The market regime classifier

Almost every rules-based growth system starts from the same observation: most stocks follow the general market, so exposure should follow market health. The app makes that operational with a computed market outlook — one of four states, recalculated daily from index price and volume data:

1. Confirmed Uptrend — the market has confirmed strength; the rules permit full exposure.
2. Uptrend Under Pressure — institutional selling is accumulating; new buying gets selective, sizing lighter.
3. Rally Attempt — a correction has stopped falling and is trying to turn; watch for confirmation before committing.
4. Market in Correction — the market is in a downtrend; the rules argue for cash and defense.

The mechanics

The classifier runs over daily bars for SPY and the Nasdaq Composite and combines the two (the more cautious call wins). The moving parts:

• Distribution days — a decline of 0.2%+ on volume heavier than the prior session: the footprint of institutions selling. The classifier counts them over a trailing 25-session window; a cluster (five or more) is how uptrends come under pressure and die.
• Rally attempt — after a correction, the first up day off the low starts the count. The attempt survives as long as the low holds.
• Follow-through day — the confirmation signal: on day 4 or later of a rally attempt, a gain of 1.25%+ on volume up from the prior day. Not every follow-through works, but historically sustained uptrends have started with one.

The same classifier drives the Summary outlook, the regime shading on the equity chart, performance attribution, the buy checklist’s market check, and practice mode’s answer key — one implementation, consistent everywhere.

Caveats — read this part

• It’s a labeled proxy. The thresholds above are our implementation of publicly documented ideas. It is not IBD’s official market call, will not always agree with it, and everywhere the app shows the outlook it is labeled as computed.
• It’s reactive by design. Regime systems confirm turns after they start; they trade some upside for not being wrong at the extremes. Expect it to be “late” at every bottom — that’s the design, not a bug.
• Whipsaws happen. Follow-through days fail; corrections resume. The classifier changes its call when the data changes, and so will you.