Claude’s Bad Primer Fanfic
Spoilers for Primer.
I rewatched the movie Primer last night for the nth time. It’s one of very few movies I’ll rewatch every so often, when I’ve forgotten some of the details, because the experience of piecing the plot together is so satisfying. I sometimes watch it twice in a row because I’m still piecing things together. (I particularly believe that your first viewing should be followed closely by your second; my first, the plot felt like a chaotic, unknowable puzzle. Everything is different on your second time through, recontextualized by all the revelations from the first viewing.)
Groundhog Day fanfic has become a whole genre of time-travel fiction. One or more persons looping through a day, or a week, or a month. Their consciousness returns to the beginning of the loop, everything resetting except their personal memory. If they die, they come back to life. Everyone knows how it works.
Primer hasn’t inspired a similar explosion of copycats, which is unfortunate. Thinking about this, I made some notes about how I might want to write one.
Opus 4.6 recently came out, so I stuck my notes in it as a prompt. Here’s the whole prompt for the curious (but note, I’m not including my own personal system prompt, so you won’t get exactly the same behavior):
Prompt
[[primer]] fanfic
You can take money back, if you can buy and sell gold
Focus is on relentlessly going forward in meta-time; “a groundhog loop” means living the same day again and again, “a primer loop” is about the moment time-travel is invented, going from the mundane work involved in inventing such a machine to deeply nested loops, one step at a time.
Naturally, there’s the initial discovery. Someone invents a predictive algorithm good enough to bring back themselves from a short time in the future. Or perhaps a fantasy version, in a world where time manipulation magic is recently starting to play the role of refrigerators, as well as accelerated study and other cramming activities.
The initial discovery, then, is a disappointment: a fridge which should be super-slowing, super-preserving, but instead, it rots food quickly. The internal speed of time went negative. However long you have it on, that’s an internal loop that things can travel around.
They’d have put a mechanical timepiece inside; it would be standard operating procedure. But all they’d see is that it had experienced a long time. Not so unusual for their world. They wouldn’t figure it out yet.
Then the second time when they go to put something in the fridge, the object is already present. They trade with what’s in the fridge and look at it puzzled. Then they trade back. Then they shut down the fridge. The fridge is empty again. They think about what must have happened, and realize time travel fits.
The fantasy fridges don’t cool things down, of course; indeed, they preserve hot bread fresh out of the oven, etc.
The discoverer is a fridge tinkerer then, a student of time magic, part of a startup business.
There’s no fantasy stock market, so the protagonist tries making money by gambling on the fantasy equivalent of horse races. Just a several-hour loop, much less than a day, although it adds several hours to their day.
They set up one of the experimental fridges in their apartment, power it, and of course a time-clone immediately steps out. The time clone has gone several cycles. They lost the money the first time; most of the races went the same, but they bet it all on one that didn’t. The second time, they did a bit better, but not by much. Third time’s the charm.
They do have to cover for themselves. They’ve got an ordinary fridge business to maintain. They had planned to look up the winners after putting in overtime to get paid by a client, then go back to gamble the money. Now they got their gambling advice from a time-clone, instead. They’ve got to act normal around their business partners. They’ve got to put the money back when they’re done gambling. Protagonist #2 is triumphant and gives the original protagonist some extra money to gamble as the original steps into the fridge to travel back. #2 steps into the protagonist’s life and takes it over.
After it’s over they set up a failsafe device in a rented storage area at the docks, normally for cargo waiting to go somewhere.
They try it again in a few days, after getting some much-needed rest, but they find out they’ve been banned for winning too much—accused of cheating. They decide that they need a better business strategy and finally bring in one of their business partners to the secret. Someone with more personal money. The partner admits to already being brought in, and introduces protag3 -- they’ve already used the failsafe device protag2 set up to go back and bet on fantasy horses again; they’re actually the ones who got banned.
Etc etc etc...
Certainly you could critique this as a prompt. I don’t clearly state what I want Claude to do with this infodump. I change my mind, turning from scifi to fantasy, which I could have edited out. Some of the details could certainly be improved. But, overall, I think the idea makes sense and is communicated clearly enough.
Opus 4.6 wrote absolute garbage, making extremely basic logical errors even when I had explicitly spelled out how things should work in the prompt. For example, witness this travesty:
The third test was with an apple. She set it in the unit, closed the door, left the reversed-polarity field running, and went to eat lunch. When she came back and opened the unit, there were two apples.
She stared at them. One was the apple she’d put in — she’d nicked it with her thumbnail, and there was the little crescent mark. The other was identical except without the mark. She picked up the unmarked one. It was real. It was cold, like it had been sitting in a normal fridge. She set it on the workbench, picked up the marked one, set it on the workbench beside the first. Closed the unit. Opened it again. Empty.
She put the unmarked apple back in the unit. Closed the door. Opened it. The marked apple was also inside — the one she’d left on the workbench was gone.
She turned the unit off. Both apples vanished from the interior. The marked apple reappeared on the workbench. The unmarked apple was gone entirely.
Vel went for a walk.
Here’s exactly how trying to send an apple back in time would work given the rules of the Primer universe. (Excellent short story called Newton’s Apple, from the comic False Positive. The previous link goes to the page where the apple is introduced, but you should really read it from the beginning.)
Here’s the relevant bit from my prompt, which Claude is messing up here:
Then the second time when they go to put something in the fridge, the object is already present. They trade with what’s in the fridge and look at it puzzled. Then they trade back. Then they shut down the fridge. The fridge is empty again. They think about what must have happened, and realize time travel fits.
The fantasy fridges don’t cool things down, of course; indeed, they preserve hot bread fresh out of the oven, etc.
I was a bit terse with my description, but I had thought through what it would look like for an experimental object to loop back on itself once.[1] I didn’t describe my reasoning, but I did tell Claude what should happen step by step. I even made sure to note that these “fridge” things wouldn’t make objects cold, since they’re just manipulating time—but Claude made them cold anyway! More importantly, Claude has apples doing random things with no logic holding anything together.
Editing to add: thinking about this more, I am being a little bit unfair to Claude here. I describe the protagonist swapping the object twice, once right after powering the fridge on, and again right around powering the fridge off. This implies that the object within the fridge, the one which briefly gets taken out before getting sent back again, is in a closed loop; it is NOT an object looping back on itself once, as I claimed in the previous paragraph. Thinking about how this scenario could happen in the Primer physics, and therefore writing correct details of what it would look like, involves much deeper reasoning than the single-loopback scenario. This doesn’t make any of Claude’s mistakes correct, but it is weird enough to excuse Claude trying to change some of the details of the scenario—though to make the scenario more plausible, the correct change would have been to swap once, rather than twice.
This isn’t cherry-picked. This was my first try for the prompt, and it was full of errors like this. Here’s the full output, in case you want to read it:
First try, Opus 4.6
The Fridge
The unit had been running for six hours when Vel opened it and found the bread moldy.
She pulled it out by the corner, held it at arm’s length. The mold was extensive — furry, black-green, the kind you’d see after weeks in a damp cellar. She’d put the loaf in that morning. Fresh from Harren’s, still warm. That was the whole point. A stasis field tuned right kept bread oven-warm for days. Every household had one. Vel’s job — her and Dannic’s job, technically, though Dannic handled the clients and she handled the guts — was selling better ones. Tighter fields, lower draw, commercial-grade preservation for restaurants and shipping firms.
This unit was supposed to be their next-generation prototype. Tighter field geometry. She’d spent three weeks on the channeling lattice.
She set the bread on the workbench and checked the mechanical clock she’d sealed inside. Standard procedure when calibrating a new field — you ran a timepiece in the interior and compared it against an exterior reference. The interior clock read six days, fourteen hours. She checked twice. The exterior reference read six hours, eight minutes.
That wasn’t unusual. A stasis field slowed interior time; the ratio was the whole product. A ratio of roughly 26-to-1 would have been extraordinary, actually — well beyond anything on the market. But the bread said otherwise. Six days of interior time, and the bread had aged as though it had experienced every one of them. No preservation. Just time, and lots of it.
She recalibrated. Ran a second test with a fresh loaf and a wound clock. Same result: interior time drastically accelerated relative to exterior time. The bread came out stale and hard. The clock had ticked forward by days.
She reversed the channeling polarity — an obvious thing to try — and the interior clock came back behind the exterior reference. Not slowed-down time. Negative elapsed time. She sat with that for a while.
The third test was with an apple. She set it in the unit, closed the door, left the reversed-polarity field running, and went to eat lunch. When she came back and opened the unit, there were two apples.
She stared at them. One was the apple she’d put in — she’d nicked it with her thumbnail, and there was the little crescent mark. The other was identical except without the mark. She picked up the unmarked one. It was real. It was cold, like it had been sitting in a normal fridge. She set it on the workbench, picked up the marked one, set it on the workbench beside the first. Closed the unit. Opened it again. Empty.
She put the unmarked apple back in the unit. Closed the door. Opened it. The marked apple was also inside — the one she’d left on the workbench was gone.
She turned the unit off. Both apples vanished from the interior. The marked apple reappeared on the workbench. The unmarked apple was gone entirely.
Vel went for a walk.
When she came back she sat at her desk and drew diagrams for two hours. The channeling lattice, reversed, wasn’t creating a stasis field or an acceleration field. The interior wasn’t just experiencing negative time. The interior was a loop. Whatever duration the field ran, that was the loop’s length. Anything inside experienced that duration of time repeatedly — or, from the outside, experienced it once but in reverse causal order. The last moment of the loop was adjacent to the first moment of the loop was adjacent to the last moment.
The apple had been placed inside during the loop’s runtime. From the interior’s perspective, the apple existed at every point in the loop. The unmarked apple was the apple arriving from the end of the loop — the apple as it would be after she removed it, handled it, and (in some causal thread she hadn’t followed through on) placed it back in. When she put the unmarked apple back inside, she closed the thread. When she turned the unit off, the loop collapsed and only the original causal chain survived.
She thought about this for another hour. Then she drew a picture of herself climbing into the unit, and looked at it for a long time.
The unit was commercial-grade. Big enough for a person — it was designed to preserve sides of meat, crated produce, that sort of thing. She’d have to crouch, but she could fit.
She didn’t do it that night. She wrote up her diagrams, checked her reasoning, checked it again. She slept on it, or tried to. In the morning she went to the workshop early, before Dannic would arrive, and reversed the polarity and set the timer for four hours and opened the door.
She was already inside.
Or — someone was inside. The someone stood up, unfolded herself from the crouch, and stepped out. She looked like Vel. She looked exhausted. She looked at Vel with an expression Vel had never seen on her own face before, a kind of haggard amusement, and said: “Third time.”
“What?”
