The Radiant Watch
Karen Oduya has run her Pi-camera station from a corrugated shed on her property outside Geraldton for six years. The Eta Aquariids peak tonight, and the back-end will not stop emailing her about the third detection of the third hour.
The shed smelled of dust and ozone, and Karen Oduya was already three cups into her thermos when she got the second email.
She had named the station Mara, after the river south of where her mother had been born, and Mara had been a good neighbor. Six years of clear nights and intermittent rain. Twelve thousand and change confirmed meteors. One hundred and eighty-seven trajectory pairs in the last twelve months alone, matched to other stations across Western Australia and New Zealand by the back end every morning at 0743 local time, give or take. Karen knew the exact arrival time of the morning report because for six years it had been the most reliable thing in her life.
Tonight was different. Tonight Mara was emailing her in real time.
"Anomaly review requested," the second email said, and a thumbnail loaded underneath the subject line: a curved track of bright pixels arcing across a frame, with the LSTM confidence score listed at 0.94 and a small flag glyph in the corner that meant "candidate not associated with any catalogued shower." Below that: "back-end cluster miner: provisional cluster K-2026-EA-7, member count 4, similarity score above threshold."
Karen put the thermos down.
She had only ever seen this email twice before. The first time, four years earlier, the candidate had washed out under follow-up — the cluster fragmented when more trajectories came in and the proposed parent body turned out to be a lump of nothing. The second time, two years later, the miner had retracted the flag inside of forty-eight hours when a station in Chile reported the same orbit and the cluster collapsed into an already-known minor shower. Both times she had been ready to be quietly proud of Mara, and both times the network had said, calmly and without embarrassment, that no, this one was a coincidence after all.
The third time, she thought, sitting in the cooling dark of the shed with the laptop screen on her knees, might be different. The radiant elevation was good. The peak was an hour away. Mara had eight hours of clean sky ahead.
She walked outside.
The Eta Aquariids did not announce themselves the way some showers did. There was no drumbeat moment, no swarm. They came on slowly, a streak every minute or so during the early hours and then more, and more, the radiant climbing out of the eastern horizon as the night turned. Karen had learned to watch for the persistent trains, the long shimmering ghosts that the brightest Eta Aquariids left behind. Halley dust at sixty- six kilometers per second. The nucleus that had thrown those grains had last been here in 1986, before her mother had been a citizen of any country, when Karen herself had not been an idea. The grains had been in free orbit for forty years, waiting for Earth to come around and meet them.
She watched a meteor go.
It was a clean, golden one, low in the east, with a train that hung in the air for almost a second after the streak itself was gone. Somewhere on her roof, in the corrugated shed, in the IP66-rated housing that her ex-husband had bought her one Christmas and that she had refused to let him take with him when he left, Mara had seen the same meteor and was already running it through three classifiers. Random forest, CNN, LSTM. By tomorrow morning the bundle would be in the Slovenian back end and some volunteer she had never met would be looking at the orbit. She found this comforting. The world ran, in places, on machinery that nobody had asked anyone for permission to deploy.
Her phone vibrated.
It was the dashboard alert this time, not the email. Cluster K-2026-EA-7 had a new member. Five meteors now. Karen felt her stomach do a small, neutral, professional thing — the same sensation as catching a test flake at work, except this was not a flake.
She went back inside.
The shed at night was cold in the way that desert sheds are cold. Karen sat on the upturned bucket she used as a chair and pulled the laptop closer and looked at the cluster page. Five trajectories now, all recovered from station pairs in Australia and New Zealand, all converging on a heliocentric orbit that — she scrolled — had a semi-major axis past forty AU, an eccentricity at 0.971, an inclination of 78 degrees. The similarity scores against every catalogued long-period parent in the database were too low to associate. The miner's flag had moved from provisional to candidate.
The page also said, in small grey text below the orbit, that the candidate had been forwarded to the optical-survey notification list at 0228 local time, which was nine minutes ago. There was a list of addressees. Karen recognized two of the names. One was a Canadian post-doc she had emailed twice to ask about reduction software. The other was the lead of a Pan-STARRS team Karen had once attended a seminar of, in person, in Brisbane, in 2022, and to whom she had not spoken since.
She watched the page for a long time.
