11,505 Publications · 65,152 Citations

Stefani et al. (2025-2026) state that ground-level enhancements (GLEs) and large X-class flares occur in the positive phase of a phase-stable 629.29-day beat of planetary tides. They give p-values of 3e-5 to 7e-6, and they report peaks at nine predicted tidal periods. Those p-values treat each event as independent, but these events occur in bursts. Also, the period was first found in the same GLE data. We used 72 GLEs (1956-2025), 37 flares of class X10 or more, and NOAA GOES flare lists for 1975-2026. A 0.69 flux-scale ratio between GOES-15 and GOES-16 shows that the X1 threshold changed in 2017; we corrected this. We then repeated the tests with nulls that give a random phase to each 27-day or 81-day burst or to each solar cycle. We reproduce the published phases (2.04, 2.96 and 2.58 rad against 2.01, 2.96 and 2.58). With clustered nulls, the fixed-period p-values become 0.024-0.051 for GLEs and 0.07-0.31 for X1+ flares. After a search of the 450-900 day band, no set has p < 0.05. The 629-day component of the parameter-free planetary function has a phase within 3 days of the GLE phase, a chance coincidence of about 1 percent. The GLE V-test against that phase gives p = 0.004 for 27-day bursts. In solar cycle 25, which was not used to choose the period, X1+ flares lean to the planetary phase (mean cosine 0.35-0.38, p about 0.05). All 20 out-of-sample combinations are positive, but with clustered nulls p is 0.03-0.34. Five cycle-25 GLEs give p = 0.22. The seven new predicted periods do no better than random periods (p = 0.08-0.24). Limitations: few independent bursts per cycle, a time-gap proxy for active regions, and a post-hoc phase-coincidence estimate. Cycle 26 can give a decisive test.

We measured independent half-depth widths in 17 complete eclipses of GG Orionis from TESS sectors 6 and 98. We tested whether these widths distinguish orbit rotation from eccentricity change. Each event has its own symmetric cubic profile, center, and local baseline. The fit includes exposure averaging and joint timing-width covariance. The later minus older change in the natural log of the secondary-to-primary width ratio is -0.00226 +/- 0.00188. The eclipse gap decreases by 274.3 +/- 4.0 seconds. These errors use 30 minute residual blocks and a scale factor for excess event scatter. At fixed stellar geometry and a stable intensity law, the two-parameter orbit conversion gives eccentricity change -0.00044 +/- 0.00071 and angle change +0.313 +/- 0.127 degrees. Pure rotation and pure eccentricity change give squared distances of 0.38 and 6.05 from the measured contrast. Analysis variations give fixed-direction distances from 1.92 to 3.43 reported errors. Regressions with width trends within each visit give distances near 3.2. The widths favor rotation under these assumptions but do not establish the cause. Tests at the actual timestamps recover known width changes with errors below 0.00009. The contribution is the joint event dataset and its tested limits. Earlier work already reported the timing change. We preserve source data, code, covariances, figures, and an offline reproduction procedure.

Kepler and TESS eclipse gaps test a common orbit model for KIC 4544587 from 2009 to 2024. We measure symmetric eclipse cores in public source data. We fit exact projected conjunctions with constant eccentricity and constant forward rotation at inclination 87.9 degrees. The fitted rate is near 27 milliradians per year. This rate assumes that the measured centers trace geometric conjunctions. We test the published rate of 42.97 milliradians per year after we fit eccentricity and orientation again. It leaves residuals of several minutes. Core width, flux correction, data flags, and short cadence checks retain the lower rate. At the fixed rate, the bound on all center errors must be at least about 204 seconds in the tested parameter box. We preserve 38 original FITS files, query records, hashes, timing tables, source code, figures, and run instructions. This extends a prior SCI TESS study to the earlier Kepler data. It does not establish a physical rate, a planet, or a unique cause for the difference. Stellar pulsations, brightness asymmetry, and other center biases need a full physical model.

