Guide to the data

Two kinds of finds live here:

By hand

Stars analysed properly in a notebook, with their own plots and a status such as Confirmed by me or Known (independent confirmation).

Automatic

Status New candidate, found by hunt.py (shown as "Found by: hunt.py"). It runs every light curve that NASA's TESS satellite recorded in a sector — hundreds of thousands of stars — through two period searches (Box Least Squares for eclipses, Lomb–Scargle for smooth variations) and a single-dip search.

A star only reaches this site if it passed every filter:

Not known

No entry within 30″ in the main variable-star catalogues: VSX (AAVSO), ASAS-SN and ZTF.

Not a glitch

Its period isn't shared by an unusual number of other stars, and (for single dips) other stars weren't dipping at the same moment — both signs of scattered light or spacecraft events.

Not a neighbour

No other flagged star within 3′ shows the same period.

Strong enough

Single dips must last several data points and stand at least 20× above the noise.

Only the best-scoring survivors are published (the ★ counter at the top shows how many, out of how many found). Even so: a candidate is a lead, not a discovery. Most will turn out to be known under another name, blended light, or instrument effects the filters missed.

Eclipsing binary

Two stars orbiting each other, taking turns blocking each other's light. Flat light with sharp dips, often two of different depth per orbit. The most promising automatic type.

Transit candidate

Shallow, box-shaped dips with no second dip. Usually a shallow eclipsing binary; occasionally something planet-like.

Periodic variable

Smooth, continuous brightness changes: pulsating stars, spotted rotating stars, or contact binaries (for those the true orbit is usually twice the period shown).

Single dip / odd event

The star dimmed once and didn't repeat within the ~27 days TESS watched. Could be a long-period binary, a single transit, a dusty young star — or a glitch. Treat with the most suspicion.

Some finds show a picture of what the two stars could look like up close. Each one is made in two steps:

1. From the data

A 3D layout is built in Blender: the stars' sizes, colours (from their temperatures), spacing and the tilt of the orbit come from fitting a model to the measured eclipses plus Gaia's temperature.

2. Painted by AI

An image AI (DreamShaper XL, a Stable Diffusion XL model) paints over that layout in the style of NASA/ESA artist's impressions. It runs locally on a home PC (AMD Radeon RX 7800 XT) — no cloud services.

So the arrangement is real, but the surface texture, glow, light rays and background are the AI's artistic additions, not observations. If the AI wasn't available, the plain 3D render is shown instead; the label under each picture says which it is.

They're made automatically only for the strongest eclipsing-binary and transit candidates, and only when the model matches the eclipses well and there are no warnings. A picture of a candidate is still a picture of an unconfirmed object. The model-fit plot next to it shows how well the model matches the data. Tick ✦ With impression on the front page to see them all.

Period (days)

Time for one full cycle. For eclipsing binaries found automatically it can be off by a factor of 2 — the plot shows both P and 2P folds so you can judge.

Eclipse / dip depth

How much of the light disappears at the deepest point (e.g. 5% = the star drops to 95% brightness).

Amplitude

Spread between the bright and faint ends of the light curve.

Detection score

How far the signal stands above the noise. Higher is stronger; below ~20 is weak. It ranks candidates, it doesn't prove them.

TESS mag / Gaia G mag

Brightness. Larger numbers are fainter; 13 is about 600× too faint to see by eye.

SIMBAD name

The star's best-known name in SIMBAD, the database of every star mentioned in published research, with its type in brackets (e.g. EB* = eclipsing binary, * = ordinary star). Click it to see the papers.
“not in SIMBAD” means SIMBAD has no record of any study mentioning the star — normal for most 12th–13th magnitude stars, and a good sign for a new find.
A note saying SIMBAD already lists it as a variable means it was probably studied after all.

Nearest known variable

The closest star in the VSX variable-star catalogue within 5′, with its type, period and distance; click it for its VSX entry. TESS pixels are 21″ wide, so a bright variable within about a minute of arc can leak its light into our star.
“(same period!)” means it varies on the same period as our candidate: almost certainly the signal is that star's, not ours.
“none within 5′” is the best case.

Data used

Which TESS sector(s) the signal comes from. A sector is ~27 days of watching one patch of sky.

Coordinates

RA / Dec in degrees (J2000). The Look it up links open the position in SIMBAD, VSX, ESASky (sky images) and ExoFOP (everything known about the TESS target).

A quick way to separate real stars from junk:

1. Warnings

Any "same period" or SIMBAD warning in the notes? Then it's almost certainly known or blended.

2. The plot

Clean, repeating dips at the same phase in the folded panels? Ragged, one-sided or edge-of-data dips are suspicious.

3. Neighbours

Open ESASky: is there a brighter star very close?

4. Other data

Check other TESS sectors, ZTF or ASAS-SN light curves for the same period.

5. Wider search

Search VSX and SIMBAD with a 1′ radius. Still nothing? Then it may really be new — worth a careful hand analysis and, eventually, a VSX submission.

Download everything on a find gives a zip with all its files. FITS files are the original TESS light curves (QLP pipeline); CSV light curves have columns time_btjd, flux, flux_err (BTJD = BJD − 2457000); meta.json holds every value on the page.

/api/finds.json lists every find with its metadata and file URLs, for use from Python: requests.get(url, auth=(user, password), verify=False).