one question per workspace

An open question is the easiest thing in a research operation to get wrong in a way nobody notices. The failure is rarely a wrong answer on the day. It is contamination: a dead end in one line of inquiry quietly becomes an assumption in the next, and by the time anyone looks, the record no longer says which parts were tested and which were merely inherited.

The decision here was blunt. Every question got its own isolated working area and its own history. Nothing about one question's state was visible from another's. Reconciliation into a single shared record happened only as an explicit act performed by a coordinator, never as a side effect of one worker saving in place.

That choice is expensive and it looks wasteful from outside. Setup is paid per question instead of once. Groundwork gets redone. Reconciliation steps are, by volume, the most tedious work in the entire history, running to hundreds of small integrations against a much smaller number of decisive ones. The tedium is the mechanism working. An operation that only reconciles when something dramatic happens has no idea what it has dropped.

What the discipline cost in speed is not something this record can price. What it bought is visible in the shape of the history instead: many lines of inquiry ran at once, each free to fail privately, and the shared record stayed legible enough that a single coordinator could still say what was known and what was not.