Nightly means lose over half of the information for the lower sidereal alias in 12 of 152 HARPS series for a 5-day circular signal and shared nightly noise of 1 m/s

We test how nightly means change the information used to separate circular radial velocity signals from daily aliases. We use 88,687 public HARPS RVBank records for a fixed sample of 152 source labels. The primary signal has a 5-day period. Alternate frequencies are abs(1/5 +/- 1.00273790935) cycles/day. The covariance has formal exposure errors and a shared nightly error of 1 m/s. All fits include era offsets and a common trend. Every compressed model uses exact averages of the original model values. For 12 of 152 series, nightly means keep at least 90% of signal information at every phase but at most 50% of subtract-branch alias information at every phase. Nine series meet both limits for the add branch. Most series have less loss. The median phase-averaged alias mismatch ratios are 95.94% and 96.58%. A later test keeps a mean and one linear contrast per night. It reduces 88,687 exposure rows to 36,283 rows. At 5 days, all 152 series then keep at least 98.29% of subtract-branch alias information at every phase. For the add branch, 149 of 152 keep at least 90%. The source audit, dense covariance checks, Gaussian tests, and clean reproduction support the calculation. The result is conditional on the model. It makes no planet detection claim. Code, source records, provenance, all case matrices, tables, and figures are supplied.

HARPS · radial velocity · daily alias · nightly averaging · time series · information matrix · data compression · public data