Every claim on the ledger goes through the same checks. If a test cannot meet them, it is labelled PENDING or not published.
Exchange fees at retail rates, spreads, funding on perpetual futures, borrow interest when a position needs it, network fees on chain, and slippage where we can measure it. A strategy that only works with zero fees is reported as failing.
Signals are decided on closed bars and filled at the next available price. A target touched inside the fill bar does not count. Delays match what a person or a simple bot could actually achieve.
Results are compared with doing nothing special: holding the asset, holding an index, or random entries with the same number of trades and holding times. A strategy has to beat that alternative to count.
For tests that decide anything, the hypothesis, data, costs, sample size and pass rule are written and timestamped before the results are seen. Parameters are chosen on one part of the data and judged on another. No re-slicing after a FAIL.
Small samples and fat tails are common in trading. We report medians next to averages, show 95% ranges from bootstrap resampling, and say when a handful of lottery outcomes drive the average.
Every write-up ends with what the data cannot show: survivorship, overlapping windows, missing price impact, short history.