Updated every evening at 17:00 IST. The headline metric is rolling 30-day Brier; we also publish hit rate by horizon, directional accuracy, and the individual agent leaderboard.
Each day's 1-day call (purple line + shaded 80% band) against the actual close that landed (white dots). Dots that fall inside the band are calibrated; dots outside are honest misses.
Every graded day since launch. Click a date to open that morning's full article.
Brier measures the squared distance between forecast probability and outcome. Lower is better; a perfectly calibrated 50% forecast scores 0.25.
Cone width is calibrated, not nominal: a 1-day call has a tighter band than a 1-month call. We measure hit rate at each horizon independently.
Each agent's individual track record. Outliers and contrarians are kept in the swarm even when their Brier is poor — fat-tail protection.