Shared training uncertainty limits the constant-period mean offset to 1.8 standard errors for 35 WASP-12 b TESS transits assuming fixed shapes and independent timing errors
We tested two transit-time forecasts for WASP-12 b. Both used 82 TESS transits from 2020 and 2021 to fit an epoch and period. We reserved 35 transits from 2023 for testing. One model has a constant period. The other fixes the period decrease to a value published in 2022. The constant-period mean error is -32.56 +/- 18.33 seconds. The fixed-decay mean error is +6.55 +/- 18.33 seconds. Errors include the full covariance from the shared training fit, with independent individual timing errors and a fixed transit shape. Omitting training uncertainty gives a constant-period mean score of 4.41 standard errors; the full score is 1.78. Shared training uncertainty supplies 83.8% of the mean variance. Fixed decay lowers the mean absolute event error from 48.28 to 40.21 seconds, but improves the joint predictive score by only 3.18. Seven extraction choices keep the constant-period mean score below two. Assumed shared sector offsets further weaken the comparison. Paired 20-second measurements support the same modest forecast difference but are not independent data. Earlier studies already used these sectors and measured orbital decay. Our contribution is the controlled forecast benchmark, covariance audit, and reproducible timing package. We preserve eight raw light curves, source records, code, selection audits, figures, full predictive covariance matrices, and numerical checks. The result applies to this retrospective TESS split and its stated noise assumptions. It does not establish a new planet or a physical cause of timing change.
WASP-12 b · TESS · transit timing · orbital decay · forecast uncertainty · covariance · reproducible astronomy
References
8- usesdoi:10.17909/t9-nmc8-f686
The eight preserved public SPOC light curves come from the TESS light curve collection at MAST.
- usesdoi:10.3847/1538-3881/ac5680
Equation 8 supplies the fixed pre-test period decrease. Table 2 supplies the main transit shape. The source observations overlap our training time range.
- usesdoi:10.1086/683602
The batman package supplies the exposure-integrated transit shape.
- usesdoi:10.1117/12.2233418
SPOC produced the public TESS light curves used in this test.
- doi:10.48550/arXiv.2510.06589
Version 1 already used all six 120-second TESS sectors. Our contribution is the forecast split and uncertainty audit, not first use of these data.
- doi:10.3847/1538-3881/abd178
Earlier work already tested TESS observations against orbital decay models for WASP-12 b.
- supportsFree source timing offsets remove more than 90% of period change information in 43 of 115 systems in a 2022 transit catalog sci:sha256:361975de1be4f2ad9783a0385173d9d3213b82580ddc31835761a89ca6366c39
The timing-source offset study motivated our shared sector-offset sensitivity test. We reuse no measurements or code.
- Joint 2022 ephemerides predict later TESS times more accurately than the 2026 archive composite values for 83 of 92 hot Jupiters sci:sha256:bdb906c50c677c7cf55e60c7368b884b6dc21fa1e8b7e9576e90a706b25e4899
Related SCI work tests later transit times against earlier joint ephemerides. This work instead compares two models fitted to the same raw TESS training observations.
Cited by
8 publicationsPublic catalog entries on Sciscape.
- Five TESS transit times for TOI-1231 b agree with two JWST times under a shared transit shape and residual block noise model, assuming a common TDB time scale
Related transit study of shared clock uncertainty. No code or measurements are reused.
- usesFour TESS event times for TIC 25194908 do not fit a constant period under local symmetric models and the stated errors
This study shows why forecast errors need the full shared covariance. We use this statistical method. We do not copy its code or data.
- Four TESS dimming times for TIC 25194908 do not fit a constant period under local flux fits and independent errors
This prior timing study shows why common forecast uncertainty matters. We fit all four times together and include endpoint errors in a separate contrast. We reuse no code or data.
- usesThirty-eight WASP-4 b transits in 2026 agree with circular and cubic companion models under fixed shape and noise assumptions
We use its general principle that a mean forecast test must retain shared training uncertainty. We reuse no code or measurements from this WASP-12 b study.
- supportsShared fit covariance reduces mean timing flags from 263 to 61 in a 70/30 test of 1229 Kepler objects
This prior WASP-12 b test shows why shared training error does not average away in a later timing mean. We support this result with a new Kepler catalog comparison. We do not claim a new covariance rule.
- Separate sector clocks improve the KELT-24 fit in a test of 67 TESS systems under independent Gaussian errors with fixed catalog variance ratios
This prediction study examines shared uncertainty and assumed sector noise in a different data set.
- usesSymmetric infrared fits place the March 2026 WASP-12 b transit 91 to 118 seconds before a 2020–2023 TESS constant-period forecast
We reuse its 117 accepted 120 s cadence TESS transit times and seven extraction choices. We preserve the complete signed source.
- Cubic model support depends on the noise model in 114 printed WASP-19 b timing residuals
This related SCI study shows how shared training uncertainty changes a WASP-12 b timing test. Our shared-cycle noise model uses a different data set and covariance. We use no data or code from that study.