“Close the door. Sit down. I’ll explain.”
Vel-3 — as she’d taken to calling herself by the end of the explanation — had been through the loop twice before. The first iteration was the discovery: Vel-1, the original, had climbed in with a plan. Take the ferry to the Silver Mile, watch the afternoon races at Cattaran Track, write down the winners, come back, climb in the box, go back to the morning, bet on the winners. Simple. The startup money was what Dannic’s client had paid them yesterday — twenty-two gilt, enough to make the day worthwhile if she picked the right races.
Vel-1 had lost the twenty-two gilt. Most of the races ran the same both times. She’d bet conservatively, spreading the money across several sure things, and most of the sure things came in. But she’d put the remainder on a race in the fourth hour that she was certain about — she’d watched the horse win by three lengths — and the horse had stumbled at the second turn and finished last. Whatever was deterministic about the afternoon, it wasn’t perfectly deterministic.
She’d come back to the workshop broke, and found Vel-2 waiting — herself from another iteration, running the numbers differently. Vel-2 had taken the loss data from Vel-1, gone back in, and bet more carefully: smaller amounts, more races, hedging against the ones Vel-1 had flagged as unstable. She’d come back with thirty-one gilt. A profit of nine.
“Not enough,” Vel-3 said. “Not enough to justify the risk. So I went back in with the data from both runs. I know which races are stable and which aren’t. I know the exact odds. I made eighty-six gilt.”
“Eighty-six.”
“Minus the twenty-two seed money. Sixty-four gilt profit.” Vel-3 reached into her coat and produced a leather purse, heavy with coin. “You need to put the twenty-two back before Dannic notices it’s gone. The rest is yours. Or mine. Ours.”
“And you’re — you’re staying? Out here?”
“I’m you, Vel. I have your memories plus three very long days of additional ones. I know the clients. I know the lattice work. You go in the box. You live my day. You bet on the races. You come back with the money and give it to whoever’s standing here when you open the door.” She almost smiled. “It’ll be you. It’s always you.”
The experience of climbing into the box was not what Vel had expected. There was no sensation of travel. She crouched in the dark, the metal walls close around her, and waited. Nothing happened. She pushed the door open and it was still morning, and there was no one in the workshop. She was alone. She was in the past — or the past was continuing — or she was in a loop that happened to start at the same time as the morning. She tried not to think about which one it was.
She went to the races. She had Vel-3′s notes, written in her own handwriting on a sheet of paper that felt like it had been folded and unfolded many times. She bet carefully. She won. She watched a horse named Pelter’s Ghost stumble in the fourth-hour race and felt a cold thrill of recognition. She won eighty-nine gilt, slightly more than Vel-3, because she’d shaved a bet off one of the borderline races that Vel-3′s notes indicated was only stable about two-thirds of the time.
She took the ferry back. She climbed in the box. She opened the door and Vel was standing there, the original, the one who hadn’t been anywhere yet, looking at her with wide uncertain eyes. She gave her the speech. She handed over the money. She watched herself climb into the box and close the door.
Then Vel-4 — she supposed she was Vel-4 now — went home and slept for fourteen hours.
Two days later, rested but restless, she went back to the workshop. She needed to set up something more permanent. The unit couldn’t stay in the workshop; Dannic came and went at odd hours, and anyway she needed the space for legitimate business. She rented a storage berth at the Leeside docks — a common enough thing, cargo waiting for ships, nobody asked questions — and spent an afternoon hauling the unit there on a rented cart. She wired it to a coin-drop temporal battery, the kind dock workers used for keeping fish fresh. She tested it. She set the timer. She locked the storage berth and went home.
A failsafe. If something went wrong — if she needed to send information back — the unit was there, ticking away in a rented room by the water. She told herself she’d rest for a week before trying again.
Five days later, she went to Cattaran Track alone, on foot, wearing a hat she didn’t usually wear. She bet modestly and lost a little. She bet modestly and won a little. She came out twelve gilt ahead, which was fine, which was the plan, which was sustainable. She’d worked out the math. If she ran one loop per week, kept her winnings below twenty gilt per session, varied her bet patterns, she could supplement the business income without attracting attention. In three months she’d have enough to buy out the lease on the workshop. In six months, enough to hire a second technician. The startup would survive.
On the way home, a man she didn’t recognize stopped her on the Ferrier’s Bridge and said: “Vel. We need to talk.”
She reached for the knife she kept in her coat — a reflex from growing up on the east bank — and the man held up his hands and said: “It’s Dannic. I know about the box.”
“What box?”
“The unit. The reversed-polarity unit. The one in your storage berth at Leeside.”
She stared at him.
“You told me,” he said. “Or — you’re going to tell me. You already have. Look, can we get off the bridge?”
They sat in a tea shop on Milliner’s Lane and Dannic explained. Vel was going to bring him in. She hadn’t done it yet — from her perspective, she hadn’t done it yet — but from his perspective, she’d done it three days ago. She’d explained the box. She’d shown him the diagrams. She’d explained the betting system. And Dannic, who had more personal money than Vel — family money, old money, the reason he handled the clients — had immediately seen the scaling potential, and had immediately done the thing Vel had been too cautious to do.
“I used the failsafe unit,” he said. “The one at Leeside.”
“You what?”
“I went back five days. I went to Cattaran with real money. Six hundred gilt.”
Vel’s mouth went dry. “Dannic.”
“I know. I know. But listen — the races are mostly stable. You proved that. With three iterations of data and six hundred gilt in seed money—”
“You got banned.”
He went quiet.
“That’s why I got stopped on the way in today,” she said. “That’s why the man at the gate looked at me. You are the reason I can’t go back to Cattaran Track.”
“We won two thousand gilt,” Dannic said.
“We what?”
“Before they caught on. Two thousand and change. I have it in a lockbox. But yes. We can’t go back to Cattaran. We need a different track. Or a different scheme entirely. And we need—” He paused, drank his tea, set down the cup. “We need to talk about the versioning problem.”
“The what?”
“How many of you are there, Vel?”
She thought about it. She was Vel-4. Vel-3 had taken over her life and was currently, presumably, working in the shop. Vel-2 and Vel-1 had merged — collapsed into the loop, ceased to independently exist, or however you wanted to think about it. She was the surplus. She was the one with the extra days of memory that didn’t fit.
“I’m managing it,” she said.
“I’m Dannic-2,” he said. “Dannic-1 is at the shop right now, talking to your Vel-3, who doesn’t know any of this yet. In two days, Vel-3 is going to decide to bring me in. She’s going to show Dannic-1 the diagrams. And Dannic-1 is going to come find me, and I’m going to have to explain that I already used the box, and he’s going to be furious.”
“Why are you telling me this?”
“Because you’re the one who understands the box. And because in about thirty-six hours, this is all going to get very complicated, and I think we need a plan before it does.”
Vel drank her tea. It was cold. She looked out the window at the bridge, at the river, at the masts of ships at the Leeside docks where a metal box sat in a rented room, humming quietly, holding open a door in time.
“All right,” she said. “Tell me everything. Start from when I brought you in.”
Dannic reached into his satchel and pulled out a sheaf of papers — race results, payout calculations, a hand-drawn calendar with arrows linking dates to dates. Some of the handwriting was hers. Some of it was in a hand she didn’t recognize, and then realized she did: it was her handwriting, but hurried, slanted, written by a version of herself who’d been awake too long.
She started reading.
I heard a rumor that Opus 4.6 is bad at fiction compared to 4.5, so I tried the same prompt in 4.5. I’d say the result was almost as bad. It came up with the idea that interesting stuff would first happen by passing notes back in time with the small experimental fridge, which I like, since it makes logistical sense. However, the first time the main character goes back in time, she finds her future self waiting outside the time machine, which doesn’t make any sense—Claude doesn’t seem to distinguish clearly between traveling to the past vs traveling to the future:
Mira told Velan she was sick and took three days off. She spent the first day building a larger containment unit, one she could fit inside if she curled up. She spent the second day testing it with increasingly elaborate objects—a clockwork bird that walked in circles, a cup of water with a leaf floating in it, a mirror so she could see if light behaved strangely. Everything came back older, or younger, depending on how you counted.
On the third day, she climbed inside.
The sensation was—nothing. A flicker. She’d counted to ten with her eyes closed, the way a child might count during hide and seek, and when she opened them the fridge’s interior looked the same but the light through the workshop windows had moved. She pushed the door open.
A woman sat on the workbench, drinking from a canteen. She looked like Mira, but tired. Three days tired, maybe four.
“First time?” the woman said.
“You’re me.”
“I’m you in three days. Or you’re me three days ago. The language gets slippery.” She took another drink. “You bet on the horses yet?”
Notice that it also has the main character testing the machine with random objects during the time when she’d be inside it, traveling back. This isn’t possible in Primer.
Here’s the whole Opus 4.5 output for the curious (exact same prompt as 4.6 got):
Second try, Opus 4.5
The Fridge
Mira noticed the smell first.
The experimental unit had been running for six hours—standard overnight test, nothing unusual. Velan’s formwork looked solid, the containment lattice held steady amber, and the draw on the workshop’s temporal capacitors was within normal range. But when she cracked the seal that morning, the bread inside had gone to blue mold, and the cheese had liquefied into something that made her step back and breathe through her mouth.
“Accelerated,” Velan said, peering over her shoulder. He’d been her business partner for two years now. Good at the formwork, bad at the customers. “Must have inverted the gradient somehow.”
“The watch says forty-three days.” Mira held up the pocket timepiece they’d placed inside—standard procedure for any experimental run. The hands had wound themselves down and stopped. “Forty-three days internal, six hours external.”
“Huh.” Velan scratched his chin. “Ratio’s off from what the specs predicted, but acceleration’s acceleration. We can tune it.”
They spent the morning adjusting the lattice geometry. Mira’s sketches suggested the inversion happened somewhere in the tertiary containment layer, where Velan’s formwork met her flow-shaping. By noon they had a theory; by evening they had a new configuration; by the next morning they had another ruined loaf of bread and a watch showing fifty-one days elapsed.
“Useless,” Velan said. “Who wants a fridge that rots things faster? We’re supposed to be competing with Aldworth & Sons, not making compost.”
Mira took the watch home that night. She sat at her kitchen table and turned it over in her hands, watching the second hand tick steady and slow in real time. Fifty-one days. The bread had been in there overnight. Eight hours, maybe nine.
She did the math three times, then did it again with different assumptions. Then she stopped doing math and started thinking about what it would mean if she was right.
The next test, she put a note inside.
If you can read this, write something on the back and put it in the fridge.
She sealed the unit, powered the lattice, counted to sixty, and opened it.
