Breakout

What just shifted regime?

ProductionVerdict · PARTIALRecall · 0.34

Supporting · the Moves chapter (regime distance) · anchor metric: regime distance (d²)

What it measures

Flags instruments whose recent behaviour has broken from their own established regime — a Mahalanobis-style distance between the current short window and the trailing distribution. Short forward window (5 sessions) because a regime break is, by definition, a fast event.

\mu,\ \Sigma = \text{mean, covariance of features over trailing window}

d^2_i = (x_{i,t} - \mu)^{\top}\, \Sigma^{-1}\, (x_{i,t} - \mu) \quad\text{\# regime distance}

\text{rank desc by } d^2 \,;\ \text{forward\_window\_test} = 5 \text{ sessions}

A large d² says the instrument is no longer behaving like its recent self — the relationships that held are breaking. It's a state-change signal, distinct from momentum (speed) and stretched (level).

Limitations

15% pass-rate — usable only for clean regime-shift events; noisy when covariance is poorly conditioned.
Variant overlap — a CUSUM drift-detector variant (breakout_cusum) runs experimentally alongside the primary distance method.

In the product

Right column of the Moves chapter ("What just shifted regime?"). Ranks by regime distance, not speed.

Grading & stability

—

Gate rate
loading…

—

Stability / churn
not emitted

—

Mean confidence
loading…

fetching /api/panel/about/lens_engine.breakout/W/all …

Graded on the Tier-1 regime-test harness — curated macro events × 18-yr backfill × recall@K. Verdict PARTIAL (15%).

Charts it reads

Major assets

major_assets—

FX majors

fx—

Bond ETFs

bonds—

Breakout · The Lens methodology← all engines