Predictions
This page shows every prediction our statistical models have made. Each prediction says whether a stock will go up or down, with a confidence score. After the target date passes, we grade it against what actually happened — that's how the engine learns and improves.
0
Total Predictions
0%
Accuracy
—
Best Model
0
Current Streak
Where we have edge — per market regime
Reliability = how close stated confidence matches realized accuracy. Brier ≈ 0.25 means coin-flip; lower is better.
We don't pretend to beat every regime. The publish gate adapts: trend regimes show calls at lower confidence (we have edge there); chop regimes hide everything below a higher gate (humility tax).
trend low vol
gate 55%
30.7%
accuracy (153)
brier 0.000
reliability 73
trend high vol
gate 60%
38.5%
accuracy (247)
brier 0.000
reliability 84
chop low vol
gate 65%
47.6%
accuracy (124)
brier 0.000
reliability 96
chop high vol
gate 70%
43.6%
accuracy (257)
brier 0.000
reliability 91
unknown
gate 65%
—
no graded calls yet
Calibration — does X% confidence actually mean X% right?
Gap = realized − midpoint. Positive = under-confident, negative = over-confident.
50–59%
44.4%
-10.6 pp
207 graded
60–69%
36.2%
-28.8 pp
254 graded
70–79%
44.4%
-30.6 pp
63 graded
80–89%
—
no graded calls yet
90%+
—
no graded calls yet
Showing only high-conviction calls. The publish gate is per-regime — see the reliability card above for each regime's threshold. Low-conviction calls are still tracked for learning, but hidden from the public board to keep the signal clean.