Sift
The first act on the change, and the first that requires a role. Capture is open to anyone; the sift is not — someone accountable rules on what arrived. Two things settle here, and only two.
What kind of thing it is
The request is given a type, or the type it arrived carrying is verified. The ordered test is on work types, and it always resolves — there is no request the practice cannot type.
Whether it will be done at all
| Outcome | Means |
|---|---|
| Accepted | Real, and not already known. It becomes work, carrying a type and an agreement that it will be done |
| Duplicate | Already known. It attaches to the existing item rather than starting a second one |
| Rejected | It will not be done — recorded with a reason, so the same request does not arrive again unanswered |
The sift does not shape the work. It does not work out what right looks like, how the work will be done, or what it will cost. Nor does it decide when, or by whom, or how urgent it is. It settles three questions only: is this real, is it already known, and what kind of thing is it.
The count of unsifted requests is a query, not a report. The requests waiting at New
are the unsifted work, so the number cannot be rounded down by anybody having a bad week. A
rising count is a fact about the rate of sifting against the rate of arrival — never a fact
about whoever is capturing, and reading it that way reintroduces the cost at the door that
free capture exists to remove.
Capture carries no quality bar, so every bit of variance it lets through arrives here. A cheap door with an expensive sift has not removed the friction — it has moved it somewhere less visible, and the queue grows instead of the backlog.
If answering "is this real?" needs investigation, that investigation is itself a question. Raise it as work; do not perform it inside the sift. The sift rules on what is in front of it, and hands off anything that cannot be settled in the moment.