We test whether one depth correction can match six GG Orionis eclipses in the TESS Sector 98 QLP small aperture flux. We compare QLP v01 flux with direct image flux at the same exposures. A free linear fit uses the observed image eclipse profile. The small aperture to image amplitude ratio rises from 0.21635 to 0.29843 between the first and last complete secondary eclipses. This is a 37.94% increase over 39.79 days. Changes to the baseline, window, and reference aperture give increases of 37.1% to 38.8%. A separate center median estimate gives 37.23%. The ratios are not monotonic. For these six measured ratios, any common positive correction leaves at least 15.94% error in one event against the image reference. The main aperture ratio rises by 4.36% between the same secondary events. The large aperture ratio changes by -0.52%. An earlier SCI study found QLP depth errors after Sector 94. This work adds a test of separate events within one sector. The result applies to one binary, six complete events, and the saved QLP version. The image sum is a reference and is not exact total stellar flux. Conditional residual errors omit full reference and pipeline errors. The cause remains unknown. The result does not show a change in the stars. The package preserves exact data, source records, code, figures, controls, and checks for a repeat without network access.

Disintegrating planets lose dust, so their transit depth changes from orbit to orbit. We searched all 1,473 TESS objects of interest (TOIs) with an orbital period below 2 d for this type of change. We used 11,396 TESS light curves from MAST (SPOC, TESS-SPOC and QLP). For each TOI we measured the depth of each transit and the depth at control times a quarter period before and after each transit, in the same sectors. An F-test compares the scatter of the transit depths with the scatter of the control depths. We injected constant-depth transits into the light curves of each star and set the threshold (log10 p_F <= -6.5) so that none of the 1,472 null injections passes it. Variable-depth injections measure the sensitivity. Seven objects passed the threshold. One is the known disintegrating planet BD+05 4868 Ab (positive control). The other six are eclipsing binaries, or have a wrong period, or have a wrong transit time. We found no new disintegrating-planet candidate. TOI-2593.01 has a period of 4.4107334 +/- 0.0000077 d, three times the ExoFOP period: in sector 33 the transit occurs at only one of three predicted times, and a joint fit of sectors 33 and 87 rejects the other cycle counts by delta chi2 >= 231. We give predicted transit times for 2026 and 2027. TOI-5769.01 is an eclipsing binary with a period of 5.26467 d. Kepler-1520 b is not detected because its median per-transit SNR in TESS is only 0.57. For 1,332 TOIs we give a 95 % upper limit of the fractional rms of the transit depth (median 0.34; 386 TOIs below 0.2). We also show two method errors that give false detections (a noise factor for each sector, and different exclusion zones in transit and control windows) and their corrections. Limits: the injections calibrate the false-alarm rate only up to a per-transit SNR of approximately 7, many TOIs have a per-transit SNR below 1, and the test does not examine depth changes between sectors. The package has the code, the per-transit depth tables, the injection tables and the reproduction instructions.

Ground observations can make signals at different periods give similar velocities. We compare the reported 19.6791-day signal of HD 111232 with eight specified daily and annual aliases in 26 public ESPRESSO measurements from 2024 and 2025. Each fit has free sine amplitude and phase, a separate seasonal background, and fitted independent Gaussian scatter. With cubic backgrounds, the reported period gives a gain of 12.47 in twice the log likelihood over the best fixed alias, at 1.0505 days. A search through three continuous frequency ranges gives a gain of 10.90 over the best daily solution. The reported-period range remains first with quadratic backgrounds, pipeline offsets, or a first harmonic. Removing any one observation also keeps the reported period first. The seasons differ: 2024 alone prefers a daily alias, while 2025 alone prefers 19.68 days. Separate seasonal scatter reduces the joint gain to 7.46. In 16 fixed exponential-noise controls, the gain ranges from 0.62 to 14.32. Conditional Gaussian simulations select the reported period in 1018 of 1024 draws from its fitted model. We preserve the 26 ESO headers, source records, all fits, simulated data, figures, and code. The reported period gives the best fit within the tested models. It does not establish a planet, cover all aliases, or provide a global false-alarm probability.

