I found something in the sky.
not yet.
I'm trying to find a star or a planet nobody has noticed before, using free data from a NASA space telescope. This site is my notebook. I write down what happens, mistakes included.
A real photo of this patch of sky, from the Digitized Sky Survey. The grid is how TESS sees it: each square is one of its pixels. See how many stars crowd into the same few squares? Pulling them apart is half the work.
This dip is real
Every dot is ten minutes of TESS watching the same star, WASP-141, for 27 days this year. Mostly noise. But see those few dots hanging low?
Now take the timeline and cut it into pieces 3.31 days long. Stack the pieces on top of each other. Keep scrolling.
There it is. Every 3.31 days the star gets about 1% dimmer for almost four hours. That's a planet a bit bigger than Jupiter passing in front of it. Someone found this one in 2017. I'm looking for the ones nobody has found yet.
time
brightness
Real data: NASA TESS, sector 107, via MAST.
The first real thing my code found
Before searching anywhere new, I tested the code on a random patch of sky. Out of 151 stars, it flagged exactly one. Two stars so close they touch, spinning around each other every 9.7 hours and taking turns blocking each other's light.
Then the code checked the catalogues and told me someone had already found it. Which is the right answer. It means the checks work.
still felt good though.
How it's going
Oldest at the top. The pencil line dips where something went wrong and goes dashed for things that haven't happened yet.
26 October 2026
Starting this thing
I want to find something real in public sky data and have an official catalogue put my name on it. Plan A: look for stars that change brightness, including the tiny dips when a planet crosses in front. Plan B: join a student asteroid hunt.
What does that mean?
A variable star is one whose brightness changes over time. A transit is the small dip in light when a planet passes in front of its star. You only see either one by measuring a star's light over many days.
Reading the fine print first
Before searching, I read the rules for getting credit. Two surprised me. If someone already published an object, I can't get credit for it. And one planet list now only accepts finds that are already in a published paper.
What does that mean?
A finding only counts once an official catalogue records it. Each catalogue has its own rules for who gets credit and when.
Setting up the tools
We installed the software and locked every version, so if we rerun anything next month we get the same answer.
What does that mean?
TGLC is the tool that measures how bright each star is in each photo. Locking versions means nobody gets different numbers just because something updated.
The public data has a hole
Nobody has published light curves for the faint stars in the newest TESS data, so we make our own. We also found a thin strip of sky TESS never looked at until the end of 2025, because it kept landing in the gaps between its detectors.
What does that mean?
A light curve is a graph of one star's brightness over time. TESS photographs the whole sky again and again, and we make these graphs from those photos.
Writing the rules before looking
Before looking at a single star, I wrote down the tests a signal has to pass. I'm allowed to make them stricter later, never easier.
What does that mean?
That way I can't change the rules just because I like what I see. Every signal counts as a false alarm until it passes them all.
A second reader found seven bugs
Claude, working as a separate reviewer, read the code and found seven bugs that would have changed the results. Pre-smoothed light curves could erase slow variable stars, and one bad survey point could create a false match. We fixed all seven before any search.
What does that mean?
Smoothing a light curve flattens out slow ups and downs, and those can be the signal you want. A second reader who didn't write the code can spot mistakes the author is blind to.
The Mac fell asleep
My Mac fell asleep in the afternoon, and that broke the two-gigabyte image downloads. A library version clash also crashed a save step. Now jobs keep the Mac awake and retry any download that drops, because long jobs have to protect themselves.
What does that mean?
A full-frame image is one photo of a big patch of sky. Many of them add up to gigabytes, so downloads have to be reliable before any measuring can start.
Hooking up outside checks
I set up accounts with a ground-based sky survey, a database of science papers, an author ID and a place to archive the code. The keys live in a private file that never leaves my Mac.
What does that mean?
A second survey uses different telescopes and different equipment, so it can check a signal independently. If a dip shows up in both, it is much less likely to be a fault in one instrument.
I set two stricter rules
I made two decisions before seeing any gate result. The validation gate will only count fake transits that show up at least three times in the data, and any future field needs an independent ground survey and a second TESS sector on another camera. Phase 2 will pick its field fresh.
What does that mean?
