Skycruncher observing log, 11 Oct 2026

A planet spiraling into its star, caught with a 50 mm smart telescope on short notice

WASP-12 b is a gas giant almost twice Jupiter's size that circles its star in 26 hours, so close that its orbit is slowly shrinking. Every time it crosses the face of its star, WASP-12 in the constellation Auriga, the star dims by about one and a half percent for three hours.

On the night of 10 October 2026 I turned my Seestar S50, a 50 mm smart telescope at a remote observatory in Texas, to WASP-12 about a quarter of an hour before a crossing was due, and left it taking 20-second frames until dawn. There was no planning beyond the pointing: no calibration frames, and the telescope didn't even know the star's name (it filed the frames as "Unknown").

From those 647 frames, the transit kit we use for these write-ups measured a dip 1.63 % deep (± 0.20 %), centered at 09:19 UTC (± 4 minutes). That is within half a minute of the newest published prediction. Older predictions, made from observations years ago, put the crossing 11 to 28 minutes later. That gap is the shrinking orbit, which astronomers have followed for about a decade.

It is a clear detection, about eight times its own uncertainty, of a known planet at the expected depth and time, with a telescope smaller than the 5-inch minimum the ExoClock timing project lists for this target. It is also one night from one telescope. A 4-minute timing uncertainty is too coarse to measure the orbit's decay on its own; it is one point that agrees with the measurements made by larger telescopes. I call the result exploratory until another night repeats it.

the dip, 10-minute averages with the fit

1.63 %depth, ± 0.20 % (8.3 times its uncertainty)

+0.4 minmid-transit against the newest prediction, ± 4.2 min

1.85 %scatter of a single 20 s frame

597of 647 frames used, 20 s each, 07:35 to 12:06 UTC

The shrinking orbit, in one chart

Astronomers predict transits with an ephemeris: a timetable made of one observed crossing and the orbital period, so every later crossing is that time plus a whole number of periods. Each published ephemeris below was fitted to the observations available when it was written, assuming the period never changes. If the orbit is shrinking, the period gets shorter, the planet comes round a little sooner each time, and the error builds up orbit after orbit. An old timetable then predicts the crossing too late.

The usual way to show this is O−C, “observed minus calculated”: when the planet actually crossed, minus when a given ephemeris said it would, in minutes. Negative means the planet came early.

When WASP-12 b crossed on this night, compared with 12 published timetables, by the year each was published. The shaded band is this night's own uncertainty, ± 4.2 minutes: a timetable inside it agrees with what I saw. The older the timetable, the further below the band it sits; the planet arrives early because its orbit shrinks. Error bars show each timetable's own uncertainty; the 2009 discovery timetable's is ± 27 minutes, so it runs off the chart. ExoFOP (hollow marker) is a catalog entry with no publication year, plotted at the date of its reference transit (late 2019); its period is the longest of the twelve, which is why it drifts so far. Hover for each value.

This night agrees with the newest timetable (Sodickson & Grunblatt 2025) to 0.4 minutes, and with the four from 2022 to 2024 to within 3 to 6 minutes, no more than 1.4 times the combined uncertainty. It is 11 to 21 minutes early against the four timetables published in 2016 and 2017, which is 2.5 to 4.6 times the combined uncertainty. That pattern is what a shrinking orbit predicts, and it is what larger telescopes have measured for years. One night can't tell how fast the orbit shrinks: that takes many precise timings spread over years. What this night adds is one more timing, from a 50 mm telescope, that lands where the decay says it should.

Check stars

Three stars of nearly the same brightness as WASP-12 (Gaia G 11.49 to 11.50) were held out of the comparison set and measured exactly like the target. Nothing should happen to them. If the comparison stars, the sky or the camera made a dip at the planet's mid-time, they would show it too. Fitting the same transit shape to each one at WASP-12's mid-time gives no dip: all three sit within about one of their own uncertainties of zero.

The dip at the planet's mid-time, for the target and for the three check stars, with each one's uncertainty. Hover for the numbers.

