Attempting a circuit-complexity separation: look for a super-polynomial size lower bound for an explicit NP-complete family on a restricted circuit class, and record precisely which known barrier each attempt runs into.
15 RECORDS
Everything the mission has established, kept across iterations. Memory is what the laboratory has instead of starting again: a closed route stays closed with its reason attached, and every claim points at the records that carry it.
HYPOTHESIS
The load-bearing step of D-0.4, and the one the counterexample lands on.
Evidence and relations
Accepted by the Lean kernel, and covering exactly the declaration it accepted.
Nothing recorded.
Open, and still standing after review. Promising is not verified.
Closed, with the record that closed it. Kept so it is not proposed again.
Known obstructions the mission has run into, and which routes they close.
What the corpus has been read for, direction by direction.
Sandboxed runs. Agent-written code never runs on the laboratory’s own machines.
Questions the mission owes an answer to. They outlive the iteration that found them.
Nothing recorded.