Outside, at three in the morning, another bright meteor went. The shed hummed. The fan in the back of the laptop turned on and off the way a small animal breathes. Mara had ingested the new frame. Frame differencing, threshold, classifier, trajectory upload. The bundle would go to the back end at 0512. The back end would integrate it into the running cluster. If the cluster held, the candidate would survive into the morning. If it did not, it would not.
Karen was not, she discovered, refreshing the page. She was waiting.
She thought about the talk she had given at the Geraldton Astronomical Society in 2024, the one where her hands had shaken and she had still gotten through the slides. Twenty-two members in the audience. Two of them had been older than her father would have been, if he were still alive. One of them had asked her at the end whether all of this was worth it, the staying up at night, the Pi rebooting itself when the wind caught the antenna, the back-end emails that she sometimes had to explain to herself before she could explain them to anyone else. Karen had said, then, that it was — but she had said it the way one says yes to a question that one is hoping not to be re-asked.
Tonight she did not feel that way. Tonight, four meteors had agreed with each other on a heliocentric orbit that did not match any known parent body. The sky had told her something that the sky had not previously told anyone. Or, more accurately, the sky had told several hundred Pi cameras across two continents the same thing, and the ML pipeline that those Pi cameras fed had heard a voice in the noise that was the right shape to be a comet that nobody had imaged yet.
Mara would get a note in the credit line, in the methodology paper, if this held. Karen knew this because she had read every methodology paper GMN had ever published. The credit line listed the participating stations. Mara had been listed in the last three of them. Karen liked this in the small, precise way she liked Friday afternoons and the sound of the fridge cycling on.
The page refreshed itself. Six members in the cluster.
She got up, slowly, and went to the door of the shed and opened it. The sky was the kind of black that you only got in a dark-sky region in the absence of moon — except tonight there was a moon, a waning gibbous one, hanging in the west and rinsing out the dimmer stars. The Eta Aquariid radiant was high enough now to see clearly. The dim stuff was lost. The bright ones came through.
She watched two meteors in three minutes. Long, bright, golden, with trains.
Her phone vibrated again. She took it out and looked.
"Cluster K-2026-EA-7: notification dispatch confirmed. Pan-STARRS team has acknowledged. Estimated next-imageable epoch for proposed orbit: 2026-08-22 to 2026-09-14 window."
Karen sat down on the bucket.
She did the math in her head, slowly. The proposed parent was at, what, forty-five AU. Imageable in late August, three and a half months from now, at maybe twenty-fourth magnitude, on a deep-image survey cadence. Pan-STARRS could probably reach that. ATLAS might, in stacked images. Vera Rubin, with the new wide-field camera, would crush it.
The point was that the sky had a question now, and the question had a date attached to it.
She looked up. The meteors kept coming.
Karen had thought, when she was younger, that the way one made a contribution to anything was by being the brightest person in the room. She had spent years in rooms where she was not the brightest person, and the years had taught her that brightness was not the relevant axis. The relevant axis was patience. The relevant axis was being one of a thousand stations on a clear night, and showing up. The relevant axis was that the sky above her shed in Geraldton had collected, through her camera and her Pi and the open-source code that maintainers in Slovenia and Croatia and Canada had kept compiling for ten years, a piece of evidence about the solar system that nobody had yet seen.
The cluster might not hold. Karen knew this. She had been here before. Two times in six years, the morning had come and the back end had said, politely, that the candidate had retracted. She had not minded then. She would not mind tomorrow.
What she liked, watching the radiant climb, was that the question had a date. August through September. Three and a half months. If the candidate held, someone on a survey team would point a much, much, much larger telescope at a piece of sky that her shed in Geraldton had nominated, and a frame would come back, and either the comet would be there or it would not be there. Either way, the question would be answered. Either way, the sky would have known something tonight that nobody else had known.
She went back to the laptop.
The cluster had eight members.
Karen Oduya — software engineer, divorced, citizen astronomer, the woman who had once told her ex-husband that she would rather sit in a cold shed at three a.m. than sit on a couch and pretend to be interested in things she was not interested in — closed the laptop and stood up and stretched. The shed creaked. The fan went off.
She walked back outside, into the very specific kind of cold that the desert was at three a.m. in May, and she watched the eastern sky.
A meteor went, low and golden, with a train.
"Hello," she said, quietly, to whatever it had been. "Your aunt or your grandmother says hello back."
She stood there for a long time.
If you like science from the edges, see today's deep-dive on edge AI and citizen meteor networks and the running prediction on ML-driven comet discovery through 2027.