Methodology

We deleted a strategy that looked great. Here's why.

Jun 1, 2026 · 5 min read

← All articles

It's easy to add trading strategies. It's hard — and more valuable — to delete them. One of the cleaner-looking setups we tested, pullback-to-VWAP, didn't make it into TradeScaner's live engine. Here's the reasoning, because how a tool decides what to ship tells you more than the list of what it shipped.

The setup

Pullback-to-VWAP is a popular intraday idea: a stock trends up, dips back to its volume-weighted average price, and you buy the bounce. On in-sample data it looked solid — clean entries, reasonable win rate.

Where it failed

Under walk-forward, out-of-sample testing — tuning on the past, then measuring on data the rules had never seen — its expectancy went negative. The in-sample performance was largely curve-fit: the rules had learned the noise of the test period, not a durable edge. Across regimes, it didn't hold.

Why we cut it anyway

A scanner's value isn't the number of setups it lists; it's the quality of what it surfaces. Shipping a setup with negative out-of-sample expectancy would pad the list and cost users money. We'd rather show fewer, proven setups — gap-up momentum, opening-range breakout, and short opening-range breakdown all earned their place by surviving the same test pullback-to-VWAP failed.

The principle

Be suspicious of any tool that only ever adds features and never removes them. Honest validation produces deletions. That discipline — keeping only what survives out-of-sample, regime-split testing — is the whole point.

See it in the product

Decade-validated, ML-scored setups on the full SIP feed.

Keep reading