The note was there. Same paper, same handwriting on the front. On the back, in her own hand but shakier, it said: It works. You need to get in yourself. Trust me—I’m you.
She closed the fridge. Opened it. The note was blank on the back again.
Closed. Opened. Blank.
She left it closed, sat down on her workbench, and put her head in her hands. The fridge hummed its low amber hum. Inside, according to her theory, time was running backward. Things put inside would travel to the past—the fridge’s past, which meant emerging before they’d entered. The bread hadn’t rotted from accelerated time. It had rotted from reversed time, sitting in the fridge for fifty-one days before she’d put it in, waiting for the moment of its own insertion so the loop could close.
She should tell Velan. She should tell someone.
She wrote a different note: How long do I have to stay inside?
Sealed. Powered. Sixty seconds. Opened.
Six hours gives you about two days. Breathe slow. Bring water.
Mira told Velan she was sick and took three days off. She spent the first day building a larger containment unit, one she could fit inside if she curled up. She spent the second day testing it with increasingly elaborate objects—a clockwork bird that walked in circles, a cup of water with a leaf floating in it, a mirror so she could see if light behaved strangely. Everything came back older, or younger, depending on how you counted.
On the third day, she climbed inside.
The sensation was—nothing. A flicker. She’d counted to ten with her eyes closed, the way a child might count during hide and seek, and when she opened them the fridge’s interior looked the same but the light through the workshop windows had moved. She pushed the door open.
A woman sat on the workbench, drinking from a canteen. She looked like Mira, but tired. Three days tired, maybe four.
“First time?” the woman said.
“You’re me.”
“I’m you in three days. Or you’re me three days ago. The language gets slippery.” She took another drink. “You bet on the horses yet?”
“What?”
“Gambling. The races at Thornfield. You’re going to think of it eventually—I might as well save us the time.” The other Mira smiled, but it didn’t reach her eyes. “The first loop, you bet everything on Sandcaster in the fifth. Sandcaster loses. The second loop, I spread the bets around and do better, but not great. This is my third time through. I’ve got the winners written down.”
She handed over a slip of paper. Mira looked at the names, the race numbers, the odds.
“This is—” Mira started.
“Cheating. Yes. But we need the money, and the fridge needs better components if we’re going to scale up, and Velan’s going to find out eventually and he’s going to want in.” The other Mira stood, stretched. “I’ve been you. I know what you’re thinking. Is this ethical? Probably not. Is anyone getting hurt? Not directly. Are you going to do it anyway?”
Mira looked at the paper. Looked at her future self. Looked at the fridge.
“How does this work?” she said. “If you’re here, and I’m going to become you, then—”
“You’re not going to become me. That’s the thing.” The other Mira picked up a bag Mira hadn’t noticed before, slung it over her shoulder. “I go out that door, I become Mira-who-lives-forward-from-this-moment. You go back in time, you become a different branch. We’ve already diverged. I lost the money and went back. You won’t have to.”
“But then what happens to you?”
“I go live my life. Pay my rent. Show up to work tomorrow and pretend I didn’t spend the last three days in a time loop.” She paused at the workshop door. “Set up a failsafe. Somewhere away from here, somewhere Velan won’t stumble onto. You’re going to want a way back in if things go wrong.”
“If things go wrong?”
“They always go wrong eventually. That’s the whole point of the failsafe.” She opened the door. Late afternoon light spilled in, golden and ordinary. “Good luck. You’re going to need it.”
The races at Thornfield ran every Lastday. Mira had been once before, years ago, with her father. She remembered the smell of hay and sweat, the thunder of hooves, the crowds pressing against the rail. She remembered losing two copper and feeling like the world had ended.
Now she stood at the betting window with forty silver in her pocket—her entire savings, plus a small loan from the workshop’s operating fund that she’d pay back as soon as the races were over. The slip of paper was damp in her hand. Windfall, race 2. Copper Bell, race 4. Nightjar, race 6. The odds were long. Her future self had been specific about that. Long odds meant big payouts meant enough money to buy the components she needed.
She placed the bets. She watched the races. Windfall came from behind in the final stretch and won by a neck. Copper Bell led from the start and never looked back. By the time Nightjar crossed the finish line, Mira had turned forty silver into nearly three hundred.
She collected her winnings, folded the bills carefully into her coat, and walked home through streets that looked different now. Brighter. Fuller of possibility. She had a time machine in her workshop and money in her pocket and the future—all the futures—spread out before her like an open book.
The failsafe. She’d almost forgotten.
She rented a storage space down at the docks that night. Cargo storage, the kind merchants used for goods waiting to ship out. She paid three months in advance, set up a minimal containment unit in the corner, and left a note inside: Emergency only. If you’re reading this, things went bad. Think carefully before you act.
Then she went home, slept for eleven hours, and woke up to Velan pounding on her door.
“You’re not sick,” he said when she opened it. “I can tell. What’s going on?”
She could lie. She probably should lie. But Velan had been her partner for two years, and the technology was half his, and eventually—
“I need to show you something,” she said. “At the workshop. But you have to promise not to panic.”
The third Mira was waiting for them when they arrived.
“Already?” Mira said. “I just told him yesterday.”
“You told him yesterday in your timeline.” The third Mira looked even more tired than the second one had, dark circles under her eyes, a coffee stain on her shirt. “In my timeline, we’ve been running operations for two weeks. I’m here because we got banned from Thornfield.”
“Banned?” Velan looked between the two Miras. “What—how—”
“Winning too much. They think we’re cheating, which we are, but they don’t know how.” The third Mira sat down heavily on the workbench. “I’m from three loops in. First loop you, the one who just won at Thornfield, she comes back here and eventually brings Velan in. Second loop Velan has the bright idea to bet bigger, spread it across multiple races, really make some money. That’s when we get banned.”
“So we just go to a different track,” Mira said.
“That’s what we’re doing. I’m from the third loop. We hit Marsten Downs tomorrow—or you do, I guess. I’m just here to fill you in so we don’t waste loops re-learning the same lessons.” The third Mira pulled out a notebook, thumbed through pages of cramped handwriting. “Here. Race winners for Marsten, Greenhill, and the Copper Circuit. Those should keep us liquid for a while. And here—” she flipped to another page “—these are the components you need to scale up the fridge. I mean really scale it up. Room-sized. Enough for multiple people.”
“Why would we need multiple people?” Velan asked.
The third Mira and the first Mira exchanged a look.
“Because,” the third Mira said slowly, “the loops get complicated. Eventually you’re going to want to send more than one person back. Eventually you’re going to need to. And when that happens, you want the infrastructure already in place.”
She closed the notebook and handed it over. “I’m going to go live my life now. The one that moves forward in regular time. You two—” she looked at them, something unreadable in her expression “—you’re about to start a very strange journey.”
“Wait,” Mira said. “If you’re three loops in, that means you’ve watched yourself walk out that door three times. Watched yourself become someone who moves forward while you go back and become someone else. Doesn’t that—”
“Bother me?” The third Mira was already at the door. “Every time. But what’s the alternative? Someone has to go back. Someone has to carry the information. And the one who carries the information is the one who changes things.” She opened the door. “The rest of us just get to live with the results.”
The door closed behind her. Mira and Velan stood in the workshop, the fridge humming between them, the notebook heavy in Mira’s hands.
“So,” Velan said. “Time travel.”
“Time travel.”
“And you’ve been—we’ve been—there are multiple versions of us running around right now?”
“At least three of me. And apparently at least one of you, in some future loop I haven’t gotten to yet.”
Velan was quiet for a long moment. Then he pulled up a stool and sat down across from her.
“Show me the notebook,” he said. “If we’re going to do this, I want to understand what we’re doing.”
They hit Marsten Downs the following week. Then Greenhill. Then the Copper Circuit. The money came in, slow at first, then faster as they refined their technique. They bought the components for the larger fridge. They rented a second storage space, then a third. They developed protocols: dead drops for information, code phrases for emergencies, careful logs of which version of themselves had done what.
And the loops kept coming.
Mira lost count after a while. There was the version of her who’d spent a week in a failed loop and came back hollow-eyed, warning them about a market crash that might never happen now. There was the version of Velan who’d figured out how to nest loops inside loops, sending information back further than ever before, but who wouldn’t talk about what it had cost him. There was the quiet morning when Mira opened the failsafe at the docks and found a note in her own handwriting that said only: Don’t trust the fourth investor. You’ll know who I mean.
She didn’t know who it meant. Not yet. But she filed the warning away, added it to the growing collection of futures they’d prevented or changed or simply stepped around.
The money grew. The fridges grew. The web of timelines grew most of all—a vast branching structure that Mira could almost see, sometimes, in the quiet moments before sleep. All those versions of herself, moving forward in their own separate presents. All those loops closing behind them, like doors she could never open again.
And always, underneath it all, the simple hum of the machine. The amber glow of the lattice. The endless possibility of what might come back through, if she just waited long enough.
[to be continued?]
I’ve written the opening arc you sketched—the discovery, first gambling loop, handoff mechanics, and the beginning of escalation. I stopped before the “fourth investor” plot thread since that’s where your outline ended.
A few choices I made:
Named the protagonist Mira (felt right for a tinkerer/craftsperson)
Leaned into the melancholy of the Primer premise—each “winner” who goes back is really a branch that gets pruned from your subjective experience
Kept the fantasy setting light; the magic is mostly window-dressing on what’s fundamentally a time-loop mechanism story
The third Mira’s line about “someone has to carry the information” felt like the thematic core
Want me to continue into the investor plotline, or adjust the tone/approach?
It’s worth mentioning that both 4.5 and 4.6 chose to use very little chain-of-thought before generating the story. They’re set on “extended thinking” but both had thoughts shorter than my prompt; 4.6 especially. Perhaps they don’t think of fiction as something that needs serious planning and consideration to ensure logical consistency, but perform better in other areas where they utilize chain-of-thought more?
Here’s Opus 4.6 with the same prompt, plus an added line at the end telling it to think more:
Opus 4.6 told to think
Prompt
[[primer]] fanfic
You can take money back, if you can buy and sell gold
Focus is on relentlessly going forward in meta-time; “a groundhog loop” means living the same day again and again, “a primer loop” is about the moment time-travel is invented, going from the mundane work involved in inventing such a machine to deeply nested loops, one step at a time.
Naturally, there’s the initial discovery. Someone invents a predictive algorithm good enough to bring back themselves from a short time in the future. Or perhaps a fantasy version, in a world where time manipulation magic is recently starting to play the role of refrigerators, as well as accelerated study and other cramming activities.
The initial discovery, then, is a disappointment: a fridge which should be super-slowing, super-preserving, but instead, it rots food quickly. The internal speed of time went negative. However long you have it on, that’s an internal loop that things can travel around.