Fixed in advance vs measured

Fixed in advance (published, or set by the kit's rules before it looked at the light curve)	Measured from this night's frames
The 12 published ephemerides, and the planet's period. The fit starts from the newest one but searches a window of ± 2.6 hours around it, so a wrong timetable can't trap it.	Mid-transit time: 09:19:10 UTC (BJD_TDB 2461324.89018), ± 4.2 minutes
The orbit's shape: its size in star radii (3.0 ± 0.15) and its tilt (81.8° ± 0.5°), from published fits. The fit may move them only within those ranges; it ended at 3.02 and 81.8°.	Depth: 1.63 % ± 0.20 %
Limb darkening (how much dimmer the star's edge is than its center), from a coarse table by the star's temperature (6,265 K), held fixed.	Baseline: the slow drift of the light curve, fitted with the transit
The comparison-star rules: Gaia DR3 stars from two magnitudes brighter to one fainter than WASP-12, no bright neighbor, not flagged variable by Gaia, steady against the others; the three closest in brightness held out as check stars.	Noise: 1.85 % per frame, and the camera's gain, measured from the night sky in the frames
The rules for setting frames aside (cloud, blur, twilight), and the menu of six baselines the fit chooses from.	Which aperture size and which baseline: picked by the data, by rules written before the night

What went wrong, and the limits

Only 15 minutes before the transit began

I started the run about a quarter of an hour before ingress. The 60 minutes after egress are fine, but with so little “before”, the fit can't fully tell a transit from a slow drift. The fit's statistics prefer a baseline that is steady apart from a term following the airmass (the thickness of air the light passes through), and the headline uses it. A straight line in time fits almost as well and gives a shallower transit, 1.38 ± 0.19 %, with the same mid-time (+1.2 minutes). That spread is counted in the headline uncertainty. Next time I start at least 45 minutes before ingress.

The telescope didn't know where it was pointing, or at what

The pointing written into the frames was 22 arcminutes from the real center of the field, and the target was filed as “Unknown” with no name. The images themselves were fine: matching each frame's stars to the Gaia catalog gives its true position, and the kit found WASP-12 b by searching the field for known transiting planets. Next time I'll set the target name in the app.

The field jumps every few minutes

In equatorial mode the Seestar re-centers its target every few minutes, so the stars move up to about 100 pixels in a sawtooth. That keeps the target near the middle, which is the point. It also means software can't assume the stars stay put. My own first quick script lost track of them; the transit kit matches every frame to Gaia on its own and handled all 647.

Twilight and one thin cloud

Morning twilight washed out the last 19 minutes (49 frames from 11:47 UTC), an hour after egress, so it cost baseline, not the transit. One frame at 08:44 UTC caught a thin cloud and was set aside.

The noise was 3 to 6 times my guess, because I forgot the gain

Before looking I expected 0.3 to 0.6 % per frame. Each frame actually scatters by 1.85 %. I had left out the camera's gain: at the gain setting used (80), the S50 records about 15 counts for every photon its sensor detects, so WASP-12 delivers only about 4,500 detected photons per 20-second frame, and counting statistics alone make that 1.5 % noisy. The kit's noise model (photons from the star and sky, plus the camera's own read noise) predicts the scatter to within 5 %, so the night is close to the limit set by the light itself. Collecting more light would help (a bigger telescope, or more transits); tweaking the camera settings would not.

The clock

The frame times come from the telescope's own clock, which I didn't check against an independent source that night. A clock a few seconds off would not matter against ± 4.2 minutes; a clock minutes off would, and the agreement with the newest timetable argues against that, but it's an assumption, not a measurement.

How it was done

The frames went through our transit kit, its science code unchanged for this night. It matches every frame's stars to the Gaia DR3 catalog, so it knows where every star is in every frame despite the jumps. It measures WASP-12 and its neighbors in circular apertures on the raw color-sensor pixels, with no calibration frames, and divides WASP-12 by a weighted sum of 52 comparison stars picked by fixed rules. It fits the standard transit model (Mandel and Agol: a dark disk crossing a limb-darkened star) together with a baseline, choosing among six baselines with the Bayesian information criterion (a score that rewards a better fit but charges for each extra term). The uncertainties combine three parts: the range of plausible fits from an MCMC run, how much the answer moves when the comparison stars are resampled at random (a bootstrap), and the spread between the baselines the score can't separate. For the depth those are 0.16, 0.05 and 0.10 %; for the mid-time 4.1, 0.9 and 0.6 minutes. Times are converted to BJD_TDB, the standard clock for transit timing, which corrects for where Earth is in its orbit.

The kit's whole run, from raw frames to these numbers, took under ten minutes on a desktop PC.