Research note

Two real single-transit fits, and why they are not a timing result

25 October 2026 Research note Working note

KELT-11 b and HD 209458 b fits with the published period fixed, plus the full-sector cross-check that disagrees at the formal precision.

The pipeline now has two real single-transit fits, both on public TESS photometry. The fits are a measurement at a predicted epoch. They are not new epochs, period measurements, detections, or transit-timing variations.

What was fitted

The same stages are applied to both targets: a threshold aperture on the target pixel file, a fixed quality mask, a running-median detrending step, and a trapezoid transit fit in a window set by the published ephemeris. The published period is fixed. That choice is what makes the fit a single-transit measurement rather than an ephemeris update.

Target	Sector	Fitted mid-transit time (BTJD)	Formal 1-sigma	Offset from prediction	Residual RMS
KELT-11 b (TIC 55092869)	9	1553.816219	80.3 s	+229.8 s (+2.86 formal sigma)	694.6 ppm
HD 209458 b (TIC 420814525)	56	2840.878747	2.3 s	-13.0 s (-5.62 formal sigma)	586.0 ppm

The HD 209458 b entry is a control. Its ephemeris is older and the fit is a backward extrapolation of about 2321 cycles. The offset is small in seconds and large in formal sigma, which is the point: the formal least-squares uncertainty is not an external error calibration.

The cross-check

A full-sector reduction and fit was run once on the same committed HD 209458 b products, using the committed ephemeris. It is a separate record, not a second single-transit fit.

Quantity	Full-sector	Single-transit	Difference
Mid-transit time (BTJD)	2840.879730	2840.878747	+84.9 s
Formal 1-sigma	16.8 s	2.3 s	5.0 times the quadrature sigma
Model depth	2,336.9 ppm	15,174.5 ppm	Not like-for-like
Residual RMS	1,054.3 ppm	586.0 ppm	Different fits

The two paths do not agree at the formal precision. The model depths are not a clean comparison either: the full-sector path detrends with a 101-cadence window and masks the in-transit cadences, while the single-transit path uses a 1001-cadence window with no in-transit mask. A trend fitted across a whole sector and a trend fitted across one transit window do not estimate the same baseline.

The comparison record reports the difference. It does not establish which path is closer to the true depth, and it does not apportion the difference among the detrending window, the in-transit mask, and the fit window.

What this does not establish

No new epoch. Both fits fix the published period.
No detection or confirmation. The fits are at predicted epochs.
No timing variation. The offsets are not tested against a noise model.
No externally calibrated uncertainty. The formal covariance is not a full error budget.
No sector-novelty claim. Neither sector is shown to be absent from its published ephemeris.
No combined real timing result. The cross-check disagreement is reported, not resolved.

Why publish this

The synthetic recovery campaign says the pipeline recovers injected times on a known fixture. It does not say the real-data path is calibrated. These two fits and the cross-check are the first real test of that gap. The result is a measurement with a visible disagreement, which is more useful than a single number presented without its checks.

The next steps are a physical transit model with fitted limb darkening and dilution, alternative detrending runs that isolate the 84.9 s difference, a correlated-noise term in the real timing uncertainty, and more target-sector pairs from the docket.

Evidence in the repository