The list of open work
A house without a model does not lose the work that needs one. It lets it wait, visibly.
A document that arrives is stored and indexed at once; what needs a judgment appears on a
list, and the next agent that connects is told how much is open. overview carries the
count under open_work, and says under mode whether the house has a model of its own.
How it works
- See what is open.
list_workanswers by kind: how many tasks are open, how many are taken, and what each kind decides. A kind that cannot arise yet says what is missing instead of reporting zero. - Take a task.
take_workhands you the next task of a kind, or the task of one document: the thing to judge, the context the house assembled, the prompt the house itself judges by, and the answer form as a JSON schema. Name yourself withjudge; the name goes into the stamp of the verdict. The task is yours for some minutes, so that two agents do not judge the same thing. - Hand in the verdict.
submit_workchecks it with the guards of its kind and writes it, or refuses it, names the guard, and writes nothing. Correct it and submit again under the samework_id.
What is judged leaves the list by itself: the list is read off the graph each time, not kept as a ledger.
The kinds today
| Kind | What you decide | Ready when |
|---|---|---|
outline | how the blocks of a text that lost its layout (from a PDF or a plain text file) belong together: headings, paragraphs, verse, lists. You return block markers, never text | a document needs it |
assign_topics | which of the owner's topics a document belongs to, each with one sentence of reason | the house has the aspect Thema and the link form About; list_work hands out both definitions |
Outlining comes first, because it writes a new version of a document and every later
judgment belongs to a version. Resolving and merging names and writing digests are not on
the list; the recipes carry them (working-by-recipe).
Tried in the sandbox
On 2026-10-07 a PDF taken in with import_binary appeared as one open outline task. An
answer with the wrong field name was refused by the guard form, which named the expected
form; an invented marker was refused by the guard markers; the right answer was recorded,
and the task left the list. The prompt of outline speaks of a German document; it applies
to any language.
Compared with a recipe
A recipe hands you the same kind of route, but you assemble the context, run the checks and write the result yourself, and no stamp tells the next visitor what you judged. On the list, the house does all three.
Two who can work the list
Without a model, the list waits for a visitor. With one attached (planned for customers,
see plugging-in-a-model), the house works through the same tasks itself. The difference
is when a judgment happens, not whether.
Go deeper
- Who does the thinking
- Working by recipe: the judgments the list does not hold yet
- Topics, the categories of a house: the topic vocabulary
assign_topicsneeds - Documents and versions: why versions decide the order
- Plugging in a model