Every hypothesis tested against recorded venue data, including the ones that did not survive. Each page is the executed notebook: the charts are live, the numbers are reproducible from the committed fixtures.
Five gates, applied in order. The first one an idea fails is its verdict, so a rejection always names a number rather than a misgiving. Nothing reaches a live book without clearing all five, and conditional means it clears them only under the constraint stated on its card.
Gross carry is not edge. Both legs pay taker fees on the way in and again on the way out: 4.40 bps on Hyperliquid plus 5.00 on Binance is an 18.80 bps round trip for a cross-venue pair, 25.56 for spot-vs-perp. An idea is costed at taker on both sides before it is quoted.
Fixed cost divided by daily carry gives the day the trade turns positive. Past two weeks, margin risk, redeployment and the chance of the spread closing dominate the carry, so a breakeven beyond that is a financing position rather than a trade.
Funding over the holding period against the price dispersion of the same window. One ratio, cheap to compute, and it rejects in one number: at 37x smaller, collecting funding on an unhedged book is a rounding error on the position. Delta-neutral structures are exempt, which is the only reason idea 3 survived idea 2's rejection.
A mean carried by its top decile is a description of a few windows, not of the strategy. If the median window earns nothing, the edge is a tail and is sized as one.
Non-overlapping windows only. Thirteen at 30 days is thin; three at 90 days is an anecdote, and no confidence interval is quoted around it because with three samples one would be theatre.
Falsified as stated, and not a basis trade. The 25.56 bps round trip costs ten days of funding, so the position turns positive only past two weeks. What it earns over that fortnight is funding, not basis.
Rejected outright. Over 30 days the position collects 76.8 bps against a 2,885 bps standard deviation in the same window: funding is 2.66% of the bet, so the subsidy is noise against the risk taken to collect it.
Not tradeable as stated. Mean +3.95% annualized against a median of exactly zero: a third of windows sit at both venues' floor and the top 10% carry 94% of the total. Breakeven is ~17 days.