The Witness

The dilution fridge hummed at 1.4 kelvin, and Haneul had his back to it because the readout was easier to ignore that way.

Minjun stood at the bench with his laptop open, the process-matrix code running its final pass. It was 2:14 a.m. The lab had the particular quiet of a building full of people who had gone home, and the quiet made every keystroke sound like a decision.

“Again,” Minjun said. “The LHS under the diagonal noise model is 1.07. Threshold is 1.0. We’re above.”

“I know the number,” Haneul said.

“You knew it last week. You knew it when you ran the off-diagonal model and got 0.98.”

“0.979,” Haneul said. “Rounded.”

Minjun looked up. “Does the rounding matter to you?”

“It matters to the sentence after ‘violates.’”

The fridge cycled. A pump somewhere in the wall ticked. Haneul had brought a thermos of coffee and a folder of printed logs, and the logs were the reason he was still here at 2 a.m. instead of home in bed pretending he had not seen them.

Minjun’s claim was clean on paper. Control qubit in superposition, photon through two gates in both orders at once, process matrix reconstructed from tomography, causal witness inequality computed, left-hand side above threshold. Indefinite causal order. A result the field would cite, because the field cited the ones that shipped.

Haneul’s problem was the noise model. The process tomography gave you a matrix, and the matrix gave you a number, but the number depended on how you modeled the noise, and Minjun had modeled it diagonal, uncorrelated, the simple case that made the math tractable. Haneul had spent three weeks on the alternative. Off-diagonal correlations, the kind that show up in real devices, the kind that the tomography itself cannot fully rule out at this sample size. Under that model the LHS came out 0.979. Below threshold. No violation. No discovery.

The gap between the two models was 0.091. The margin above threshold under Minjun’s model was 0.07. The margin was smaller than the gap, and Haneul had said so three times already and Minjun had heard it three times and here they were.

Minjun closed the laptop halfway. “You want me to hold the paper.”

“I want the tomography redone. Six weeks. Better sample size, full correlation matrix, and we settle it.”

“Six weeks.”

“The grant review is in November. I know.”

Minjun smiled. It was not a warm smile. “You know.”

“I know.”

The grant was Minjun’s, not Haneul’s. Haneul was a postdoc on a two-year appointment that ended in March. If the paper went in now, Haneul’s name would be on it, second author, and the second author of a quantum-switch paper that got cited was a person who got interviews. If the paper went in now with a claim that did not survive the off-diagonal model, Haneul would be the second author of a paper that might not survive contact with a referee who asked the noise question. Haneul had been that referee. He knew what the question looked like from the other side of the desk.

“Let me tell you what six weeks costs,” Minjun said. He held up a hand and counted. “The department already knows the result. I presented it in September. If I walk back a retraction in November, the renewal committee reads it as a failure of judgment, not caution. The grant is 340,000 euros. My lab, my students, the fridge, the cryostat, the tomography software you want to use for your six weeks. All of it. You understand that.”

“I understand it.”

“And you’re asking me to risk it on a model that gives 0.979, which is, I will note, a number that is consistent with indefinite causal order. It is not inconsistent. It is compatible.”

“Compatible with,” Haneul said.

“Compatible with indefinite causal order. Which is true. The data do not exclude it. That is a real statement, and it is a publishable statement, and it is honest.”

“It is true and useless,” Haneul said. “Every quantum-switch paper from the last five years is compatible with indefinite causal order. The ones that got cited are the ones that demonstrated it. You know that.”

Minjun did not answer for a moment. He looked at the fridge, at the readout Haneul was not looking at. “Yes,” he said. “I know that.”

Minjun turned back to him. “You think I don’t know what I’m asking. I’m asking you to let me publish a number that is true under one assumption and false under another, and to call it a discovery, and you are asking me to let you hold it until the assumption is settled. Both of us can prove our logs. I can show you the diagonal reconstruction, the tomography counts, the witness computation, step by step. You can show me the off-diagonal reconstruction, the correlation terms, the recomputed LHS. Both of us are right about what our own numbers say. The only thing we disagree on is which numbers get to be called the result.”

“Yes,” Haneul said.

“And the reason I get to decide is not that I’m right. It’s that I’m the PI and the grant is in November. That is the actual reason. You know it and I know it and it is not a scientific reason.”

Haneul said nothing. The fridge hummed.

Minjun opened the laptop again. “I’ll offer you this. We publish the weaker claim. ‘Compatible with indefinite causal order.’ We state the noise model explicitly. We note that under an alternative correlation model the witness is not violated. Full transparency. A referee can read it and decide. We do not claim a demonstration. We claim compatibility. That is honest.”

“It is honest,” Haneul said. “It will not be cited. It will be filed. The next group that demonstrates a violation will cite the ones that demonstrated, not the one that said it was compatible. You know that too.”