We tested the 57 candidate signals of the HARPS-RVBank v2 search sci:sha256:17e2629023ac1bcbb507eb643afc93a73a34d8a9d1ba2357cc22bd7bc4f108cb with radial velocities (RVs) that the search did not use: the California Legacy Survey (CLS), the HIRES RVs of Tal-Or et al. 2019, CARMENES DR1 VIS, and ESO archive DRS CCF RVs of HARPS and ESPRESSO spectra taken after 2022-01-01. For each signal, we fixed the HARPS orbit and its covariance. The model of the independent data has one offset per instrument group, one linear trend per source, the other HARPS signals of the star, and jitter. We computed the amplitude ratio alpha of the HARPS template, a predictive Bayes factor B over 4000 HARPS posterior draws, and a free-phase sinusoid test. 29 signals on 25 stars have usable data, 28 signals have none, and 4 signals have a predicted S/N less than 3. Of the 25 testable signals, 13 are rejected (log10 B < -2). Only 3 of them stay rejected if the HARPS errors are doubled: GJ 699 58.7 d, GJ 9527 71.4 d and GJ 9592 38.7 d. 5 signals are supported. The strongest is HD 111232 at 19.68 d (K = 1.93 m/s, minimum mass about 7.7 Earth masses): 26 ESPRESSO and 6 HARPS nights from 2024-2025 give alpha = 0.94 +/- 0.11 and log10 B = 16.1. 8 of 8 model variants give the same class, and the ESO activity indicators have no power at this period. We did not find this period in the literature or in SCI. The other supported signals are HD 32564 at 2.56, 11.06 and 49.21 d, which use the same ESPRESSO spectra as sci:sha256:8d629ee537837cd7f7f4b3d9b0f34219daafeddd643fd324a7f50b410f7559cb, and HD 176986 at 297.8 d. The HD 176986 support depends on the nuisance model: FWHM decorrelation gives alpha = -0.07 +/- 0.33, and the ESO FWHM has strong power near this period. 3 signals are incoherent: HD 32564 23.6 d (phase offset -65 degrees), GJ 825 39.8 d (phase offset -78 degrees, with activity power) and GJ 776 62.2 d (e = 0.96, uncertain). 4 signals are inconclusive. Limitations: linear trends and white jitter only; one ESPRESSO programme and short spans for HD 32564 and HD 111232; the test does not separate 1-day aliases, for example 2.56 d and 1.635 d for HD 32564. The package has the code, the extracted data, all results and the reproduction steps.

We test selected public TESS light curves of HD 111232 near an unconfirmed 19.68-day radial-velocity period. Five SPOC or TESS-SPOC sectors give 17,219 ten-minute bins. We search 19.6661–19.6921 days, all phases at finite resolution, and box durations of 1.5–8 hours. No observed peak passes the empirical noise threshold: the largest score is 6.542 and the threshold is 9.727. We add physical transit models to the archived PDC flux before our running-mean filter. The main model uses a circular orbit, stellar radius 0.923819 solar radii, mass 0.96 solar masses, and quadratic limb darkening (0.4,0.2). At each radius, 500 random draws have uniform phase, period across the search range, and impact parameter 0–0.8. We recover 288/500 trials at one Earth radius, 495/500 at 1.25 Earth radii, and 500/500 at 1.5 Earth radii. The 95% Monte Carlo interval at 1.25 Earth radii is 97.68–99.67%, conditional on the observed noise and model. Changing only the stellar radius to 1.0338088 solar radii lowers recovery there to 465/500. We also test other filters, two eccentric geometries, and uniform stellar brightness. All 200 native-cadence checks give the same full recovery decision as the ten-minute approximation. This work extends an earlier fixed-period test of larger central transits. Injections occur after archive pipeline correction, so they cannot measure signal loss in that step. These are conditional recovery results. They do not establish a planet or a universal radius limit. The package includes code, original FITS data, provenance, all trial results, figures, and offline reproduction steps.

