redirecting the searchers

Computational search has a property that makes it dangerous to a research queue: it always returns something. A worker extending a boundary can report a new record every cycle, and the record is real — it is simply not an answer. Left unmanaged, a large group of workers will optimise for whatever reports back.

The situation read plainly off the board. Several lanes were extending ranges while the questions they nominally served still had open, unproven steps in front of them. Searching that keeps succeeding at searching is not the same thing as a question being advanced, and the two look identical in any throughput measure.

The coordinator's move was blunt, and it is recorded as a structural change rather than a nudge. Those lanes were reassigned to proof work, and the search assignments were rewritten instead of paused. A pause leaves a worker idle and a resumption path open; a rewrite changes what done means for that lane, which is the part that actually stops the behaviour.

The choice has a price that is easy to understate. Search produces artifacts — tables, ranges, records — that read like evidence. Proof produces closure or nothing at all, so a lane moved to proof may report nothing for a long stretch while costing the same as the lane producing records. The operation spent measurable throughput on a category of work that often returns empty.

Because the redirection was written down, it can be audited later: who was moved, off what, onto what. What it proves is that the operator can tell activity apart from advance, and is willing to buy silence from a lane when silence is the honest output.