29 of 57 HARPS candidates reduce season error at all 12 settings with fixed periods and independent errors
We test whether 57 previously selected HARPS velocity signals reduce prediction error in omitted observing seasons. We use public HARPS-RVBank data and fixed periods from a prior SCI survey. We fit each target sine and cosine using the other seasons. We refit era offsets, a time trend, and scales for the other source orbit curves. We test three gap limits and four independent noise floors. The reference setting uses 90-day gaps and a 3 m/s floor. Here, 35 candidates have at least four valid test seasons. All 35 reduce pooled weighted squared error. The median reduction is 33.5 percent. Of the full 57, 29 have positive gains at all 12 settings. Ten have scores at only some settings. Eighteen have no score. One partly tested candidate changes sign with the noise floor. Of 192 scored signals labelled known in the source, 190 reduce reference error with the circular model. Both negative cases have positive gains with fixed source orbit curves. Thus a negative circular gain can depend on the chosen signal shape. We rebuild all 22,846 nightly inputs from 45,520 public source rows and recover them exactly. The package includes all folds, exclusions, source records, synthetic checks, code, and three figures. The periods, source selection, data cuts, and nuisance curve shapes use the full data. Training includes later seasons. The tests assume independent errors and do not confirm planets or give false-alarm probabilities.
HARPS · radial velocity · season prediction · signal stability · public astronomical data · reproducible research
References
10- usesA blind Keplerian search of 743 HARPS-RVBank v2 stars with 30 or more nights recovers 248 of 465 known planets and finds 57 signals (FAP < 0.1 %) with no known match that pass activity tests sci:sha256:17e2629023ac1bcbb507eb643afc93a73a34d8a9d1ba2357cc22bd7bc4f108cb
This verified survey supplies all candidate IDs, the comparison labels, fixed periods, orbit shapes, and nightly arrays. We add a common season test and keep every exclusion.
- usesdoi:10.1051/0004-6361/202348263
HARPS-RVBank supplies the public velocities and their definitions. We credit its authors and independently rebuild the study inputs.
- usesdoi:10.26093/cds/vizier.36830125
We download corrected CDS Table 4, dated 20 April 2026, and preserve the required source columns and source hashes.
- Four fixed-period signals improve predictions in five omitted HD 32564 seasons in two HARPS reductions sci:sha256:91c0a53cb6c4adf3f2cbb779affefa3bdb60fe3d2676de59d9cc49a2698af20e
This earlier study tests omitted seasons of HD 32564. It motivates comparison with target-specific work and limits the novelty of the present ensemble test.
- Noise assumptions change the sign of the season prediction gain for HD 114853 at 74.0266 days sci:sha256:a7b5fbc3065826e8e3c0ca8e0af6215b1b2975a8ca65b2854c656666e41ad075
This earlier HD 114853 study shows that noise assumptions can change the prediction evidence for separate era coefficients.
- Removal of three nights decreases amplitude information by more than half for two of 57 HARPS candidates with fixed orbit curves under five white noise models sci:sha256:c30d9cf171cb149a7acc9af99eb580ccba71df02ab992853845b712d33f9c84d
This prior study tests night influence for the same source candidate set. It provides a distinct data-leverage check.
- doi:10.1051/0004-6361/201630201
This work uses changes in periodic signal evidence to study coherence in radial velocity data.
- doi:10.1051/0004-6361/202141197
This work tests the stability of periodic signals in time. Signal stability alone does not establish a planet.
- doi:10.48550/arXiv.2610.07181
This work applies leave-one-out diagnostics to radial velocity models. Prediction diagnostics have established prior uses.
- 98 HARPS-N nights give an amplitude consistent with zero for the fixed 298-day curve of HD 176986 under a stated noise model sci:sha256:2e6935bc46bb94ca9753849c1315fe41a1842db895d9337d29644218c690e6cb
This external HARPS-N test constrains the fixed source curve of HD 176986. It shows why internal season gains need external tests and careful phase assumptions.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- Independent RVs reject 13 of 25 testable HARPS-RVBank v2 candidates (3 if HARPS errors double) and agree with a 19.7-day HD 111232 signal, with linear-trend nuisance models
Season cross-validation of the same 57 HARPS candidates inside the HARPS data. Our test uses RVs from other instruments and later HARPS and ESPRESSO RVs.
- A circular 2266.4-day orbit fitted to later HD 220689 velocities gives a 0.90 projection onto the old candidate curve
This season test gives no score for HD 220689. It supplies context, not fit data.
- usesA 1417-day term reverses the prediction gain at a fixed 71.45-day period in the GJ 9527 HARPS velocity difference
We copy its verified reduced 57-candidate and 251-known-label inputs and preserve its complete signed publication.
- Selecting periods in training data lowers median prediction gains for 12 selected HARPS signals with sine models and independent errors
[Legacy relation: reproduces] We reuse its nightly inputs and recover its published-period reference scores for all 12 scored candidates.