The situation
Nobody. I built Kanopi for LÏEF's own bids because outside takeoff vendors were slow and rate sheets were scattered, a problem I had first solved at Tellus in 2014.
By April 2026 the pipeline priced a plan set in minutes; nobody could say whether to trust it.
A test that separated memory from forecast, a calibration pass against real bids, a takeoff measured from the drawing's own geometry, and the discipline to publish the unflattering result.
In-sample numbers are never an accuracy claim. Only the leave-one-out figure ships, with its ground truth and corpus size beside it.
Held out one bid at a time and priced blind, mean error 18.6 percent across three single-family bids, worst bid 31.9 percent (June 10, 2026), its only accuracy claim. As of September 25, 2026, in place with eight builders and developers; over $600 million in proposals bid nationally in 2026.
Read on
The decision and the reasoning
Why we did it this way, told first.
ReadWhat we did and what it produced
The work, decision by decision, and the result.
ReadA slice of the project list
A few related projects.
ReadPrivate side
The full report, the source files on record, the timeline and every figure with its source. Password.
Enter