Workshop note · AI and automation
What the automation here is not allowed to do
Automated processes fetch the data, normalise it, weigh matches and prepare decisions. They are not allowed to publish anything. That boundary is not a statement of intent but something built into the system.
- Revised
- 2026-08-06
- Topic field
- AI and automation
- Figures in the text
- read from the database at request time
The useful question about an automated system is not how much it can do but what happens when it is wrong. A system that publishes when in doubt produces errors that reach the outside and that somebody has to retract. A system that stops when in doubt produces work — but the work stays internal.
This network is built on the second option. The import of the 367 catalogue records writes internal drafts and nothing else. No step of the processing has a path that ends in something becoming public.
Four places where the system stops
- stop
Import
- stop
Matching
- stop
Identity
- stop
Media
The price of this stood visibly on the site for a long time: open matches, an undecided question of identity, not a single image. On 5 August 2026 the owner decided, and three of those gaps closed — not because the system closed them, but because a person was handed the evidence and could read it. That is what the waiting was for.
What stays open is what cannot be decided without a second piece of evidence: one match with no counterpart on the platform account, and the image work outside the covers. Those gaps could be closed in minutes by letting the system decide. The result would be a more complete site whose statements could no longer be trusted — and that trade is not worth making.
Automation here is a tool for preparing decisions, not for releasing them. It makes sure that when a person decides, the decision is well prepared: the evidence, the contradictions and the case against it, all on one sheet. What is currently pending is listed under provenance.