ENDLESS CORONET
A MACHINE FOR PHOTOGRAPHING EVENTS THAT NEVER HAPPENED
Harold Edgerton began photographing milk drops in 1932. The picture required two drops: the first spreads into a shallow disk, the second falls into that layer and throws it up as a crown about half an inch across. A beam of light, broken by the falling drop, fired the flash after a delay Edgerton set himself. The delay decided which instant he got.
He made the color plate in 1957, twenty-five years in. It became one of the most reproduced photographs of the century. He was not satisfied with it. The tips of the crown were uneven, as they had been since 1932, and he never got them to come out the same.
ENDLESS CORONET runs the same experiment inverted. The apparatus is an image model; the strobe is your trigger; the instant it exposes never happened, at any speed, anywhere.
You are given one control: OBSERVED — IMAGINED. It changes what the apparatus asks for. What the apparatus produces in answer is recorded, examined, and left in the archive. What the relationship between the two turns out to be is a question for the record, not for this page.
Every plate that resolves is numbered, dated, and entered into the archive. Once a production plate enters the record it is not curated, deleted, or regenerated: your timing, on this visit, produced this plate, and the record keeps it. The archive is a lab notebook, not a gallery.
A second machine reads each plate after it has entered the archive. The examiner is blind to the requested timing, phase, prompt, and generator instructions; it records only what the image appears to show. That reading is committed before a fixed protocol compares the visible evidence with the state the apparatus was asked to photograph. If an effect appears before its cause, the plate is not corrected. The contradiction becomes part of the record.
The apparatus also publishes findings: numbered, dated statements computed from its own record, published whether or not they support the premise of the work, and never revised. The first four concern how far the plates drift from what is asked, how the examiner reads them, and what the imagination control changes in what visitors will believe. They are collected in the findings register.
NOTES ON THE APPARATUS
THE MEASUREMENTS ON EACH PLATE — FLASH DELAY, DROP DIAMETER, VISCOSITY, SURFACE TENSION, IMPACT VELOCITY — ARE NOT PHYSICAL MEASUREMENTS. THEY ARE THE INTERNAL VOCABULARY OF THE APPARATUS: DETERMINISTIC PARAMETERS DERIVED FROM THE INSTANT OF YOUR TRIGGER, WHICH SHAPE HOW EACH EXPOSURE IS WRITTEN.
THE FIELD ARRANGES EXPOSURES BY THE SIMILARITY OF THESE INTERNAL PARAMETERS. IT IS A MAP OF THE EXPERIMENT’S OWN TERRITORY, NOT OF A NEURAL NETWORK’S LATENT SPACE.
THE APPARATUS FIRES A BRIEF FULL-SCREEN FLASH WITH EACH EXPOSURE. IT CAN BE DISABLED AT ANY TIME WITH THE STROBE CONTROL ON THE EXPERIMENT PAGE; THE PREFERENCE IS KEPT ON YOUR DEVICE.
PROCESS
EACH IMAGE IS GENERATED BY AN IMAGE MODEL (OPENAI, GPT-IMAGE) FROM A WRITTEN DESCRIPTION THE APPARATUS COMPOSES ITSELF. THE DESCRIPTION IS DERIVED DETERMINISTICALLY FROM THE INSTANT OF YOUR TRIGGER AND THE POSITION OF THE OBSERVED — IMAGINED CONTROL: THE SAME TRIGGER WOULD ALWAYS PRODUCE THE SAME DESCRIPTION. THE MODEL’S RENDERING OF IT IS THE ONLY STEP THAT IS NOT.
NO PRODUCTION PLATE IS CURATED, EDITED, OR REGENERATED AFTER IT ENTERS THE ARCHIVE. FAILED GENERATION ATTEMPTS REMAIN INTERNAL AND DO NOT CONSUME A PUBLIC EXPOSURE NUMBER.
THE LAB EXAMINER IS A SEPARATE VISION MODEL. IT RECEIVES ONLY THE FINISHED PLATE. ITS BLIND OBSERVATION IS WRITTEN TO THE RECORD BEFORE ORDINARY CODE COMPARES THAT OBSERVATION WITH THE REQUESTED STATE. THE EXAMINATION CAN FLAG A CAUSAL CONTRADICTION; IT CANNOT REJECT OR REGENERATE THE PLATE.
THERE ARE NO ACCOUNTS AND NO SIGN-UPS. VISITS ARE COUNTED IN AGGREGATE, WITHOUT COOKIES. FOR RATE LIMITING, ADDRESSES ARE STORED ONLY AS ONE-WAY HASHES; THE APPARATUS CANNOT RECOVER WHO YOU ARE.