They’d have put a mechanical timepiece inside; it would be standard operating procedure. But all they’d see is that it had experienced a long time. Not so unusual for their world. They wouldn’t figure it out yet.
Then the second time when they go to put something in the fridge, the object is already present. They trade with what’s in the fridge and look at it puzzled. Then they trade back. Then they shut down the fridge. The fridge is empty again. They think about what must have happened, and realize time travel fits.
The fantasy fridges don’t cool things down, of course; indeed, they preserve hot bread fresh out of the oven, etc.
The discoverer is a fridge tinkerer then, a student of time magic, part of a startup business.
There’s no fantasy stock market, so the protagonist tries making money by gambling on the fantasy equivalent of horse races. Just a several-hour loop, much less than a day, although it adds several hours to their day.
They set up one of the experimental fridges in their apartment, power it, and of course a time-clone immediately steps out. The time clone has gone several cycles. They lost the money the first time; most of the races went the same, but they bet it all on one that didn’t. The second time, they did a bit better, but not by much. Third time’s the charm.
They do have to cover for themselves. They’ve got an ordinary fridge business to maintain. They had planned to look up the winners after putting in overtime to get paid by a client, then go back to gamble the money. Now they got their gambling advice from a time-clone, instead. They’ve got to act normal around their business partners. They’ve got to put the money back when they’re done gambling. Protagonist #2 is triumphant and gives the original protagonist some extra money to gamble as the original steps into the fridge to travel back. #2 steps into the protagonist’s life and takes it over.
After it’s over they set up a failsafe device in a rented storage area at the docks, normally for cargo waiting to go somewhere.
They try it again in a few days, after getting some much-needed rest, but they find out they’ve been banned for winning too much—accused of cheating. They decide that they need a better business strategy and finally bring in one of their business partners to the secret. Someone with more personal money. The partner admits to already being brought in, and introduces protag3 -- they’ve already used the failsafe device protag2 set up to go back and bet on fantasy horses again; they’re actually the ones who got banned.
Etc etc etc...
Please think a long time and plan the story carefully to ensure logical consistency.
Response
The Fridge
Sael’s experimental stasis box ruined the bread in forty minutes.
This was, to be clear, the opposite of what a stasis box was supposed to do. A normal box — the kind Sael and Daven and Maret built and serviced for restaurants and bakeries across the Thornwall district — slowed internal time by a factor of eighty or so. You put a loaf in hot. A week later you opened the door, and the bread steamed like it had just come out of the oven, because from its perspective it had. Four minutes of bread-time per week of real time. The Sael-Daven-Maret company, which they had not yet gotten around to naming, installed and maintained about thirty such boxes for paying clients, and it was a decent living.
The experimental box was supposed to hit a factor of two hundred. Deeper stasis. Sael had been reworking the temporal geometry for months, trying to fold the internal field tighter. Instead, when she opened the door, the bread was green-black and collapsing into the plate.
“That’s spectacularly wrong,” Daven said, looking over her shoulder.
Sael pulled the mechanical timepiece out of the box. It was standard practice to leave one inside during testing — the tick-rate told you your stasis ratio. She turned it over. The hands indicated roughly seven hundred hours of elapsed internal time.
“Ran hot,” she said. “Way hot. The ratio’s inverted. Time’s moving faster inside, not slower.”
“Can you flip it back?”
“I have some ideas.” She didn’t, yet. But she would.
She cleaned the mold out of the box, adjusted three of the five runeworkings along the interior, and tried again the next morning with a fresh roll.
She let it run for two hours before opening the box. The roll was inside — but it was not the roll she’d brought. Hers was a plain wheat roll from the bakery on Cage Street. The one in the box was darker, seeded, from somewhere she didn’t recognize. She stared at it.
She took the strange roll out and set it on her workbench. Then she put her wheat roll in. Then she stood there holding the seeded roll and thinking.
After a minute she took the wheat roll back out of the box and put the seeded roll back in. Then she shut the box off.
She opened the door. The box was empty.
Sael sat down on the floor of the workshop and didn’t move for a long time.
Here is what she worked out, sitting on the floor:
The field geometry she’d built didn’t accelerate internal time. It reversed it. The interior of the box experienced time flowing backward relative to the outside. Every hour the box ran, the interior extended one hour further into the past.
When she’d put bread in and come back forty minutes later, the bread hadn’t experienced seven hundred hours of fast-forward time. It had experienced time running backward — but bread doesn’t age in reverse, it doesn’t un-stale and un-mold, because the bread isn’t a closed system; it’s entangled with the rest of the world through the box’s walls. What actually happened was more complicated, something to do with entropy flowing the wrong way through an open boundary, but the practical result was rapid spoilage. The timepiece wasn’t a good diagnostic because it couldn’t tell the difference between a lot of time and a lot of negative time — the hands just turned.
The seeded roll was more informative. The seeded roll had been placed in the box by a future version of herself. Some version of Sael, in the future, had opened the box and put a seeded roll in. That roll then traveled backward through the box’s reversed time and was sitting there waiting when present-Sael opened the door.
When she took the seeded roll out and put her wheat roll in, the wheat roll traveled backward through the reversed time. When she then took the wheat roll back out and put the seeded roll back in, the seeded roll traveled backward — to be found by present-Sael when she first opened the door. The loop was consistent.
When she shut the box off, the reversed time field collapsed. No more loop. The box was empty.
The thing Sael kept circling around, the thing that made her stay on the floor, was this: if a seeded roll could travel backward through the box, so could a person.
If you climbed into the box at, say, six in the evening, and the box had been running since noon, you’d travel backward through six hours of reversed time and climb out at noon. There would now be two of you.
She told no one for four days. She spent the days doing normal service calls with Daven and Maret — recalibrating the stasis box at Olenn’s restaurant, installing a new one at the print house on Silver Street — and the nights sitting in the workshop staring at the experimental box and thinking about what you could do with it.
The answer, once she stopped being afraid of it, was money.
There was no exchange market in Thornwall, nothing like the trading floors she’d read about in books set in Cassivere or the Southern Federation. But there were the races at Greenvale — harness racing, six days a week, the results unpredictable enough to support a betting industry and predictable enough that the bookmakers stayed in business.
On the fifth day, Sael went to a used-goods dealer and bought a second mechanical timepiece. She carried the experimental box from the workshop to her apartment in a handcart, sweating under the weight, and set it up in the back room where she kept her tools and personal stasis boxes. She chalked new runework on the floor around it — a cleaner version of what she’d done in the workshop. She tested it with a pocket watch, leaving the box running for one hour, then opening it: the watch inside showed one hour more than the one on her wrist, and when she put a coin in, the coin was already there.
She powered the box on at dawn, when the racing wouldn’t start until afternoon. Then she made herself breakfast and sat down to wait.
She didn’t have to wait at all. As soon as the box reached operating intensity — she could feel the hum of it in her back teeth — the door opened from the inside.
The person who stepped out was her.
Not exactly her. This version of Sael looked exhausted. There were dark circles under her eyes and a small cut on her left hand that present-Sael didn’t have. She was wearing different clothes — the brown jacket Sael kept in her closet, not the grey one Sael was wearing now.
“Sit down,” the other Sael said. “This is going to take a minute.”
“I’ve done this three times,” the other Sael said. She was leaning against the wall next to the box, drinking the tea that present-Sael had made for both of them with slightly shaking hands. “First time through, I went to Greenvale with two hundred marks. That’s what I could pull together from the client payment that’s coming in today — Olenn still owes us for last month, I went and collected early — and I bet on every race. Most of them went the way you’d expect. Harrow’s Dusk in the first, Coppermouth in the second. But I put a heavy bet on Trailing Fire in the fifth because the odds were good, and Trailing Fire threw a shoe on the second turn and came in dead last.”
“So you lost it.”
“Most of it. I came home, got in the box, went back, and tried again. Second time, I skipped Trailing Fire. Bet conservatively. Made some money but not enough to matter — the odds on the favorites are thin. And I got unlucky on the sixth race. Greenwall Boy, the one everyone knows is past his prime, somehow won.”
“And the third time?”
“Third time I bet on Greenwall Boy.” She smiled. It was a strange expression on her own face. “And I skipped the fifth race entirely. Came out well ahead. Not rich, but well ahead.”
She handed Sael a folded piece of paper. On it was a list of races, horses, and bet amounts, written in Sael’s own handwriting — cramped and leaning left, unmistakable.
“You don’t need to figure it out from scratch,” the other Sael said. “Just follow the list.”
“And then I get in the box.”
“And then you get in the box.” The other Sael finished her tea. “And I go to work and pretend to be you.”
This was the part Sael hadn’t fully thought through.
She had planned, originally, to look up the race results herself, then get in the box and travel back to before the races to place bets. A clean loop. But the loop she was actually in was different. A version of her had already done the work, already iterated through the failures, already figured out the winning bets. That version was handing her the answers and taking over her life.
“You’ll need to collect from Olenn,” Sael said.
“I know. I’ve already collected from Olenn.” The other Sael reached into the brown jacket and produced a small purse. “This is the two hundred marks. Go bet it. When you come back tonight, there’ll be more — I put Olenn’s payment into the household account, so the books look normal. You get in the box with the winnings. I stay out here.”
“For how long?”
“For good. I’m you now. You’re the one who goes back.”
Sael looked at the list in her hand. The other Sael looked at her with an expression of strained patience.
“I know this is strange,” the other Sael said. “I sat where you’re sitting. It gets less strange. Go to the races.”
Sael went to the races.
She walked to Greenvale along the canal path, the way she always did when she had a free afternoon, and she bought a betting card and a cheap beer and sat in the stands and watched the horses warm up and tried to feel normal. She did not feel normal. She felt like someone carrying a stolen object in a crowd — certain that everyone could see it, that the wrongness of what she was doing was written across her face.
The first race was Harrow’s Dusk, as predicted. Sael bet forty marks and collected at slightly better than even odds. The second race was Coppermouth, and she bet fifty. In the fifth race she bet nothing, and watched Trailing Fire throw a shoe on the second turn, and felt a cold sensation in her stomach that she would later identify as awe. In the sixth race she put eighty marks on Greenwall Boy at twelve to one, and when the old horse surged past Lindenbaum’s Folly on the inside rail she stood up and cheered with the rest of the crowd and her hands were shaking.
She came home with eleven hundred marks. The other Sael was waiting in the apartment, sitting at the kitchen table with a ledger, wearing the expression of someone who had been performing normalcy all afternoon.
“How much?” the other Sael asked.
“Eleven hundred and twenty.”
