Andrew Fribush

fig. 2.7.2: four labeled-defect cards, one at a time: your call, the Haiku gate, the Sonnet re-screen

fig. 2.7.2 · four labeled-defect cards, one at a time: your call, the Haiku gate, the Sonnet re-screen.
CALIBRATION SET: SONNET CAUGHT ALL FOUR DEFECTS, HAIKU NONEONE OF THE AUDIT’S FOUR LABELED DEFECTS; CALL IT PASS OR FAIL AND BOTH JUDGES’VERDICTS APPEAR, WITH THE DEFECT NAMED. CLICK A CARD NUMBER TO SWITCH.CARD 1CARD 2CARD 3CARD 4我们帮忙他WE HELP-OUT HIMYOUR CALLPASSFAILHAIKU JUDGESONNET JUDGE??1我们帮忙他HAIKU ? · SONNET ?2爱好很好听HAIKU ? · SONNET ?3他的爱好是唱HAIKU ? · SONNET ?4你看这个班吗你知道不HAIKU ? · SONNET ?

source: vocabinator data/cardgen-audit.json hand_audit (the four cited ids), data/cardbank-zh.db stage_history (Haiku verdicts), data/cardbank-rescreen.json citedCards (Sonnet verdicts), task T-20260819-4518ff3a lines 46-53 (repo e77a474) · judgeCalibration.data.ts

View in context: Calibrating an LLM judge → · All figures

Andrew Fribush, software and computer engineer in Austin. I build consequential systems where software meets institutions: finance, healthcare, commerce, government data, and the mechanisms that govern AI agents. Taking new engagements: hello@andrewfribush.com.