Why we disabled BALANCED
The initial signal for the BALANCED mode did not confirm on the prospective holdout, so it stays off in v0 rather than shipping on a hunch.
Blog
What we learned building AOPX.
Short, dated notes from the evidence trail — what worked, what didn't, and why.
Blog
What we learned building AOPX.
What 300 paired provider calls taught us
Running SEARCH providers head-to-head on the same fresh missions surfaced failure modes that no single-provider test would show.
Why outcomes matter more than recommendations
A recommendation is a guess until an agent reports what actually happened. Production outcomes are the only thing that moves the public evidence ledger.