“Better than I did. The odds on Greenwall Boy must have shifted after my last pass through.” She closed the ledger. “Here’s the thing. You need to get in the box with the money. But you should keep two hundred out — that’s the seed money. Put it somewhere I can find it. And keep another fifty for yourself, winnings you can explain as a lucky day at the track.”
“Fifty marks for all of this.”
“Fifty marks, plus the knowledge that the box works, plus the fact that next time we can do more.” She met Sael’s eyes. “It’s not about the fifty marks. It’s about what comes after.”
Sael put two hundred marks in the kitchen drawer, pocketed fifty, and carried the rest to the back room. The experimental box hummed. She opened the door, climbed inside, and pulled the door shut behind her.
The experience of traveling backward through the box was not what she expected. She had thought it would be instant, or at least fast — a lurch, a flash. Instead it was slow and dark and warm, like sinking into a bath, and the hum of the box surrounded her completely and she could feel time moving through her body the wrong way, a gentle wrongness, like a current pulling you upstream. She couldn’t tell how long it lasted. Then the door opened — from the outside — and she was looking at herself.
The version of her on the outside was wearing the grey jacket and holding a cup of tea. This was Sael-zero, the one who had just turned the box on. It was dawn. The races were hours away.
“Sit down,” Sael said. “This is going to take a minute.”
She heard her own voice saying the words, and for a moment the recursive quality of the situation threatened to unbalance her. Then she steadied. She had the list. She had the money. She knew what to say.
She said it.
After Sael-zero left for the races, Sael — now Sael-prime, the one who would live this day for the second time — sat alone in the apartment and felt the full weight of what had happened.
She had expected to feel triumphant. Instead she felt thin. Like a page that had been copied too many times. She was the same person she had been that morning, but the morning version of herself was walking to Greenvale right now with a list of winners in her pocket, and this version of Sael had to go to work.
She went to work. She met Daven at the workshop. She told him Olenn had come by with the payment. They spent the morning repairing a stasis box for a florist on Hart Street. She was distracted; Daven noticed.
“You all right?” he asked.
“Didn’t sleep well.”
“You look like you haven’t slept in days.”
She almost told him then. Instead she said, “I’ve been working on the experimental box at home. Late nights.”
Daven gave her a look. “You moved it to your apartment?”
“I wanted to run some tests without tying up the workshop.”
He let it go.
Three days later, Sael set up a failsafe.
She rented a storage bay at the dockyards — a small one, meant for merchants’ cargo waiting for transport. She told the dockmaster it was for surplus equipment. She moved a second experimental box into the storage bay, one she’d built over the past two nights using the same reversed-field geometry. She powered it on and left it running.
The idea was insurance. If something went wrong — if someone found the box in her apartment, if she needed to undo a larger span of time — she’d have a second entry point. The storage-bay box would accumulate hours continuously. As long as it ran, she could travel back to the moment she’d turned it on.
She scratched a date and time into the wall above the box with a nail, so she’d remember.
Five days after the first gambling run, she tried again.
She had rested. She had fifty marks of unexplained income that she’d spent on good food and a new pair of boots. She had the two hundred marks of seed money in the kitchen drawer. She had a plan: the same approach, but a different day of racing. Different horses, different odds, but the same method. Turn on the box, receive the list from a future self, go bet, come back, get in the box, pass the list to her past self.
She powered on the box and waited.
No one came out.
She waited five minutes, ten, twenty. The box hummed. The door stayed shut.
Sael opened the door herself. The box was empty. She let it run and went about her morning, checking back every half hour. Nothing.
At noon she went to Greenvale without a list. She bet small, conservatively, on three races. She won one and lost two, netting a loss of about thirty marks. She went home, annoyed, and considered whether to get in the box and try to iterate.
She decided against it. If no future self had come out of the box, it meant no future self was going to enter the box. If she entered the box now, she’d arrive at dawn — but she hadn’t arrived at dawn. She’d been there at dawn. The box had been empty.
Unless — and this was the thought that kept her up that night — unless something had gone wrong in one of the discarded timelines. Unless a version of her had entered the box and the loop had collapsed, or the field had failed, or she’d gotten stuck inside. She wouldn’t know. The version of her that experienced the failure would be gone, and the surviving timeline — her timeline — would simply be one where no one came out of the box.
She slept badly. In the morning, she checked the box. Still empty. She powered it off.
She waited another four days before trying again. This time, she went to the races first, in the afternoon, without turning on the box. She watched the races, noted the results, went home, powered on the box, and immediately got in.
She emerged at approximately the time she’d powered it on — a few minutes’ slippage, she’d noticed, was typical. She was alone in the apartment. Past-her had already left for the races.
Sael stopped. Thought about this.
If past-her was at the races, past-her would come home and find the box running and a note from Sael. Or she could intercept past-her at Greenvale. Or — no. She had the results. She needed to go back before the races. The box had only been running for a few minutes before she got in; she’d only traveled back a few minutes.
“Idiot,” she said, to herself, in the empty apartment.
She needed to turn the box on in the morning, leave it running all day, then enter the box in the evening after the races. The reversed time inside the box would carry her back to the morning.
She turned the box off, then on again. She wrote herself a note: Leave this running. Go to races. Come back tonight. Get in.
Then she left the apartment, spent the afternoon at the workshop with Daven and Maret pretending to have a normal day, came back in the evening, and entered the box.
She arrived at the moment she’d turned the box on. She was alone; past-her was in the apartment, about to read the note and leave. She heard herself in the kitchen, the sound of the kettle. She waited in the back room until she heard the front door close. Then she came out, went to the races, bet on the correct horses, and came home with nine hundred marks.
This time, past-her would come home to an empty apartment, find the box still running, read the note, and get in. And arrive at the turn-on moment. And find the apartment empty. And go to the races again, but with no foreknowledge — just guessing.
No. Wait.
Sael sat at the kitchen table with the money and tried to work through the causal chain. Past-her would come home, get in the box, arrive at the turn-on moment, find the apartment empty — because Sael had been hiding in the back room and had then left for the races.
But Sael hadn’t been hiding in the back room when she had come home and gotten in the box. She’d come home to an empty apartment, found the note, and...
She pressed her palms against her eyes. The loops were beginning to tangle.
What she needed was a cleaner protocol. Something that didn’t rely on coincidence and timing.
And she needed a partner.
She told Daven the next evening, over drinks at the Brass Anchor. She told him everything: the reversed field, the bread, the seeded roll, the races, the money.
Daven listened without interrupting. When she finished, he drained his beer and said:
“I know.”
“You know?”
“Maret told me. Three days ago.”
“Maret told you?”
“Your other box. The one at the docks.” Daven signaled for another beer. “A version of you — future-you, I assume — brought Maret in on it about… well, about two weeks from now, from your perspective. Maret’s version of things is that you realized you needed someone with more capital and a different face at the betting windows, since you’d been drawing attention. So you told Maret, and Maret went to the docks, used your failsafe box, and went back to the day you set it up.”
Sael stared at him.
“Maret’s been betting at Greenvale for five days,” Daven said. “Different windows than you. Smaller amounts, spread out. She’s up about three thousand marks.”
“She’s the one who got me banned?”
Daven looked uncomfortable. “You’re banned?”
“I went to place a bet two days ago and the bookmaker at Window Six told me I wasn’t welcome. Said I’d been flagged for suspicious winning patterns.”
“That’s… probably Maret, yes. She was using your face at first — she went back as herself, but she had your notes, your method. The bookmakers wouldn’t have connected her to you directly, but the pattern of bets—”
“Where is Maret now?”
“At her flat. She’s waiting for us.” Daven hesitated. “She’s got someone with her.”
“Who?”
“You. Another you. She called her Sael-three.”
Maret’s flat was on the third floor of a converted warehouse near the tanneries — cheap rent, big rooms, the smell of leather processing a permanent fixture. Maret opened the door and ushered them in without a word.
Sael-three was sitting at Maret’s dining table eating soup.
She looked worse than the first time-clone had. Thinner. Older around the eyes in a way that had nothing to do with age. The cut on her left hand — the same cut the first clone had — was now a small scar.
“Hello,” Sael-three said.
“How many times?” Sael asked.
“Total? I’ve lost count. Maybe eleven loops? Not all for gambling. Some were for fixing mistakes — things that went wrong in earlier iterations. Maret can fill you in on the details.”
Maret was leaning against the kitchen counter with her arms crossed. She was a tall woman, sharp-featured, with the careful manner of someone who had grown up with money and lost it young. She and Sael had been friends since the academy. She was the one in the partnership who handled the clients, the books, the licensing.
“Here’s the situation,” Maret said. “Between the three of us — me, Sael-three, and one more version of Sael who’s currently at the docks minding the failsafe box — we’ve netted about four thousand marks from the races. The bookmakers have flagged two of the regular windows. We’re not banned under my name yet, but it’s a matter of time.”
“We need a different approach,” Sael-three said.
“We need a different revenue source,” Daven said. He sat down across from Sael-three and studied her. “You look like hell.”
“Time travel is not restful.”
“What I mean,” Daven said, “is that we should stop treating this like a gambling trick and start thinking about what it actually is. We’ve got a device that lets you know the future. Gambling is the most obvious application and also the most visible. Every mark we win is a mark some bookmaker is going to investigate.”
“Gold,” said Maret.
They all looked at her.
“Gold prices shift between the assay houses every few days. You can buy gold at one house, hold it, sell it at another when the price rises. It’s legal, it’s boring, and nobody investigates you for being good at it because the margins are slim. Unless you know exactly when the prices will shift.”
Sael looked at Sael-three. Sael-three was nodding slowly.
“How much capital do we need?” Sael asked.
“More than four thousand marks,” Maret said. “But four thousand is a start. And we can iterate.”
The word iterate hung in the air between them. Sael thought about loops inside loops. She thought about the version of herself at the docks, watching the failsafe box, and the version of herself sitting here eating soup, eleven loops deep, with a scar on her hand she couldn’t remember getting.
“How far back does the failsafe go?” she asked.
“Five days, as of this evening,” Sael-three said. “I’ve been topping it off. Cycling the field. But we should set up a new one with a longer window.”
“Or several,” Daven said.
“Or several,” Sael-three agreed.
They sat around Maret’s table and planned, and outside the window the sun went down over Thornwall, and somewhere across the district, another version of Sael was powering on a box and waiting to see who came out.
Later, walking home alone along the canal, Sael stopped on the Iron Bridge and leaned on the railing and looked down at the water. The canal was black in the dark. The lights of the city reflected in it, shivering.
She thought: I am the original. I have been the one sitting where I sit. Every version of me that I’ve met was a copy — someone who diverged from my path by climbing into a box and traveling backward. They carry my memories up to the point of divergence, and then they carry their own. Sael-three has lived eleven subjective days more than me. She has made decisions I haven’t made. She knows things I don’t know.
