NOSTRADAMUS

SUPERVISOR

INDEPENDENT LEARNING & CALIBRATION WATCH

01 / INTELLIGENCE VERDICT

ANALYZING

Comparing chronological official outcomes without look-ahead.

—EVIDENCE STRENGTH

OFFICIAL RECORD—— accuracy

RECENT HOLDOUT—latest 0 resolved calls

ACCURACY CHANGE—recent vs previous window

BRIER CHANGE—lower is better

AUTONOMOUS ENGINE—0 successful runs

02 / LEARNING CURVEEQUAL HOLDOUT WINDOWS

PREVIOUS 0—

RECENT 0—

METRICPREVIOUSRECENT

Accuracy——

Brier score——

Calibration error——

High-confidence misses——

Probability of the recent accuracy being higher: —. A streak alone is not accepted as learning.

03 / THINKING-TIME VALUEFIRST 1M vs 2:30 LOCK

—VALUE OF WAITING

FIRST-MINUTE CALL—

2:30 LOCKED CALL—

Paired on the same 0 official cycles, so missing data cannot make one timing rule look better.

04 / CALIBRATIONPREDICTED UP vs ACTUAL UP

Loading calibration bands.

MODELACTUAL

05 / WHAT IT LEARNED—

Official samples—

Effective local sample—

50/50 correction—

Direction guardREADY

Loading model identity

06 / REGIME & FAILURE LABWHERE NOSTRADAMUS BREAKS

07 / VERSION SCOREBOARDNO RETROACTIVE CREDIT

MODELNW/LACCURACYBRIERCAL ERROR

08 / POST-TRADE INVESTIGATIONSWHAT EACH RESULT TAUGHT

RESULTREGIMEDIAGNOSISPROCESSRECORDED LESSON

Structured official investigations will appear after the background research sync records resolved cycles.

09 / PROMPT LABEVIDENCE-BASED NEXT EXPERIMENTS

These prompts ask Nostradamus to research and shadow-test one bounded change. They do not let it edit production based on one streak or one loss.