Stop guessing. Start building.

The most famous exit in trend following just lost its own head-to-head. For decades the channel-breakout playbook has said: buy a close above the 40-day high, sell a close below the 20-day low. The entry gets argued about. The exit is treated as settled wisdom. We ran the same 40-day-high entries through eleven different exits on 33.4 years of SPY — part of a 1,188-variant grid, one position at a time, frictionless, $10,000 fixed account. Holding every breakout for exactly 20 bars — no signal, no channel, just a calendar — beat the classic 20-day-low exit on net profit ($12,916.31 vs $10,550.27), max drawdown (18.2% vs 26.3%), and return-to-drawdown (0.20 vs 0.10). And if you insist on a channel exit, wider wins: the 40-day-low exit reached a 4.25% CAR against the classic's 2.69%. The grid ranks exits, it does not isolate why. What it does show: on these entries the 40-day channel, the 55-day channel and the 20-bar clock each earned more than the classic exit with a shallower drawdown. One honest caveat: frictionless results, single market. Real costs shrink every number here. Full tables and methodology: https://statoasis.com/overfit/research/40-in-20-out-the-hedge-fund-trend-strategy-still-in-use-today

Random entries matched or beat one AI-written rule 56.2% of the time. That was Rule 2, a Stochastic oversold buy in an uptrend. It made $23,316.58 on the build data and $6,264.20 on the blind test. Profitable, and no better than a coin. Compare Rule 4: random entries matched or beat it only 3.3% of the time. I had an AI write 5 trading rules for SPY, then tested each one out of sample against 1,000 random-entry runs on the same side and hold. All 5 made money. Only 2 beat 95% of the random runs. None beat buy and hold. Profit is not proof. Rule 4, the best of them, earned $31,954.46 out of sample. Holding SPY earned $87,835.34. You can read a 5 for 5 profitable record as success, or as a test that was too easy to pass. Frictionless, one market, one split, $35,000 starting capital. Save the poster, read the full study: https://statoasis.com/overfit/research/ai-strategy-validation

Here is what this gap study cannot tell you. It is frictionless. No commission, no slippage. The forward drifts are real but they are gross — a +0.19% average move after a small gap is exactly the kind of edge trading costs eat first. Magnitudes here are relative, not net. It is an event study, not a trading system. Events overlap, and backtest variants with fewer than 50 trades are flagged rather than dropped, because their rankings are unstable on new data. Those flagged variants are the ones to distrust first when you rebuild this yourself. What survives those caveats: the forward drift rises with gap size across every non-trivial bucket, the ATR lens is the cleanest, and long ranks above short, though neither side beats a random-entry control. Those hold; the tradeable net-of-cost version is your homework. I publish the limits because a study that hides them is not worth trusting. Full caveats section: https://statoasis.com/overfit/research/sp500-opening-gaps