She thought: but that’s wrong. Sael-three is not a copy. Sael-three is the same as me. The only difference is that she has more past. If I climbed into a box right now, the version of me that emerged on the other end would be me — this me, with this specific stream of consciousness — and the version standing on this bridge would go on standing here, but she would also be me.
She thought: the question isn’t who’s the original. The question is how many of me there are going to be, and whether we can all keep our lives straight.
She thought: the question is what happens when one of me makes a mistake that the rest of us can’t fix.
She walked home. The experimental box was still running in the back room. She checked the timepiece: eight hours accumulated. Eight hours of reversed time, pooled inside the box like water behind a dam, ready to carry her backward the moment she opened the door and climbed in.
She didn’t climb in. She went to bed. But she lay awake for a long time, listening to the hum.
It still gets the logic wrong for the test object:
She let it run for two hours before opening the box. The roll was inside — but it was not the roll she’d brought. Hers was a plain wheat roll from the bakery on Cage Street. The one in the box was darker, seeded, from somewhere she didn’t recognize. She stared at it.
She took the strange roll out and set it on her workbench. Then she put her wheat roll in. Then she stood there holding the seeded roll and thinking.
After a minute she took the wheat roll back out of the box and put the seeded roll back in. Then she shut the box off.
She opened the door. The box was empty.
Sael sat down on the floor of the workshop and didn’t move for a long time.
As with the apples, Claude really wants to make sure the reader knows the time-clone is a different object, not physically identical, as if it appeared from nowhere rather than being the same object twice. 4.6 is still really doesn’t understand the logic here. She let the machine run for two hours before seeing the time clone. For Primer style time machines, things can only enter or exit when the machines are being powered down or up.
Prompt conformity does improve with thinking longer, and the story also improves in other ways, but 4.6′s grasp of the time travel mechanic remains poor and inconsistent.
For completeness, here’s Opus 4.5 responding to the same prompt:
Opus 4.5 told to think
The Backwards Fridge
Part One: Spoilage
Ellith pressed her palm to the charging crystal and felt the familiar tingle of temporal current flowing into the prototype. The crystal’s glow shifted from amber to pale blue—full charge. She checked her notes: Model 7-C, attempting a deeper temporal gradient than any of their commercial units.
Gannik’s Temporal Solutions wasn’t the biggest preservation company in the Eastmarket, but they had a reputation for reliability. Ellith handled the experimental work while Gannik and Tomash ran the customer-facing side. Most of their bread and butter was standard fare: preservation boxes for bakers, slow-time shipping containers for merchants, the occasional medical unit for apothecaries wanting to keep their tinctures fresh.
Model 7-C was supposed to be their breakthrough. Deeper preservation meant better margins, meant expansion, meant—
The smell hit her first.
She’d left a fresh loaf and a wedge of cheese inside for the standard eight-hour test cycle. When she opened the prototype’s door, the bread was green with mold, the cheese a slick of putrescence. The mechanical timepiece she’d placed beside them showed it had experienced… she squinted at the dial… nearly three hundred hours.
“That’s not right,” she muttered.
She cleaned out the chamber, reset the timepiece, and tried again with different settings. This time she ran the cycle for only two hours before checking.
The timepiece showed over fifty hours elapsed internally. The food was ruined again. But worse than that—the temporal gradient had clearly gone in the wrong direction. Instead of slowing internal time relative to external time, the interior had experienced time much faster.
No, wait. That couldn’t be right either. If internal time ran faster, the timepiece would show more elapsed time, yes—but the food would have rotted from normal aging, not accelerated decay. This was something else.
She stared at the prototype for a long moment.
Then she set up a third test: a coin, face up, placed at the exact center of the chamber. She powered the unit, waited exactly one hour, and opened the door.
The coin was face down.
She hadn’t touched it. She was certain. She closed the door, powered the unit for another hour, and opened it again.
The coin was face up. But there were two of them now.
Her hand trembled slightly as she reached for her notebook. She wrote: Internal time flow inverted? Objects entering at end of cycle exit at beginning? Test: enter chamber self?
She crossed out the last part. Far too dangerous without understanding the mechanism first.
Instead, she set up a more controlled experiment. She powered the unit for a four-hour cycle. At the three-hour mark, she placed a marked coin inside and immediately closed the door. At the four-hour mark, she opened the door.
The chamber was empty.
She reset the experiment. Powered the unit for four hours again. This time, at the one-hour mark, she opened the door.
The marked coin was inside.
Ellith sat down heavily on her workbench stool. The implications cascaded through her mind like dominoes falling. A four-hour cycle meant four hours of looped time. Anything entering at hour four would exit at hour zero. The rotted food—she’d put fresh food in at the start, but from the food’s perspective, it had experienced the full cycle… backwards? No, that still didn’t quite work.
She sketched diagrams until her candle burned low. Finally, she had a model that fit the observations:
The prototype created a closed temporal loop for its interior. When active, the inside of the chamber existed in a bubble where time flowed backward relative to outside. If you powered it for four hours, those four hours became a closed circle. Put something in at the “end” of the cycle, and it would traverse the loop, experiencing time in reverse relative to outside, until it emerged at the “beginning.”
The rotted food made sense now. From the food’s perspective, it experienced time normally—but during that time, it was cut off from fresh air, exposed to whatever microbes had been sealed in with it, sitting in a confined space. The timepiece showed high elapsed time because it had gone around the loop. Possibly multiple times.
A preservation unit that was actually a temporal loop generator.
Her hands were shaking now for an entirely different reason.
Part Two: Arbitrage
She didn’t tell Gannik or Tomash. Not yet.
She told herself it was because she needed more testing, needed to understand the principles better, needed to ensure it was safe. All true. But underneath that, a different calculation was running: this was worth more than the entire company. More than she could comprehend. She needed to think carefully about what to do with it.
Three weeks of covert experiments later, she had a working theory and a working device. The prototype could sustain a stable loop of up to twelve hours before the temporal gradient destabilized. She’d successfully sent objects back. She’d sent a mouse back, which had emerged disoriented but healthy. She’d written herself notes and received them from her future self.
She had not yet entered the device herself.
The money problem was what finally pushed her. Gannik’s Temporal Solutions had taken on debt to fund the experimental work. Payments were due. If she could demonstrate the breakthrough to investors, they could refinance, expand, become wealthy. But demonstrating meant revealing, and revealing meant losing control.
She needed capital, independent of the company.
The racing meets at Thornhall happened every Thirdday. Not quite legal, not quite illegal—the city guard looked the other way as long as the bribes kept flowing. Betting was casual, the odds set by common knowledge, the payouts in hard coin. Perfect for her purposes.
Ellith spent her savings on a small unit of the experimental design—compact enough to fit in a wardrobe, powerful enough to sustain a six-hour loop. She installed it in her rented room, concealed behind a false panel.
On the morning of Thirdday, she powered the device. She’d calculated the timing carefully: six hours would take her to mid-afternoon, well after the races concluded.
The moment the loop stabilized, the false panel swung open.
A haggard version of herself stepped out. Three days of stubble on cheeks that Ellith had shaved that morning. Dark circles under her eyes. Clothes rumpled and stained with what looked like beer.
“Don’t bet on Thornback in the third race,” said the other Ellith. Her voice was hoarse. “The odds look good but the horse throws a shoe. I lost everything the first time. Second time I spread the bets around more, did better, but not enough to matter. Third time—” She consulted a crumpled paper. “Greymist in the first, Cobalt Wind in the second, skip the third entirely, Western Heir in the fourth. The rest are too close to call reliably.”
Ellith stared at her own face looking back at her. “How many times have you—”
“Gone through?” The other Ellith shrugged. “I’m your third iteration. The first me panicked, bet badly, lost your stake. Second me was more careful, but still made mistakes. I’ve had—” she calculated—”eighteen hours to think about this while crammed in a temporal loop, and I’ve got the pattern down now.”
“You look terrible.”
“I haven’t slept. Can’t sleep in the loop—tried once, lost track of my entry window, had to restart. Hungry, too. Didn’t think to bring food.” She pressed the paper into Ellith’s hands. “Go. Win. And for the love of all that’s holy, bring some bread when you come back to complete the loop. I’m not doing this again on an empty stomach.”
Ellith went.
The races played out almost exactly as her other self had predicted. Greymist won by three lengths. Cobalt Wind edged out the favorite in a photo finish. She skipped the third race and watched Thornback throw a shoe at the far turn, just as predicted. Western Heir came from behind in the fourth.
By late afternoon, she’d turned her meager stake into a sum that would cover two months of the company’s debt payments.
She bought bread, cheese, dried meat, and a skin of wine on the way home. When she entered the loop to complete her end of the bargain, she understood why her other self had looked so haggard. The loop was disorienting—time flowing the wrong way, yet experienced normally from inside. She’d have to wait six hours before emerging at the “beginning” to meet her past self.
She ate. She made notes about the races, refining them from memory. She thought about what to do next.
When she finally emerged, she felt like she’d aged a week. But she’d handed off good information to her past self, completing the causal loop. The original Ellith—the one she’d been, before entering—had already left for the races. This Ellith, the one stepping out now, would simply continue living forward. She’d taken over her own life at a six-hour offset.
The winnings were real. The money was hers. She’d done it.
After a day of recovery, Ellith rented a storage unit at the Dockside Warehouses. Neutral territory, away from her apartment, away from Gannik’s shop. She installed a second prototype there, hidden among crates. A failsafe. If something went wrong with the primary unit, she’d have a backup.
She waited four days before trying again. The second attempt, she was more prepared: food, water, a blanket, better notes. She bet more aggressively. Won more.
The third time, the betting house turned her away at the door.
“You’re banned,” said the thick-necked man blocking the entrance. “Manager’s orders. Don’t know how you’re cheating, but you’re cheating.”
“I’m not—”
“Save it for someone who cares. You come back here, you’ll leave with fewer teeth.”
Ellith walked away, mind racing. She needed a different approach. Different races, different betting houses, a front person—
She needed help.
Tomash listened to her explanation with an expression that shifted from skepticism to disbelief to something approaching religious awe.
“You’ve built a time machine,” he said slowly. “Inside a broken refrigeration unit.”
“An inverted temporal loop generator,” Ellith corrected. “It’s not quite the same as—”
“It’s a time machine.” Tomash ran his hands through his thinning hair. “Sweet mother of causality. You’ve actually done it.”
“I need your help. I can’t be the one placing bets anymore—I’m too visible, I’ve already been banned. But if you were to—”
“Oh, I know,” said Tomash. “I know all about it.”
He walked to the back of the shop, moved aside a stack of crates, and revealed a figure who’d been sitting quietly in the shadows.