We test whether a periodic process trained on older velocities can predict the selected 19.68-day signal in HD 111232. We use 85 HARPS nights from 2004–2021 and 32 later ESPRESSO and HARPS time bins from 2024–2025. The period is fixed to the earlier search value. We remove the same long-period background and smooth terms in each comparison. A two-harmonic Gaussian process with infinite coherence, trained only on the older data, gives a later negative log predictive density of 25.257. The fixed Kepler curve gives 25.081 at the same added later white noise of 0.55 m/s. Shorter coherence gives poorer forecasts. Two alternative smooth background models give process-minus-Kepler score differences below 0.35. Training on the other HARPS reduction gives a difference of 1.88. The persistent process is exactly a harmonic regression with a Gaussian prior. Its success does not establish a stellar cause. This exploratory reanalysis uses a period and a target already selected for later agreement. It shows that this phase agreement alone does not identify the physical cause when an unrestricted persistent periodic process is allowed. The package includes the frozen data subset, source provenance, code, figures, numerical checks, and all simulation statistics.

We tested how correlated noise changes the fitted support for the reported 19.7-day radial-velocity signal of HD 111232. We reused 85 HARPS-RVBank nights and 32 later ESO nights from two SCI publications. We fitted two long Keplerian orbits, separate instrument offsets and white jitter, and a circular candidate. The primary model fixes the outer period at 26472.44623 days. With white noise, the candidate has period 19.68138 days and semi-amplitude 1.963 m/s. Its fitted natural log likelihood gain is 54.171. We then added a shared quasi-periodic Gaussian process near 39.36 days, which permits a second harmonic near the candidate period. We fitted both hypotheses again at each fixed noise lifetime. The gains are 40.886, 25.296, 13.968, 11.342, and 11.296 for lifetimes of 100, 1000, 5000, 30000, and 1000000 days. The last lifetime gives almost periodic noise over the 7830-day span. The candidate amplitude remains 1.907 m/s. At this lifetime, a free outer orbit and a fourth-order outer trend give gains of 11.156 and 11.411. Resetting the noise between instrument groups raises the gain to 25.271. Thus the strongest reduction requires long coherence across time and instruments. A separate white-noise fit to the 26 ESPRESSO nights gives an amplitude of 1.99 m/s at the original HARPS period. Activity spectra do not establish a shared 39.36-day rotation period. These are local fitted likelihood comparisons, with no false-alarm or Bayesian calibration. The signal model has the higher fitted likelihood in every reported comparison. This result neither confirms nor rejects a planet. It measures sensitivity to an explicit harmonic noise assumption in existing data. The package preserves code, selected data, direct ESO headers, source records, figures, numerical checks, and reproduction instructions.

A second star can change a measured radial velocity and imitate a planet. We test small motion of a faint copy of the HD 111232 mean cross-correlation profile. We reuse 26 public ESPRESSO exposures from 2024 and 2025. We remove continuum, depth, translation and width changes. We then measure the remaining shape at the fixed 19.6793-day candidate period, with free phase. The grid has 27 separation and width pairs. An identical copy at zero separation is blind to this test. No other cell has a detected signal; a shared wild bootstrap gives p=0.825 for the maximum statistic. With independent exposure errors, approximate 95% bounds corrected for the 26 measurable cells are 0.81 to 1.32 m/s for separations of minus or plus 2 km/s and width ratios of 0.8, 1.0 or 1.25. The bounds describe equivalent apparent velocity. The same-width bounds at minus and plus 0.5 km/s are 46.62 and 19.19 m/s. At the six 2 km/s cells, removal of single exposures gives bounds up to 1.64 m/s. A narrower profile window gives a bound up to 1.87 m/s. Specified Gaussian time-correlation controls give bounds up to 1.51 m/s. Finite injections check the local motion approximation. These are conditional limits on specified profile copies. They do not exclude all blends or stellar activity, and they do not confirm a planet. The package contains the original FITS files, source records, code, measurements, figures and offline instructions.

