Finding a planet
you were told was there
One dip proves nothing. What makes it a measurement is that the dip comes back, on a schedule, and keeps coming back for years.
The folding trick
If a planet crosses in front of its star once a day, can years of public brightness measurements tell you exactly how long its year is?
A planet passing between us and its star blocks a sliver of the light — for WASP-18 b, about one percent, for a couple of hours at a time. One such dip proves nothing; a telescope has a thousand ways to flicker. What makes it a measurement is that the dip comes back, on a schedule, and keeps coming back for years.
The trick for finding that schedule is called folding. Guess how long the planet's year is. Cut the entire record into lengths of that guess and stack them on top of each other. If the guess is wrong, the dips land in different places and smear into the noise. If the guess is right, every dip lands on the same spot and a shape appears that no single observation could show you. The right guess is the one that makes the picture sharpest.
The lab did this with eight official TESS light curves — the same public files anyone can download from the mission archive — spanning six years of observations. It found the dip in one sector without being told where to look, then used the full six-year baseline to time 177 individual transits and refine the period from those timings alone. The published catalogue value was read only at the end, to grade the answer. Both the period and the depth landed inside the published error bars.
The panel below is the folding trick, live. It opens by sweeping the trial period across the whole search window on its own; take the slider whenever you like. Almost everywhere you put it, you get fuzz. Somewhere very near 0.941 days, the fuzz collapses into a planet.
left · the record as it arrives — — cadences, gaps where the spacecraft was doing something else · right · the same points folded at the trial period, with the binned profile drawn heavier over the scatter, and the box the search actually fitted in moss · below · the fold score across every trial period the sweep tried. The box depth readout runs low against the 1.04 % that was injected: a box has square edges and a real transit does not, so the window swallows some ingress and egress. That bias is real, and quantifying it rather than asserting it away is what A04's injections are for.
The scan panel is where the search stops looking easy. The tall spike is the planet. The shorter ones on either side are aliases — a trial period of twice the true one stacks the transits at two phases instead of one, and half the true one stacks every transit at phase zero but throws away half the phase coverage. Both produce a real dip. Neither is the answer. A search that took the first convincing dip it found would take the wrong one about as often as the right one.
The published values are 0.94145223 ± 0.00000024 d for the period and
1.041 ± 0.022 % for the depth. The measurement lands 0.00000013 d from the first
and 0.019 % from the second — inside both bars. It used eight official mission-produced
TESS SPOC light curves for TIC 100100827, each pinned by URI, byte count
and SHA-256, across sectors 2, 3, 29, 30, 69, 96, 103 and 104. Once those files are
local the whole thing runs in 7.1 seconds.
The 177 transits are the part that matters. A single fold gives you a period good to a few seconds. Timing each transit individually across a six-year baseline and fitting a straight line through the timings gives you one good to eleven milliseconds — and the scatter of those timings about that line, 0.79 minutes, is what tells you the line is real rather than fitted.