Whisper runs on the same local-embedding description clusters as Radar’s granular section — specific things like Coding Agent or Autonomous Drone, not broad categories like “Machine Learning” or “Security”, which have existed for over a decade and don’t say much ticking up on their own. It uses Radar’s same idea with a looser, earlier trigger: 3 companies instead of 6, a 1.3× rise instead of 1.5×, +2 companies instead of +3.
PRECISION69%
Lower than Radar’s granular section on purpose — a looser trigger means more of what fires won’t pan out.
BASE RATE61%
What an equally-active cluster does anyway, whether or not Whisper fired.
LIFT1.15×
A real edge over that baseline — actually higher than granular Radar’s own 1.12×.
RESOLVED153
Historical firings since Winter 2018 with a known outcome (106 confirmed).
Roughly 31 in 100 Whisper signals won’t pan out — meaningfully more than Radar’s miss rate, and off a thin sample (there are only ~60 total description clusters). Read this list as “glance at it,” not “trust it” — that trade is the entire point of a faster, noisier tier.