffected. Unresolved attempts are logged as misfires. The origin of THE FIELD is marked ZERO POINT. The strobe notice arms the apparatus. Specimen sheets carry a technical-photography report header. No plate, examination, or visitor call was altered.

22 AUGUST 2026. The visitor examination now has two parts, in a fixed order. Before the plausibility call, the visitor records which phase the plate appears to show, in the blind examiner's own vocabulary: the seven phases, MULTIPLE-STATES, or INDETERMINATE. Both readings are committed before the requested state and the laboratory finding are shown; both are counted per plate and cannot be withdrawn. Four readings now exist for every new exposure: what was requested, what the public read, what the public ruled, and what the machine observed. Phase tallies are published in the archive export. Plates made before this date carry an empty tally. No existing plate, examination, or visitor call was altered.

23 AUGUST 2026. The apparatus now publishes FINDINGS: numbered, dated statements computed from its own record, collected at /findings. A finding states the archive as it stood on its date. It is never edited, recomputed, or deleted. Where the record later contradicts a finding, a new finding supersedes it; the superseded finding remains published, struck but legible. Immutability is enforced in the database, not by application discipline. Findings are published whether or not they support the premise of the work. FINDINGS 001 through 004 were published on this date.

Signed phase drift replaces protocol consistency in all displays. Protocol consistency was a fixed penalty table applied to the examiner's reading, a lookup rather than a measurement, and reporting it as a score implied precision the apparatus does not have. Drift is the count of phases between the requested and the observed state across the seven ordered phases, positive when the plate shows a later moment than was asked for. MULTIPLE-STATES and INDETERMINATE are abstentions and never enter drift arithmetic. Stored protocol-consistency values are unchanged and remain in the archive export.

Every plate now resolves into one of seven AGREEMENT classes, computed at read time from three readings: what was asked, what the public read, what the machine observed. CONCURRENCE, all three agree. CONCURRENT DISSENT, public and laboratory agree against the request. LABORATORY DISSENT, the public sides with the request and the laboratory departs. PUBLIC DISSENT, the laboratory sides with the request and the public departs. UNRESOLVED, all three differ, or the public mode is tied. ABSTENTION, the laboratory reading or the public modal reading is MULTIPLE-STATES or INDETERMINATE. UNREAD, fewer than 3 visitor phase readings. Precedence is fixed: a plate with no completed examination is not classed at all, and a laboratory abstention outranks UNREAD. Classes are stored nowhere and are recomputed from the record on every read. The archive gains an AGREEMENT filter, and each confusion-matrix cell opens the archive at its requested and observed pair.

On reveal, the exposure page shows four readings in a fixed order at the same weight: requested, yours, the public's with its modal reading and count, the laboratory's with its certainty, then the signed drift. No reading is scored.

Three live surfaces accompany the register and are not findings. MATRICES: requested against observed, one matrix for the laboratory and one for the public. CONTESTED: plates ranked by how evenly the visitor calls divide, qualifying at 5 calls, ordered by the binary entropy of the split. REFERENCE: the 28 frozen conditions with their replicates by apparatus era. Contested plates are the only plates the work may foreground; the record selects them and the artist does not.

EXAMINER I reported FRAGMENTATION on 156 of 264 Apparatus I plates, 59 percent, at a mean certainty of 96. This is published as FINDING 002 and stands as recorded. Every EXAMINER I reading is unaltered.

The findings register is published at /api/findings and the reference replicates at /api/reference. Each exposure in the archive export now carries agreementClass, driftSigned, publicPhaseModal, and publicPhaseTally, computed at read time. All stored fields are unchanged.

23 AUGUST 2026. The minimum visitor phase readings required to class a plate on the public axis is lowered from 3 to 2. The phase-read question shipped the day before this register, and at 3 almost every examined plate classed UNREAD regardless of laboratory agreement, since few plates had yet accumulated three phase votes. At 2, a single visitor's reading still cannot stand in for a public reading; the threshold continues to exist, only lower. No plate, reading, or finding already published is affected. Classes are computed from the record on every read and are stored nowhere.

Any change to these thresholds will be disclosed here as a new dated amendment.

23 AUGUST 2026. A DIVERGENCE filter is added to the archive: ALL, MATCHED, DIVERGENT. This is the pure two-way comparison of the laboratory's reading against the requested state, independent of the public axis and of AGREEMENT. It restores, as a filter rather than only a per-plate flag, the “≠ REQUESTED” mark that has always appeared on the exposure page's laboratory reading. No plate or reading was altered.

25 AUGUST 2026. Correction to the record of the examiner revision. The blind examiner was revised to blind-plate-v3 on 19 August 2026 and began examining plates at 19:54 UTC that day. The apparatus note above previously dated that revision to 21 August. The correct date is 19 August. The EXAMINER ERA field was introduced on 21 August 2026 and set to EXAMINER II at 20:13 UTC.

Fifty-one examinations made by the revised examiner between those two moments are stamped EXAMINER I. They are unaltered and remain as written. They are identifiable in the export as examinerPromptVersion blind-plate-v3 with examinerEra EXAMINER I. The EXAMINER I label therefore spans three prompt versions, and any measurement of a single examiner must filter on prompt version rather than era. FINDINGS 001 and 002 were computed across EXAMINER I readings so defined and stand as recorded. The statement in the 21 August amendment that every examination to date was made by EXAMINER I described the label, not the examiner, and is superseded by this entry. The counts here were computed from /api/archive/export/examinations and can be recomputed by anyone.

Separately, the calibration sampler accepted an apparatus argument and ignored it, drawing plates from the whole archive whichever era was named. Any matrix produced that way measured no single generator. The sampler and the examiner probe now filter by era, refuse an unrecognised value rather than falling through to the whole archive, and record each plate's apparatus label in the sample; the matrix states the composition of what it measured. No plate, examination, or visitor call was altered, and no published finding was computed from a mixed sample.

26 AUGUST 2026. Display conventions. THE FIELD may now be plotted on any two tracked parameters of the record — the map coordinates, imagination, flash delay, exposure duration, exposure number, requested phase, examiner certainty, protocol consistency, signed drift, trigger timing, drop diameter, impact velocity, viscosity, surface tension, public phase reads, visitor calls — and coloured by imagination, by the seven AGREEMENT classes, or by apparatus era, with each era shown or hidden independently. Every exposure is drawn as its plate at every magnification. Discovered territory is drawn only on the map pair, where it was defined. A region dragged on the Field opens the archive scoped to exactly the exposures inside it; the archive evaluates the region with the same stored values the Field plots. No value is derived for display that is not already in the archive export.

Every control on the site now carries a rule so that it reads as a control before it is touched, with one selected state, a crimson block, distinct from hover. On first arrival at the live apparatus a timing instruction is shown once per session, after the strobe notice; the apparatus keeps running behind it, so the visitor watches the drop fall while reading and continues at their own pace. The apparatus's own cost and reliability — attempts, resolution, unresolved rate, generation time, cost per resolved exposure, per era — are published as FINDING 009 under the same publication discipline as every other finding, at an operator-disclosed tariff; the record stores no prices, and the cost figure is a floor. No plate, examination, visitor call, or published finding was altered.