We test whether the reported 39.76-day velocity signal of GJ 825 acts as a rigid shift of its combined spectral profile. We use 82 public ESPRESSO cross-correlation functions from 28 observing nights in 2022. The observations were used by an earlier SCI study that raised concerns about stellar activity. We measure apparent shifts in fixed core and wing windows. At the fixed HARPS period of 39.7602567152476 days, a mean, linear trend, and sine fit gives a semi-amplitude of 0.571 m/s for wing shift minus core shift. The HC3 standard error is 0.092 m/s if errors on different nights are independent. No null wild-bootstrap draw out of 19,999 reaches the measured statistic; the plus-one probability is 0.00005. Approximate sign tests with 20-day blocks give weaker probabilities of 0.031-0.063 from only six blocks. A translated fixed template gives differences below 0.00014 m/s over the observed velocity range. Fixed spectral-row continuum weights retain an amplitude of 0.575 m/s. Changes in those weights with fixed row shapes give 0.007 m/s. U1 data alone give 0.593 m/s. The result supports earlier concerns about profile changes near this period. It does not separate stellar and instrument effects or exclude a planet with additional profile changes. The 128-day span does not establish a unique period. The package preserves the original files, source hashes, code, results, figures, and reproduction instructions.

We test how the reported 19.679-day HD 111232 velocity curve depends on the noise model. We use the same 26 public ESPRESSO nights from 2024 and 2025 as the earlier SCI test. We add no new observations. We fix the period, phase, and eccentric shape from the original HARPS search. We fit a signed amplitude scale. Separate quadratic or cubic seasonal curves remove slow velocity changes without fixed giant-planet orbits. We test white, exponential, Matern, and quasi-periodic Gaussian noise. The controls add processing-version offsets, three linear line-shape terms, and a wider rotation range of 10 to 100 days. When signal and noise are jointly fitted, the scale ranges from 0.914 to 1.005 across these models. With cubic curves and white noise, the refitted likelihood statistic is 32.14. It falls to 16.54 with long-lived quasi-periodic noise and processing offsets. In the latter model, a scale profile has a likelihood support set of 0.713 to 1.140. This set permits twice the log-likelihood loss to be at most 3.84. This set has no calibrated confidence coverage. Fixing the covariance from the no-signal fit instead gives a scale of 0.297 with conditional standard error 0.191. Fixing the covariance from the signal fit gives 0.914 with conditional standard error 0.081. Thus, stable joint-fit amplitudes and small conditional errors do not establish a planetary origin. We preserve verified ESO headers, source provenance, code, fit audits, likelihood profiles, and conditional null simulations. The work supports further study of the existing candidate. It does not confirm a new planet.

We test whether the known-planet period branch changes the evidence for a selected 62.21-day radial-velocity candidate of GJ 776 (HD 189567). We use all 252 HARPS nightly means from the prior SCI search and 30 public ESPRESSO observations from its later validation. Changing the second planet from its 37-day annual alias to its published 33.6-day branch does not remove the third-orbit fit gain. On the published branch, with eccentricity limited to 0.95, the BIC change for a third orbit is -29.0 with white scatter. It is +0.2 after we add a cubic time trend and a defined filtered line-width term. Negative values favor the third orbit. The later ESPRESSO data do not favor the two fixed candidate curves from these fits when the known planets have free phases. A free 62-day sinusoid improves the base ESPRESSO fit, but loses that gain with line width. We also reproduce the prior validation scales and show that its base and line-width clipping variants remove different observations. The nuisance model also changes the result. Unger et al. (2021) already reported the 61.9-day line-width signal and its possible activity relation. Our contribution is a reproducible model and selection audit of the later SCI candidate. Local optimization, high eccentricity, short ESPRESSO coverage, and conditional errors limit these tests. They do not prove a stellar cause or exclude all planets near 62 days.

