Calibration
When you say “70% probability,” do things actually happen 70% of the time? This is the track record of your probabilistic predictions — scored with Brier (lower is better) and visualized as a calibration curve.
Total forecasts
5
Resolved
0
Brier score
0.000
(none resolved)
Bias
+0.0%
underconfident
Forecast ledger (5)
NBIS FY2027 revenue >= 9B USD
55%
unresolved
CBRS FY2026 core revenue >= $850M (guide floor)
62%
unresolved
MU FY27 non-GAAP EPS >= $95
58%
unresolved
CRWV remains FCF-negative through FY27 (no positive annual free cash flow)
65%
unresolved
A second pure-play PIM acquisition by NVIDIA, AMD, or a hyperscaler within 18 months of Qualcomm's UPMEM deal (by 2026-12-31)
30%
unresolved
How calibration works
Every forecast is a probability assignment. When one resolves, it contributes (p − actual)² to the Brier score. Lower is better; perfect prediction is 0.0, coin-flip is 0.25. The calibration curve binning shows whether your 70% forecasts actually resolve true 70% of the time. Record forecasts via scripts/research/forecast.ts, resolve them as outcomes land.