24h moves net of Bitcoin beta

Each contract needs 193 contiguous complete hourly closes c0…c192, ending at the exact latest completed hour, on the same market ID; the BTC reference needs the same. From them come 192 log returns. Beta = cov(coin, BTC) / var(BTC) and intercept = mean(coin) − beta × mean(BTC) use only the first 168 returns; the last 24 are never part of the fit.

Residual (log percentage points) = 100 × (Σ coin log returns − 24 × intercept − beta × Σ BTC log returns) over the evaluation hours. Coin 24h and BTC 24h are ordinary compounded returns, 100 × (c192 / c168 − 1), so they are not in the same units as the residual. Outperformers sort by residual descending, underperformers ascending; ties use coin and market ID.

This is a descriptive comparison with one statistical baseline, not causal attribution, alpha or a trading forecast. Missing, partial, stale or invalid candles, non-finite arithmetic or a BTC training variance at or below 1e-12 leave a row unranked; gaps are never bridged and other contracts' prices are never stitched in. If BTC is absent from the selection or its history is incomplete, nothing is ranked. Up to 50 reference perpetuals selected by completed 24h quote volume, from stored exchange public-feed candles. Refreshed every two minutes and at each UTC hour.