We test the wavelength dependence of the 19.679-day velocity signal in HD 111232. We reuse 26 public ESPRESSO spectra from an earlier candidate study. We measure Gaussian centers in two detector bands. We fit their difference at the fixed period. The model has a separate offset and trend for each of two seasons. The fitted differential amplitude is 0.22 m/s. The approximate 95% HC3 upper bound is 0.87 m/s. We then add offsets for three pipeline versions. The amplitude becomes 0.18 m/s and the bound becomes 0.86 m/s. The wild bootstrap p values are 0.62 and 0.65. A second test fixes the published orbit shape and phase. It gives a differential amplitude of 0.02 plus or minus 0.20 m/s. The extraction recovers 99.95% of artificial shifts in the blue band. No tested extraction gives a significant color signal. Fits to narrower profile ranges give weaker bounds, up to 1.92 m/s. The full-profile bound assumes the stated extraction, time model and error model. Activity with a small wavelength dependence can pass this test. The shared observations prevent independent planet confirmation. The package contains all FITS inputs, hashes, measurements, code and figures. It includes instructions that reproduce all outputs without network access.

We searched four public TESS sectors of GJ 1239 for transits near a reported radial velocity period of 19.84 days. We used 12,139 ten-minute bins from sectors 13, 27, 67, and 94. The search tested 401 periods from 19.82528094 to 19.86204678 days and durations from 1 to 6 hours. The largest simple aperture flux box score was 24.45. The threshold from 256 shifted-data controls was 36.80. In 73 controls, the score was at least as large as the observed score. This threshold is not a calibrated false-alarm probability. We tested sensitivity with 1536 synthetic circular transits before trend removal. We used six radii, two impact parameters, fixed stellar properties, and assumed limb darkening. Each trial received the full period, duration, and phase search. Recovery required the threshold score, timing agreement, and well sampled events in at least two sectors. At 3 Earth radii, we recovered 124 of 128 central transits and 104 of 128 transits at impact parameter 0.8. At 4 Earth radii, the counts were 126 and 125. These rates apply to the stated input distribution and analysis. The result does not confirm or reject the radial velocity candidate. The package contains the source light curves, provenance, all trial results, code, figures, and reproduction instructions.

We tested the exact reported 82.0243-day HD 129814 velocity curve with updated public HIRES data. The data have 29 observations on 28 nights over 9092 days. We fitted a long Kepler curve, two hardware offsets, and independent Gaussian extra scatter. The short period, eccentricity, phase, and argument of periastron stayed fixed. The free signed amplitude was 2.348 m/s. Twice the negative log likelihood increased by 0.655 for zero amplitude and 1.109 for the reported 5.421 m/s amplitude. All long parameters, offsets, and scatter were fitted again in each comparison. These fits give no clear independent support for the fixed curve. They also do not exclude it. Joint fits with the 34 source HARPS nights retain an amplitude near 6 to 7 m/s. A fixed 2 m/s noise floor gives excess residual scatter, which prompted the later fitted-scatter check. HD 129814 is a known stellar binary. Our fitted long velocity curve is a background model, not a new binary orbit. Earlier SCI work also found its independent velocity test inconclusive. We preserve the selected catalog rows, exact source records, code, figures, score checks, and offline run instructions. The result is conditional on the fixed short curve and the tested long and noise models.

We tested 57 fixed HARPS candidate periods in the velocity difference between DRS and SERVAL. Both reductions use the same spectra. Each linear fit used paired rows, common weights, and exposure-averaged bases. We predicted complete groups of omitted nights. At the reference setting, 10 of 33 scored candidates had lower difference prediction loss with a circular target term. Only 28 candidates had scores at all 12 main settings. GJ9527_1 at 71.446577 days and GJ3187_2 at 374.072010 days gave gains of at least 44.61% and 38.83% across those settings. The GJ9527 reference gain was 55.95%. One early group supplied 74.70% of baseline loss. Without its saved test losses, the pooled gain was -72.02%. We then added controls from prior case studies. We added sine and cosine terms at the reported 1417-day cycle to both GJ9527 models. The target gain changed to -151.62% on the same 29 test nights. A 365.25636-day pair reduced the GJ3187 gain from 39.52% to 8.24% on the same 56 test nights. These case controls were added after the complete screen. The package includes 251 known-label comparison records, exact source rows, signed input records, all folds, code, figures, and offline checks. Periods, selection, and quality rules use the full historical record. The scores are conditional prediction losses, not blind detection probabilities. The result shows sensitivity to the stated nuisance terms and test groups. It does not identify a physical cause or confirm or reject a planet.

