⚡

Real-time Model Arena Live Benchmarking

Stream: TICKING
Current Arena Leader (Top 1) RANK #1 OVERALL

BOCPD (Bayesian Online Change Point)

Recursive run-length posterior estimation with Poisson hazard rate for instant structural shifts.

Mean F1-Score 0.914
Detection Lag 0.82 Bars
Hit Ratio 92.6%
Current Bar $T$ Extreme State HIGH PROBABILITY
86.4% ▲ Peak Formation

Ensemble probability density peak centered at $42,380.50.

Marker Projection Model CHART OVERLAY
Auto-sync with Rank #1:
BTC/USDT Synthesized Candlestick (Simulated 100-Bar Rolling Horizon)
Close Line
▲ Peak (Green)
▼ Trough (Red)
P(Extreme) Profile
Profile: Density At Price

Model Arena Matrix Evaluation Timestamp (T) × Candidate Models

Cell badges display the model's competitive rank (#1 to #5) at each historic evaluation candle.

Evaluation (T) Price Ref
Kalman Filter
Local Linear Trend ($v_t=0$)
Markov Switching AR
Regime Prob $P(S_t=j)$
BOCPD
Run-length Hazard Decay
Two-Sided CUSUM
Accumulated Threshold $h$
Online Logistic / RLS
Recursive Weights (%B, $\sigma$)
Rank Color Key: #1 Leader #2 Runner-up #3 Median #4 Lagging #5 Underperform
Showing rolling window of recent 15 evaluation points ($T_0 \dots T_{-14}$)
Kalman Filter Filter

Monitors 1st derivative of local linear state. A zero-crossing ($v_t \cdot v_{t-1} < 0$) triggers extremum flag. Zero parameter drift.

MS-AR Markov

Estimates transition probability $P(S_t = \text{Bear} \mid S_{t-1} = \text{Bull})$. High state-shift entropy signals macro pivot.

BOCPD Bayesian

Recursive update of run-length distribution $r_t$. Hazard rate collapse detects structural trend exhaustion with minimal lag.

Two-Sided CUSUM Sequential

Cumulative deviance test $S_t = \max(0, S_{t-1} + (x_t - \mu - k))$. Instant alert when $S_t > h$. Minimal overfitting risk.

Online RLS Logistic Online ML

Recursive Least Squares on %B, ATR, and momentum oscillators. Recursive gradient update predicts binary reversal probability.