AI
Four small tools that are supposed to connect. This desk is where each one says what it does and, more usefully, what it refuses to be.
Right now it mostly says what it is waiting for, because none of the four has a spec or shipped code in scope.
The four tools
One page each, with what each does and what it refuses to be, taken from each tool's own notes. All four are currently open lines in the fact ledger.
How they connect
The intended shape: fixturegen feeds evalset, runlog shows the run, skilldiff catches drift. That is a statement of intent, and it is labelled as one until there is code to check it against.
What is decided
One thing on this desk is actually on the record rather than intended: evalset grades with deterministic graders and not with a model acting as judge.
The decision, and what it cost
Changelog
Dated entries, starting from the first one there is a source for.
What is live
None of the four. Evalset, fixturegen, runlog and skilldiff are not live, so they are not linked. Rubric and Skillfile, which are adjacent to this work, are built but not live either.
There is nothing to click here yet, and a page of buttons that go nowhere would be worse than saying so.