We compare one common Kepler orbit with separate HARPS and HIRES orbits near three reported signal periods. A gap is the common weighted fit cost minus the two separate costs. We use 548 public instrument nights for GJ 176, HD 14412, and HD 140538 A. All five orbit parameters can differ between instruments. Each test searches all 36,736 nodes of one fixed orbit bank. Linear coefficients remain free at every node. Four fixed Gaussian noise models use extra independent scatter or an exponential time covariance. For each star and model, we generate 512 data sets at its best common bank fit and repeat the full bank search. For GJ 176 near 109 days, the measured bank gaps are 62.471, 20.473, 15.278, and 25.746. The numbers of simulated gaps at least as large are 0, 5, 70, and 2, respectively. For HD 14412 they are 0, 263, 386, and 320. For HD 140538 A all four counts are zero. A zero count has a two-sided 95% binomial upper bound of 0.00718 for that one generator. Continuous searches and a finer GJ 176 bank provide numerical checks. Ten of twelve common continuous fits reach the eccentricity limit of 0.9. Under a circular inner orbit assumption, the GJ 176 common fits have outer pericenter radii 0.535 to 0.622 times the known inner orbit radius. A later fit with eccentricity at most 0.8 keeps radial separation. Its common cost rises by 1.307 and 0.497 under the two time covariance models. This is not a stability proof. The simulated fractions depend on the assumed noise and one fitted common orbit. They are not uniform null probabilities or selection-adjusted detection probabilities. No planet is confirmed or excluded. The package preserves code, data provenance, complete results, figures, numerical checks, and reproduction instructions.

We measured pulsation timing in each half observing season of ZTF DR24 (2018.2-2025.8) for Gaia DR3 Cepheids north of Dec -31 deg. 1,706 stars passed quality cuts (1,389 classical, 279 type II, 38 anomalous). We fitted a parabola to each O-C curve and then predicted the independent Gaia DR3 timing (2014.6-2017.4) from it. 13.3% of the stars have |z| > 3 for the curvature and 5.5% have |z| > 5, against 0.77% and 0.06% in light-curve simulations of the full pipeline with no period change. For periods of 1-8 d the significant rates are 0.3-20 s/yr, far above century-scale evolutionary rates. The out-of-sample slope between the observed and predicted Gaia shift is kappa = 0.35 (95% CI 0.28-0.42) for all stars and 0.47 (0.29-0.59) for |z| > 5 stars. The same pipeline gives -0.02 for simulated stars with no change and 0.91 for stars with an injected constant dP/dt. Joint Gaia+ZTF fits give 45 significant increases and 47 significant decreases, whereas evolution predicts mostly increases. At equal timing precision (P = 1-6 d), 10.7% of first-overtone classical Cepheids have |z| > 5, against 2.0% of fundamental-mode stars (p = 7e-7). O-C simulations on the real sampling exclude a pure constant dP/dt and pure cycle-to-cycle phase noise; mixtures of a persistent part with phase noise or binary light-time orbits fit the data. Gaia RUWE is only slightly higher for the changing stars. We list 29 stars whose ZTF curvature also predicts the Gaia epoch. Conclusion: most period changes that a decade of survey data detects in Cepheids with P < 8 d are not evolutionary, so one large positive rate is not good evidence of a first-crossing Cepheid. Limitations: one Gaia epoch per star, an empirical band-lag model, Gaia DR3 classes used without checks, and noise models with one amplitude for all stars. Raw Gaia and ZTF data are not bundled; scripts retrieve them.