connecting…
—
The fade (run_down_ge_3) is the strategy.
close_near_low has no run requirement and fires on ~39% of bars
against the fade's 16%, so it is kept as a labelled observational arm and
never pooled into a headline — a pooled number is dominated by whichever
rule is noisier, not by whichever is right. When both fire on one window
the fade takes it.
Cash plus open positions at cost. Open positions are deliberately not marked to market — a 5-minute binary's mid is noise.
Point estimate with its 95% Wilson interval. The marker is the average entry price — the win rate this arm must beat to make money. RESEARCH_PROTOCOL §4 ranks on the interval's lower bound, never the point.
Do the orders we miss win more often than the ones we get? near_expiry.md found unfilled windows resolved Up ~100% of the time. A positive gap here means resting orders are being picked off, and no backtest survives it.
Our Binance candle vs how the venue actually resolved. 5m and 15m settle on Chainlink, so disagreement is the basis risk Gate 1.0 put at ~7% of any edge — measured here rather than assumed.
Resting orders and filled positions awaiting settlement.