A forecast model is only as trustworthy as its calibration record. TIDAL Drift is POSEIDON's
public scoreboard: every backtested prediction, every outcome that resolved, plotted against
expectation. If we say 70% and we're right 70% of the time, we're calibrated. If we drift, you'll
see it here first.
Hold-Out Games
—
2025-26 season · 2025-10-21 to 2026-06-04
Overall Hit Rate
—
vs — home-team baseline
Brier Score
—
vs — uninformed baseline
High Confidence Hit Rate
—
picks ≥ 60% confidence (n=—)
Calibration Curve
// Predicted vs actual · 10 buckets
Perfect calibrationPOSEIDONDot size ∝ sample size
When POSEIDON predicts 60-70% probability, actual frequency is — across
— games. When we predict 70-80%, actual is — across
— games. The closer our line tracks the dashed diagonal, the better-calibrated our forecasts.
By Confidence Tier
// Hit rate by predicted probability range
≥ 70%
—
n=—
≥ 60%
—
n=—
All picks
—
n=—
By Market
// Each market graded honestly
MoneylineA
—
2025-26 hold-out
SpreadB
derived
not backtested
TotalC
derived
playoff bias
Margin RMSE: — pts · Total RMSE: — pts.
Totals systematically over-predict in playoffs — graded C accordingly.
Below this line · DEMO DATA
Live tracking activates with TIDAL Drift. The metrics below show plausible example data
labeled DEMO so you can see what the live tracking will look like once we begin collecting
day-by-day predictions and outcomes. None of these numbers are real.
Last 30 Days
// Daily picks resolved · most recent first
Wins
18
Losses
11
Pushes
1
Hit Rate
62.1%
Calibration Drift
// Recent calibration vs historical baseline
When live tracking is active, this panel will show whether recent predictions are
drifting from the historical calibration baseline. Sample drift indicators below.
7-Day Brier
—
live tracking pending
30-Day Brier
—
live tracking pending
Activates with TIDAL Drift: live performance tracking, downloadable JSON of every prediction
and resolution, and automated drift alerts when recent calibration deviates significantly from
historical baseline. The demo data above is replaced with real numbers the moment we go live.
Calibration · does the math match reality?
// 2025-26 hold-out · — games
Predicted vs Actual Win RatePerfect calibration = diagonal
Each dot is a probability bin. X: model prediction. Y: actual outcome. Dot size = sample count. Closer to diagonal = better calibrated.
Confidence-Stratified Hit Rateselectivity edge
At >60% model confidence we hit —; at >70%, —. Selectivity is the foundation of profitability.