The detectors we route to, scored against real decisions made by people who reviewed the same images and videos. Only measured numbers, with the sample size and a range, and only from teams that agreed to share.
| # | Detector | Reviews | Right when it flags | Catches | Score range (F1) | F1 |
|---|---|---|---|---|---|---|
| - | BitMind | 0 | Not enough reviews yet (0 of 200) | |||
| - | Hive | 0 | Not enough reviews yet (0 of 200) | |||
| - | Sightengine | 0 | Not enough reviews yet (0 of 200) | |||
Of the files it called AI, how many people agreed were AI. Low means false alarms.
Of the files people said were AI, how many it caught. Low means misses.
One score from both. The range shows how sure the number is: fewer reviews, wider range.
Call Sightengine, Hive and BitMind through one key, set the threshold per check, and see every result next to its cost.