Seven Facts That Look Like One
Once the workshop began uploading its own films, the obvious next status was "published." It is the wrong shape for a status, and the workshop wrote down why. What happens after a render is a chain of separate facts, each owned by a different system, and each able to be true while the next one is still false: the private upload went through; the owner watched it; the film was observed to be public; a copy reached the network archive; the transcription engine registered it; a transcription was accepted; the transcript was approved for the searchable corpus.
Seven facts, not one. A film can be public and not yet archived. It can be registered with the transcription engine and not yet transcribed. It can be fully transcribed and still not approved for anyone to search. Collapsing those into a single word does not simplify them; it hides which one is false today.
Read Receipts You Do Not Own
The workshop owns none of those facts. Uploads and archive copies are recorded by the brain's repository, beside the channel credentials, where a scheduled pass follows each public upload into the archive and on to the transcription engine, and manual passes are serialized with it so two cannot race. Transcriptions are recorded by the transcription engine itself. So the workshop does not store a delivery status of its own. It projects the owners' durable receipts: one command reads the receipts for a single production, and the queue shows each language's state beside its job. When a step has stalled, the workshop asks the owner to resume follow-through for exactly that production's uploads, and nothing wider. Registering a film is recorded as registering, never as transcribing, and corpus approval is never the workshop's to grant.
One more condition protects the link between them. An authored production reference — the script and speaker record that the transcription engine keeps beside its own recognized text — is attached only after both the source movie's hash and the archived copy's hash match, so the reference can never describe a different file than the one transcribed.
Label an Observation as an Observation
The workshop sees the moment a video turns out to be public; it does not see the moment the owner pressed the button. So the time it records is labelled as an observation, and nothing reconstructs a publication time from an upload time. The same discipline governs approval. A film being public is not evidence that the owner approved that exact cut — he publishes in the platform's own studio, outside every record the workshop keeps, and a public video says nothing about which file he watched first. Approval is recorded from what he actually said, bound to the file's hash. And when a stage has to be logged after the fact, it is logged as a backfill with its original evidence, never back-dated to look as if it had been recorded at the time.
The Producer's Own Scope Is a Fact Too
The most uncomfortable fact on the list is the producer's own. A session that renders a film can measure it, sample its frames, and read its subtitles; it may not have listened to every minute. The workshop's records now say so in those words — technical checks and sampled visual review, no full listening claimed — instead of a single "QA passed." When the owner has already published a film and a production is finalized afterwards, the acceptance keeps the producer's actual review scope, incomplete playback included, rather than inventing a completed watch.
The same rule shows up at the lowest level. A voice take's model is recorded from the voice engine's completed job receipt — what actually ran — rather than from the request. And a dub delivered in three languages re-reads every render's bytes in full on every run, so a file changed after validation is caught however it changed.