Time Series
Forward targets, look-ahead leakage, horizon decay, and why a single series gives you far fewer independent looks than it seems.
Forward targets and look-ahead leakage
A target is built from what happens after the row's timestamp; a feature from what was known at it. The six ways future information leaks into a feature, a one-line detector for the worst of them, and a deliberately leaked example that scores absurdly well.
Single series or panel: where the repeats come from
A panel gives hundreds of independent looks per day; a single series gives one. Why the same question is far harder to answer at the aggregate level, and what to use as the repeats when there is no cross-section.
Feature decay across horizons
The same signal scored against forward outcomes of increasing horizon, cumulative and day-by-day, with the overlap correction applied. Two examples with opposite shapes, sign versus size.