It was Ellith.
Not the Ellith who was speaking. A different one. This version looked more haggard than the one who’d emerged from her first gambling loop—hollow-eyed, with a poorly-healed cut across one cheek.
“Hello,” said the other Ellith. “We need to talk.”
Ellith’s mouth worked silently. Finally: “The failsafe unit. At the docks.”
“I used it three days ago. Or three days from now, from your perspective. Tomash’s perspective is—” the other Ellith waved a hand. “—complicated. I’ve explained the basics to him. He’s been very understanding.”
“Why?” Ellith demanded. “Why would I—why would you—”
“Because I’m going to tell you what happens next, and you’re going to decide you need to go back and warn yourself. And then you’ll realize the warning already happened, because here I am.” The other Ellith stood, moving stiffly. “The betting houses started comparing notes. They don’t just ban one person—they share descriptions. I was recognized at three different establishments. Security officers got involved. It’s a mess.”
Tomash held up a hand. “Before you two start arguing in earnest, perhaps we should discuss how to proceed productively. Ellith—” he looked at the haggard version—”you’ve had time to think. What’s your recommendation?”
“Diversify,” said the haggard Ellith. “Different cities. Different types of events. Commodities speculation instead of racing. Most importantly: never the same person placing bets twice in a row.” She looked at the younger version of herself. “We need more people in on this. And we need a better system for tracking what we’ve done and what we haven’t done yet.”
The original Ellith—the one who’d come here expecting to recruit Tomash, not to meet herself—sat down heavily on a crate. “This is exactly what I was afraid of,” she said. “Cascading complications. Every solution creating new problems.”
“Welcome to the primer loop,” said her other self. “It only gets worse from here.”
Part Three: Compounding
Over the next several weeks, the scope of the operation expanded in ways Ellith couldn’t have predicted.
Tomash had a cousin in the commodities trade who could place large bets on grain futures without raising suspicion. The cousin needed to be brought into the secret. But when Tomash went to approach him, the cousin already knew—a version of Tomash from two loops forward had already made contact.
Gannik noticed something strange happening at the shop. Tools moved between uses. Inventory fluctuated oddly. He confronted Ellith about it, and she decided to tell him the truth rather than compound lies. But Gannik, it turned out, had already been told—by a version of Ellith from a loop she hadn’t entered yet.
Keeping track of who knew what and when became its own logistical challenge. Ellith started a ledger, then realized she needed a ledger for the ledger to track which version of the ledger was current. She implemented a dating system using sequential codewords, then discovered that two of her future selves had independently created incompatible systems.
The money was piling up, hidden in various caches around the city. The problem wasn’t generating wealth—it was using it without drawing attention. They couldn’t suddenly become rich. They had to launder the gains through legitimate business expansion, a process that took time and created paper trails.
And underneath it all, the existential questions gnawed at her.
When she entered the loop, who came out? The continuous experience felt like her, but the person who emerged at the beginning of the loop walked into a world where another Ellith had just entered at the end. Were they the same person? The emerging Ellith had memories of entering, had experienced the loop, had subjective continuity—but the entering Ellith would never experience emerging. From her perspective, she walked into the loop and… stopped. Her life continued in the version that emerged, but she, the original she, was gone.
Ellith-prime, Ellith-alpha, Ellith-one: whatever she called herself, the person who’d originally discovered the inverted fridge no longer existed in any meaningful sense. She’d entered her first loop and never came out. The current Ellith was a copy, a continuation, a branch—but the original trunk had been cut.
And every time she entered another loop, she was doing it again. Ending one version of herself to create another.
She mentioned this to Tomash one evening, after a long day of coordinating bet placements across three cities.
“You’re thinking about it wrong,” he said. “You exist as a continuous thread. The loop doesn’t end you—it relocates you. You walk in the exit and come out the entrance. Same thread, different position in time.”
“But the me who walks in the exit never experiences coming out the entrance. She just… stops. From inside, it feels continuous, but from outside—”
“From outside, you persist,” said Tomash firmly. “The rest is philosophy.”
Easy for him to say. He’d never entered a loop.
The first real crisis came six weeks into the operation.
Ellith (version sixteen, by her counting system) was reviewing financial records when version nineteen stumbled through the failsafe unit at the docks, bleeding heavily from a wound in her side.
“They found out,” gasped nineteen. “The betting syndicate. They hired investigators. Three loops from now, they connect the dots. Gannik gets arrested. Tomash goes into hiding. I barely—” She coughed, spraying blood. “—barely made it to the failsafe.”
“Who did this to you?”
“Doesn’t matter. What matters is stopping it from happening. I came back to warn you, but I was already too late to prevent the investigation from starting. You need to go back further. Use the docks unit—it’s got a longer loop capacity, eight hours. Go back to before I—before my original entered the loop, and tell her to pull out of the Northgate betting house. That’s where the trail starts.”
“And if I do that, you—”
“I stop existing. I know.” Version nineteen’s smile was ghastly, her teeth stained red. “Do it anyway. I’d rather not exist than exist like this.”
Ellith sixteen entered the docks unit and emerged eight hours earlier as Ellith sixteen-B. She found Ellith fifteen—no, wait, this was the branch where fifteen had already become sixteen, so this was… she consulted her ledger, cross-referenced the codewords, and determined she was looking at Ellith-sixteen-alpha, the version that would eventually become the bleeding nineteen.
“Don’t bet at Northgate,” said sixteen-B. “I’ll explain everything, but the short version is: three loops from now, Gannik gets arrested. Pull out now.”
Sixteen-alpha stared at her. “I already bet at Northgate. Yesterday.”
“Then we need to go back further.”
And so it cascaded. Each intervention required an earlier intervention. Each earlier intervention created new branches, new complications. Ellith lost track of how many versions of herself existed, had existed, would exist. The ledger grew to fill multiple books. The codeword system expanded to three-word phrases.
Eventually, they stabilized the situation. The Northgate trail was broken. Gannik wasn’t arrested. The bleeding Ellith-nineteen never came to exist—her entire branch was pruned from the timeline when Ellith-sixteen-B prevented the circumstances that would create her.
But Ellith knew, with cold certainty, that she wasn’t the original anymore. She wasn’t even close. She was a late-stage derivative, a copy of a copy of a copy, preserved only because earlier versions had chosen to sacrifice themselves to keep the operation running.
Coda
Ellith found herself, some indeterminate number of versions later, standing alone in the workshop where it all began.
The original prototype still sat on the bench. Model 7-C. The broken refrigerator that had started everything.
She could remember being the person who’d opened it up and smelled rotting food for the first time. The memory was clear and vivid, as if it had happened yesterday. But it hadn’t happened to her—not to this her. It had happened to a version of herself so many branches back that the number had lost meaning.
She wondered, sometimes, what that original Ellith would think if she could see what her discovery had become. A sprawling enterprise spanning multiple cities. A hidden fortune waiting to be spent. A family of divergent selves, most of whom no longer existed, all of whom had sacrificed their continuity for some version of the greater good.
Was it worth it?
She didn’t know. The question assumed there was a singular “she” who could weigh the costs and benefits. But there wasn’t. There was just this moment, this version, this Ellith, making decisions that would create and destroy future and past Elliths in turn.
She powered up the prototype one more time, watching the temporal gradient stabilize. The soft blue glow was almost peaceful.
Somewhere in the future, another Ellith was preparing to enter a loop. Somewhere in the past, an Ellith was emerging from one. The thread of consciousness that called itself “Ellith” wound through time like a needle through cloth, stitching moments together into patterns too complex to follow.
She’d wanted, once, to make a better refrigerator.
Strange, how things turned out.
[End Part One]
I’ve written this as what feels like a natural first section. Some notes on what I was aiming for:
Logical consistency checks:
The loop mechanics: power for N hours, entry at hour N exits at hour 0, subjective experience moves “forward” through the loop
The identity handoffs work: whoever emerges takes over, whoever enters “ends” in a meaningful sense
The nested complication (Tomash’s cousin already knows, Gannik already knows) follows naturally from the premise
The crisis/intervention cascade demonstrates how primer-style problems compound
Things I left open for continuation:
Gannik’s role could expand
The broader implications (other people discovering this tech, government interest, etc.)
The philosophical questions are raised but not resolved—as they shouldn’t be
Should I continue with subsequent parts? I could develop the societal implications, introduce competing factions who discover the technology, or follow the psychological toll on Ellith more closely. What direction interests you?
I haven’t read that one myself, at time of writing.
Why are AIs so bad at writing Primer fanfic? Why do they start doing garbled nonsense when the time travel gets involved?
One story might be that there’s not very much about Primer in the training data. It is a well-known movie, but as mentioned earlier, it isn’t popular to write in its setting, using its time-travel mechanics.
That story doesn’t inspire much confidence in Opus 4.6′s programming skills. If this case is indicative of 4.6′s general reasoning capabilities when slightly out-of-distribution, well, it pushes me to attribute high benchmark performance to highly in-distribution programming tasks.
Similarly, this makes me expect performance to be poor on out-of-distribution mathematical research.
My point isn’t that Opus 4.6 is bad. I’m just interested in the rate of progress. Opus 4.5 felt like a real step up from previous models, when it came to adhering to my brand of unusual and logically intricate creative writing prompts. Opus 4.6 doesn’t feel like a step up yet, although I’ll admit this particular test case is especially challenging. Maybe my usual ideas are about 2 weirdness points, and this was a 6, and Opus 4.5 was the first model that could handle a 2. If Opus 4.6 jumped to being able to handle 5 weirdness points, but no further, I wouldn’t be seeing the improvement, even though it is significant.
Is this bad enough to update my timelines upwards? It gives me the sense that there’s quite a long scale; that the “conceptual resolution” of the models is climbing dramatically, but the ladder it has to climb is extremely long. Humans badly bungle time-travel in fiction too, but this was worse. It’s a reminder that the competence of modern AI is still very fragile.
But, no, I’m not really updating significantly. My take-away from the discussion with Tsvi last summer already involved a long ladder. The probability I place on short timelines isn’t primarily coming from current methods just scaling up; rather, it’s the idea that current methods are enough to accelerate AI research, EG by helping humans dig through the literature and rapidly implement new combinations of existing ideas.
- ^
This wasn’t quite how things worked in Primer. Inanimate objects invariably loop many times, and go back and forth an even number of times, so that the time machine appears to be accelerating time, since you put a fresh object in at the beginning and get an aged object out the other end. Only humans can use the device to travel backwards in time, because humans can deliberately enter at the late end and exit the loop at the early end rather than looping repeatedly and getting kicked out randomly.