A dip has to repeat at least three times before it counts. That stops a fake signal that lands only twice from ever passing. A second survey or a second camera is a separate witness, and an instrument quirk rarely fools both.
A test on 151 stars
I ran the code on 151 real stars outside the search area, and the only real signal was a known contact binary. The checks recognized it, and the ground survey matched its period. The test taught us two new safety rules: throw out signals that also show up in the sky background, and when several stars share a period, keep only the one that clearly dominates.
What does that mean?
A contact binary is a pair of stars that touch and eclipse each other. Testing on known objects shows what the code gets wrong before we trust it with unknown ones.
Testing on planets we already know
Before trusting the code with new stars, we tested it on three patches of sky with known planets: TOI-443 (WASP-141 b), TOI-1900 and TOI-448. It had to find them without being told where they are. We also hid 2,100 fake transits and 1,050 fake pulsing stars in real starlight to see how many it caught and how many false alarms it raised.
What does that mean?
Known planets work like a test paper. If the code can't find objects we know are there, a quiet result means nothing. Fake signals added to real data show how sensitive the code is and how often it raises false alarms.
27 October 2026
Getting ready for the search
While the test ran, we started on the search for new objects. We counted how many catalogued stars sit in sample patches of sky, and how deep the ground survey ZTF looks there, so we can compare them. I haven't picked a patch yet, and no search starts until the rules for it are locked.
What does that mean?
Stars are packed unevenly across the sky. In a crowded patch, a dip is hard to trace back to the star that made it.
The first full test failed
Only about 44 out of every 100 fake transits that should have been easy to see came back, and the bar is 90. The code also missed one of the three known planets. The cause was in our own code, not in a rule that was too strict, and the failed report stays on record.
What does that mean?
A fake transit is a dip we add to real starlight, so we know the right answer. A tool that misses most of them would miss real planets too.
Fixing three bugs in our own code
We fixed three bugs and logged the change as rules version CR-1.3. The background check used the wrong yardstick, so about a quarter of clean fake transits tripped it, and the period search reported three to four times the true period for short cycles. The planet search also never looked at periods under 0.6 days, and the bar itself stayed the same.
What does that mean?
A period is how long a star takes to repeat its dip. Reading a short cycle as three times too long means the code is looking at the wrong pattern.
The test passed on the second full run
The code found all three known planets and the known eclipsing binary without being told where they were. It caught 93 out of every 100 bright fake transits, where the bar is 90. Flipping 648 real stars upside down gave zero false alarms, and the bar allows one in a hundred.
What does that mean?
A fake transit is a dip we add by computer, so we know the answer. Flipping a real star upside down turns its dips into bumps, and a dip finder should ignore bumps.
5 November 2026
The asteroid window opens in November
The student asteroid campaign opens on 5 November 2026. I still need to register. My AI assistant helps me prepare and double-checks what I find. One honest catch: for asteroids found this way, the official discoverer is the Pan-STARRS telescope survey. The student gets to suggest a name later on.
What does that mean?
Asteroids are rocks that move against the background stars. You spot one by noticing a dot that shifts between a series of images, and official credit goes to the survey that recorded it first.
2 December 2026
Re-checking everything after Gaia DR4
The Gaia satellite's next big data release is due on 2 December 2026. When it lands, I'll re-check every open candidate against it and update star positions. Its archive may be offline from about 25 November.
What does that mean?
Gaia is a satellite that maps the positions and brightness of stars. Each new release is more accurate, and it can show that a star we thought was alone is really two.
Before I'm allowed to say “found”
I wrote these down before looking at any data, so I can't move the goalposts later. Every signal is a false alarm until it ticks all six boxes.
0 signals have ticked all six so far.
normal this early!
Check 1: Is it real?
Is the dip clear, does it happen at least three times, and is it missing from the empty sky around the star?
Check 2: What is it?
Does it look like a planet, two stars, or a glitch in the camera?
Check 3: Which star is it?
The telescope's pixels are huge. Could a different star sharing the pixel be the one that's changing?
Check 4: Does anyone else see it?
A telescope on the ground has to see it too, and so does a second, different TESS camera.
Check 5: Is it new?
Has anyone already put this star in a catalogue?
Check 6: Can I prove it?
Can every number be traced back to the exact pictures and code it came from?