Stop guessing. Start building.

2,160 backtested variants. 8,397 gap events. Three ways to measure the same gap. The design, because the design is the point: Every SPY gap since 1993 measured for forward drift at 0, 1, 2, 3, 5, 10, and 20 days — overlaps kept, because this is measurement, not trading. Then the same gaps bucketed three ways: price %, raw points, and ATR % (volatility-relative). Then a flat-only backtest across the full grid, long and short, $35,000 per test, frictionless. The three lenses don't agree by accident. Price % shows a clean climb above the second bucket. Raw points hide the edge in an era-mixing dead zone. ATR % concentrates the next-day drift above 1 ATR: +0.200% against +0.039% to +0.055% below it. Over five days it does not, and the 0.5 to 0.75 ATR bucket leads at +0.345%. Same data, and the yardstick decides what you see. Every number generated by one deterministic run. Methodology and full results: https://statoasis.com/overfit/research/sp500-opening-gaps

I expected down-gaps to be the tradable ones. The data disagreed. The folklore says a down-gap is fear you fade and an up-gap is greed you sell. Across 8,397 SPY gaps, both drift up over the next week, and by almost the same amount — down +0.194%, up +0.206%. The side you gapped is not the edge. Every top-ranked variant in the 2,160-strategy grid is long. The median short variant fights SPY's structural upward drift and loses: median CAR/MaxDD is +0.010 to +0.020 for long variants and -0.020 for short. Long still sits below a random-entry control's 0.036, so that is a ranking of the two sides, not an edge. Long beats short. Neither beats random. If you short a down-gap because it looks weak, you are betting against a +0.194% five-day drift measured across 3,659 of them. The gap tells you how much to expect, not which way to bet. Full study: https://statoasis.com/overfit/research/sp500-opening-gaps

"Measure a gap in points" is where the size signal breaks. Rank SPY gaps by raw point size and a dead zone appears: gaps of 1–2 points show a 5-day drift of +0.007% — basically nothing — while smaller and larger gaps both do better. That looks like noise, and it is: raw points mix a 1995 tape with a 2026 tape, so the same "1 point" means something different in each era. Switch to a volatility-relative lens, gap size in ATRs, and the artifact disappears. Gaps larger than 1 ATR jump to a +0.200% next-day return versus roughly +0.04% for smaller ones. Same gaps, honest yardstick, clean edge. Put exact numbers on that step. The four buckets below 1 ATR averaged between +0.039% and +0.055% the next day. The 251 gaps above it averaged +0.200%. Size your gaps in points instead and the biggest bucket, over 4 points, averages just +0.075% the next day. The measurement wasn't wrong. The units were. Full breakdown of points vs price % vs ATR %: https://statoasis.com/overfit/research/sp500-opening-gaps