HAT-P-32 b, watched live and measured over the whole night

The words on this page

Depth

How much of the star's light the planet blocks, in percent.

Mid-time

The middle of the transit. Ingress is when the planet's disk first touches the star's edge, egress when it leaves.

Timetable (ephemeris)

A published prediction of when transits happen: one observed mid-time plus a whole number of orbital periods.

O−C

“Observed minus calculated”: when the transit happened minus when a timetable said it would, in minutes. Negative means early.

BJD

Barycentric Julian Date (BJD_TDB): a count of days timed at the center of mass of the solar system, so the few minutes light takes to cross Earth's orbit don't shift it. Astronomers quote transit times this way.

Comparison stars

Nearby stars of similar brightness whose combined light is the yardstick. Dividing the target by them cancels haze and the star's changing height in the sky. Check stars are three more of the same kind, kept out of the yardstick and measured like the target: they should show no dip.

σ (sigma)

One standard uncertainty. A dip at 5.1σ is 5.1 times larger than its uncertainty; above about 5 counts as a clear detection.

Noise floor (null)

Treat each comparison star in turn as if it were the target, with the same mid-time and fit, and see how big a “dip” it shows. The spread of those fake dips is what this night produces with no planet.

Baseline

The slow drift of the light curve outside the transit, from the sky and the star's height, fitted together with the transit. Airmass is the thickness of air the light passes through, 1 straight overhead.

What was written down before measuring

Saved at 10:02 UTC on 21 October 2026, before any frame of this night was measured. Nobody had looked at this night, so it counts as fresh data: nothing about the method was tuned on it.

The choices, fixed before measuring

The live mode used the transit software's own steps and defaults. Everything that needs the whole night was fixed once, on the catch-up batch: the 429 subs already uploaded when the live run started, 06:51 to 09:34 UTC.

Comparison stars

Gaia stars from 2 magnitudes brighter to 1 fainter than the target, isolated, not flagged variable, with a clean Gaia position fit (RUWE < 2); drop any whose scatter is above 2× the median, repeated until none are; the three closest in brightness held out as check stars; each star weighted by its scatter, no one star above 20 %. Chosen once, then frozen.

Aperture and color

The aperture (the circle of pixels summed around each star) with the lowest frame-to-frame scatter over 8 sizes (0.6 to 3.0 times the star width, FWHM), and the camera's color channels, picked without looking at the transit. Chosen once, then frozen.

Flat and sky

No flat-field correction of any kind (the default). The sky is the clipped median of a ring around each star.

Baseline

No extra corrections: only baselines in time and airmass, picked by the Bayesian information criterion (BIC, a score that rewards a better fit and penalizes extra terms) from a constant, a straight line or a curve, each with and without airmass. This is the software's default and the fixed-time fit's setting.

Bad subs

The software's rule, re-run on everything so far at each update.

Result of record

The software's whole-night analysis with every default (a 3,000-step Markov-chain Monte Carlo run, MCMC, to map the range of plausible fits; a 100-draw bootstrap; 50 comparison-star subsets), run after the sequence ended.

The light curve

The depth at each published timetable

Here the software holds the published transit shape and mid-time fixed, fits only the depth and a baseline, and does it for every published timetable. Nine of ten land at 1.80 to 1.84 % and 7.5 to 7.7σ by their own errors. ExoFOP's timetable, which sits 17 min earlier with a 19 min error, gives 2.35 %.

Δχ² is how much better the fit gets with the transit than without it. The fixed-time fit at Kokori et al. 2023 also checks against the comparison stars: the spread of their fake “depths” is the night's noise floor (sd 0.36 %), and the target's 1.80 % is 5.1 of those spreads. A transit of the published depth, added to the same stars, came back at 1.00 of its depth. The comparison stars scatter 1.38× wider than their quoted errors. That factor is reported, not applied; with it, the time-free depth error would be ± 0.74 % instead of ± 0.54 %.

The live updates against the whole-night answer

The live mode ran from 10:05 to 12:16 UTC, with a plain least-squares fit only (no MCMC while live). Each update took 3 to 7 min, most of it listing the folder of uploaded subs. The newest sub always lagged: the upload times put the lag at 6.9 min or more, with a 28 min median.

The shaded bands are the whole-night result ± 1σ. On a narrow screen, scroll the chart sideways.

At 11:30 the baseline rule briefly picked a curve (a second-order polynomial), which absorbed the transit (fixed-time depth −0.15 ± 0.95 %). The live and final runs differ only in the frozen comparison stars (28 chosen on 429 subs, against 19 on 762), MCMC, and the bootstrap size.

Why the mid-time came out early

The choices written down in advance stay the result of record: changing the baseline rule after seeing the answer is exactly what fresh-data testing forbids. Everything in this section is reported beside the result, not instead of it.

• Timestamps: not the cause. Exposures run 20.6 s from the start stamp to the end stamp in each file. The software uses the midpoint, the same code that gave an O−C of +0.4 ± 4.2 min on the WASP-12 b night. As an independent clock, each sub's upload time minus its own stamp is never under 6.9 min (638 subs). A camera clock 20 min behind would make that impossible.
• The comparison stars: not the cause. The check stars are flat. Dropping each of the five most heavily weighted comparison stars gives −17.2 to −22.9 min. The target against each of the 28 comparison stars alone gives a median of −20.4 min (16 to 84 %: −29.4 to −10.3). The shape is common to every comparison star, so it lives in the target against the field: differential extinction (airmass 1.04 to 1.89) or a position or flat-field term.
• The baseline: the cause. The transit fills 213 of the 298 observed minutes, with 12 min before ingress. Any baseline with a slope or curvature trades against the mid-time, and a slope taken from the 78 subs after egress alone gives −10 min.
• Re-centering. From 10:43 UTC the Seestar re-centered 7 times. After some of those, 5 to 7 of the 45 stars fell off the S50's small frame.

These timing checks ran on the live measurements (650 subs, the 28 comparison stars the live run chose), not on the whole-night run.

The night

Airmass ran from 1.04 to 1.89, and the comparison stars' combined light fell steadily with it. The software set aside 11 subs: one at 07:41 UTC, where the field dimmed unevenly, and the last ten from 11:46 UTC, where the number of stars detected fell under 60 % of the night's level as the sky brightened at high airmass.

Proposed for future nights

Proposals, not applied to this night.

• Cap the baseline's complexity by how much of the night lies outside the transit: no curve without 45+ min outside the transit on both sides, and no slope without 30+ min on one side and 15+ on the other.
• Warn when a window leaves under 45 min before ingress or after egress. Our transit planner now warns on short baselines.
• An observation, not a rule: the comparison stars' 95 %-on-frame requirement is computed on the catch-up batch, so it can't see re-centering that only starts later and moves stars off the S50's small frame.