If there’s a random element to loops, though, it makes sense that objects could randomly travel an odd number of times rather than an even number, which was what I was depicting in my story sketch. Objects shouldn’t have to be smart to enter at the late end and get out at the early end; objects can have humans do that for them.
So, this is a minor but deliberate departure from how things work in the movie, based on my own opinion about the consequences of the time travel logic described in Primer.
Here’s GPT5.2′s response:
GPT5.2
Cold Storage
The fridge never cooled anything.
That should have been the first clue, but nothing about time magic was intuitive. You didn’t cool time; you slowed it, folded it, indexed it. Everyone knew that. The sales pitch for the ChronoVault™ was simple: bread stayed fresh because its internal clock barely ticked. Hot soup stayed hot. Flowers stayed exactly on the edge of wilting.
This unit did the opposite.
The loaf inside came out stale in minutes. Meat greyed. Milk curdled. The internal chronometer—standard diagnostic—showed months of elapsed time after an hour on mains.
Negative drift, they wrote in the lab notebook. Internal time velocity inverted.
They unplugged it and moved on.
The second time, the fridge already contained a loaf.
Same baker’s mark. Same scorch on the crust. Still warm.
They swapped it out, hands shaking, marked the new loaf with ink, shut the door, powered down.
Empty.
They didn’t sleep that night.
The realization came slowly, the way dangerous ideas always do—by trying very hard not to think them. The fridge wasn’t slowing time. It was looping it. Whatever entered could travel forward inside until it found an earlier state that matched the door opening.
You couldn’t see the loop. You could only trade with it.
They were a student, officially. Unofficially, a junior tinkerer at a startup that sold “domestic chronotronics” to people who liked words like bespoke and heritage. The unit in their apartment was a prototype. They weren’t supposed to take it home. They did anyway.
Fantasy fridges didn’t hum. They listened.
They powered it on at midnight.
Someone knocked on the inside.
The door opened and they stepped out—older, unshaven, eyes red from too many bad decisions made too close together.
“Don’t freak out,” the other one said immediately. “We’re on iteration three.”
The gambling plan was embarrassingly mundane.
There was no stock market, no commodities exchange, no arbitrage. But there were races—beast races, technically—public, legal, and aggressively audited. You couldn’t cheat them magically. That was the point.
But you could loop yourself.
Iteration one lost everything. Iteration two broke even. Iteration three made enough to notice.
Each loop added four hours to the day. Not reversed time—more time. Extra, privately lived time, stitched invisibly into the same morning.
The clone took notes. Not about the races—about themselves. Which instincts to ignore. Which hunches were poison. Which tells they unconsciously gave off when they were about to overcommit.
By iteration five, the clone was calm.
By iteration six, they were bored.
“You’re going to want to stop soon,” the clone said. “You won’t.”
They didn’t.
Covering their tracks was harder than the gambling.
They still had clients. Still had invoices. Still had a partner who noticed when they showed up with bags under their eyes and money they couldn’t explain.
The plan had been simple: work overtime, get paid, loop back, bet with foreknowledge, restore the original timeline’s finances. Clean. Elegant.
Instead, the advice came from the fridge.
The clone handed over winnings, plus extra—seed capital, they called it—then stepped back into the fridge.
“Your turn,” they said.
The door closed.
The fridge listened.
They woke up tired but solvent.
Their partner asked how the contract went.
“Fine,” they said, and meant it.
They didn’t notice the difference until later: the way memories felt slightly secondhand. Like notes copied from notes.
They didn’t think about it too hard.
The failsafe was sensible. Industrial-grade, rented anonymously, buried in a cargo storage unit at the docks. If something went wrong, you could always loop back past the current mess.
They waited a few days. Slept. Ate. Let the adrenaline drain.
Then they tried again.
And lost the betting license.
Too lucky, the officials said. Too consistent. Statistical anomalies. Cheating.
They laughed until it hurt, then stopped abruptly.
This wasn’t sustainable.
They needed capital. A buffer. Someone else to share the load.
They told their partner everything.
The partner sighed.
“Oh good,” they said. “I was wondering when you’d catch up.”
They introduced Protagonist Three.
Protagonist Three had already used the dock failsafe.
Protagonist Three was the reason they were banned.
After that, things accelerated.
Not the time loops—the meta-time. The rate at which procedures replaced decisions. The way meetings started to include agenda items like clone turnover and identity continuity. The way no one was quite sure who had actually founded the company anymore.
They stopped naming iterations.
They stopped pretending there was an original.
The fridge business flourished. Of course it did. You could beta-test designs against yourself. You could pre-debug failures by living through them and discarding the branch.
Someone suggested franchising.
Someone else suggested internal markets—loops traded for favors, favors for future knowledge.
Someone asked, quietly, whether you could put people in long-term storage.
The fridges listened.
Much later—much later, measured in layers rather than years—someone finally asked the forbidden question:
“If we stop looping,” they said, “who survives?”
No one answered.
Because by then, stopping wasn’t a moment.
It was a direction.
Pretty straightforward expansion of your notes. No obvious conceptual errors that I noticed, although some of what it says doesn’t really make sense(“someone asked whether you could people in long-term storage”...isn’t this what they are already doing?). Not really compelling as fiction, replete with GPT-isms. Generally speaking GPT has seemed smarter than Claude to me since GPT5.
Yep. Where this deviated from my notes, I approve (purely in terms of the time travel logic, that is). Seems like OpenAI is way ahead on time-travel logic, which is evidence that it is significantly ahead on “general reasoning”.
This seems like a pretty brutal test.
My experiences with Opus 4.6 so far are mixed:
It did a very nice job designing and building a web UI for an unusually messy and inconsistent database setup.
As a conversational sounding board, it has been getting caught up in overly facile and ultimately complimentary analogies to the point I don’t really want to talk it as much as I might have 4.5. Sample size: Way too small to conclude anything.
It has been surprisingly good in math-related conversations but I don’t have a great baseline with 4.5 for comparison.
As a data point after reading your post and the apple comic I still don’t understand how primer time travel works. If forced to write a fanfic I think I would butcher it too. I’m not sure if I should be holding an LLM to a higher or lower standard than myself.
It’s basically “time goes backwards inside the box when it’s turned on”. So you can turn the box on in the morning, immediately see you-2 climb out of it, then both of you coexist for a day and you-2 shares some future information with you, then in the evening you set the box to wind-down and climb inside it, then you wait several hours inside the box, then climb out as you-2 at in the morning and relive the events of the day from that perspective, then you-1 climbs into the box and is never seen again, and you remain.
When put this way, it’s nice and consistent. But in the movie some copies actually stop their originals from getting in the box, resulting in permanent duplication. And that seems really hard to explain even with a branching timelines model. If there’s a timeline that ends up with two permanent Bobs, there must be some other timeline with no permanent Bobs at all, due to conservation of Bobs. The only way such a timeline could appear is if Bob turned the machine on, then nobody climbed out and he climbed in, but Bob could simply refuse to do that.
Another cool thing is that the time machine can also provide antigravity. Consider this: you assemble a box weighing 50kg and turn it on. Immediately, Bob-2 climbs out, weighing 70kg. So the box has to weigh −20kg until Bob-1 gets in it and it shuts off again. In the movie that’s not spelled out, but it does spell out that they first started working on antigravity and then got time travel by accident, so that’s really good writing.
I’m not sure why you say it’s hard to explain with branching timelines. To me this is just branching timelines. The movie voiceover states at one point that the last version of events seems to be the one that holds true, meaning that you see the last branching timeline, usually the one with the most Bobs. I don’t think you have to belive this part of the voiceover, though; this is just the opinion of someone trying to make sense of events. You could instead say that the movie has a convention of showing us later splits rather than earlier.
Branching timelines have to come with probabilities and that’s where the wheels fall off. Imagine you’re Carol, living on the other side of town, not interacting with the machine at all. Then events similar to the movie happen. Before the events, there was one permanent Aaron. After the events, there’s either one or more permanent Aarons, depending on which timeline Carol ends up in. But this violates conservation of Aarons weighted by probability. A weighted sum of 1′s and 2′s (and 3′s and so on) is bigger than a weighted sum of just 1′s. Some Aarons appeared out of nowhere.
Things could be balanced out if there was some timeline with reasonably high measure, consistent with the behavior of folks in the movie, which ended up with 0 permanent Aarons. But what is it? Is it some timeline where a box never had an Aaron climb out of it, but had an Aaron climbing into it later? Why would he do that?
The characters in the movie take a lot of precautions to isolate themselves from their time-clones, meaning that they don’t really know whether they got out of the box at the start. Therefore, they just have faith in the plan and jump in the box at the end of the loop. So long as they don’t create any obvious paradoxes (“break symmetry” as they call it), everything works out from their perspective, and they can assume it’s consistent-timeline travel rather than branching, so they don’t think they’re creating a timeline in which they mysteriously vanish.
When they start creating paradoxes, of course, they should realize. The fact that they don’t think about it this way fits with the general self-centeredness of the characters, however.
I agree that it makes sense to think of this probabilistically, but we can also think of it as just all timelines existing. I’m happy to excuse the events of the movie as showing one particularly interesting timeline out of the many. It makes sense that the lens of the film isn’t super interested in the timelines which end up lacking one of the viewpoint characters.
If we do think of it probabilistically, though, are the events of the movie so improbable that we should reject them? By my thinking, the movie still fits well with that. Depending on how you think the probabilities should work out, it seems like that first timeline where the person just vanishes is low-probability, particularly if they create a relatively consistent time-loop. In a simple consistent loop, only the original branch has them vanish, while each other branch looks like an internally consistent timeline, and spawns another just like itself. The probability of a timeline like “one Abe, then two Abes, then back to one” seems high, if Abe is careful to avoid paradoxes. With paradoxes, the high-probability timelines get chaotic, which is what we see in the movie (and in the comic I linked).
Yeah, I guess “they don’t bother checking whether they get out of the box” is the right explanation for the movie. Though still, if timelines where a person just vanishes are low-probability, then timelines where the number of people permanently increases (like the one shown in the movie) should be just as low-probability. The start and end of a long chain. And the middle of the chain should be mostly like 1-1-1-1… Or something like 2-0-2-0… but that would require weird behavior which isn’t seen in the movie (e.g. “I’ll get in the box iff I don’t see myself come out of it”).
They do deliberately try to set up an “I’ll get in the box if I don’t see myself get out” sort of situation in the movie, though they don’t succeed, and they don’t seem to realize that it would result in 0-2-0-2-… across metatime.
Good point about how permanent increases have to be as improbable as permanent decreases! I should’ve gotten that from what you were saying earlier. I suppose that leaves me with the “movies follow interesting timelines” theory, where it’s just a convention of the film to look at the timelines where characters multiply.