“I know that.”

“And you want to publish it anyway.”

“I want the paper to exist. I want the grant to renew. I want the lab to keep running. Those are not the same want, and the first one is the one I’m asking you to trade for the other two.”

Haneul set down his coffee. The thermos was still warm. He had drunk half of it over the last hour, and the half he had drunk sat heavy in his gut.

“Six weeks,” Haneul said. “The tomography is the bottleneck. We have the cryostat time, we have the software, we have the photon source. What we don’t have is the sample count. The current run is 10 to the fifth shots per setting. For a full correlation matrix we need 10 to the seventh, and that is the six weeks. It is not padding. It is the difference between a matrix that constrains the noise and one that doesn’t.”

“I know what it is.”

“Then you know it is not a request. It is a requirement. If we publish now, we publish a number that we know changes sign under a model we cannot rule out. That is not a discovery. That is a choice of model.”

Minjun closed the laptop. He did not close it gently. “And if we don’t publish now, I lose the grant, and you lose the paper, and the next group publishes first, and we get nothing. Not a citation. Not a filing. Nothing. You understand that the system does not reward caution. It rewards shipping. I am not wrong about that.”

Haneul opened his mouth and closed it. Minjun was not wrong. Haneul had watched it happen twice already in his own career. The first time, a colleague at another group had published a result with a noise model Haneul suspected was wrong, and Haneul had written a careful comment, and the comment had been published, and the original paper had been cited forty times and the comment had been cited four. The second time, Haneul himself had held a result back for three months to check a calibration, and by the time he submitted, another group had the same result with a weaker check, and Haneul’s paper was the second one, and second ones are not remembered.

He knew the system. He knew what it rewarded. He knew Minjun was not wrong, and that was the worst part, and saying it again would not change anything.

“The grant is yours,” Haneul said. “The result is not yours to publish if I don’t sign.”

Minjun looked at him. The look lasted a long time. “You would do that.”

“I’m telling you I will.”

“You understand what that costs you. Not me. You. The paper, the authorship, the letter I would write for you. All of it. You walk out of here without signing, and you are a postdoc who refused to sign a paper, and that is a thing that follows you.”

“I know what it follows me as.”

Minjun sat down. He put his hands flat on the bench, palms down, and stared at them. “What do you want, Haneul? Not what you think is right. What do you want?”

Haneul thought about it. The fridge hummed. The thermos was cold now.

“I want the number to mean something,” he said. “I want to publish a result that is a result, not a choice. And I want to be able to say, when someone asks me at a conference, yes, we checked the noise model, and the violation holds, and mean it. I can’t mean it now. I could mean it in six weeks.”

“And if it doesn’t hold? If you run the tomography and the LHS comes out 0.97? Then you have spent six weeks and a grant to learn that you were right to be cautious and wrong to be hopeful, and the paper is dead either way.”

“Then the paper is dead, and at least it’s dead honestly.”

Minjun stood up. He walked to the fridge and rested a hand on the lid without checking the temperature. “Email the data,” he said. “If you’re going to refuse, refuse clean. Send it to the shared archive tonight. Don’t leave it in your folder where it looks like you’re hoarding it. Put it where everyone can see it, including me, including the committee, including whoever reviews the paper if there is a paper. If you’re going to make this a question of record, make it a question of record.”

Haneul looked at him. Minjun did not look back. He was looking at the readout now, finally, the number on the screen that said 1.41 kelvin, stable, within spec.

“Okay,” Haneul said.

He opened his laptop. The shared archive was a folder on the group’s server, and the data was 4.2 gigabytes of tomography counts, reconstruction code, and logs. He had already organized it. He had organized it three weeks ago, on the night he finished the off-diagonal model, because he had known then that this night would come, and he had not known what he would do with it, but he had known he would need it ready.

The upload took eleven minutes. Minjun stood at the fridge with his hand on the lid and did not move. The lab was quiet except for the upload bar and the hum of the fridge and the tick of the pump in the wall.

When it finished, Haneul closed the laptop. He put it in his bag. He picked up the folder of printed logs, the ones he had brought, the ones that proved his model, and he put them in the bag too. Then he stood up.

Minjun was still at the fridge. “Haneul.”

“Yes.”

“The paper is still open on my screen. I haven’t closed it. I could still submit it tonight. You know that.”

“I know.”

“I’m not going to. Not tonight. But I could have, and you know I could have, and that is the thing you are walking out with. Not the data. The knowledge that I could have.”

Haneul put his bag on his shoulder. The bag weighed almost nothing. The data weighed nothing. The logs weighed almost nothing. He had carried heavier things to class when he was twenty.

“Goodnight, Minjun,” he said.

He walked to the door. The corridor outside was lit by the emergency lights, the ones that never turned off, the ones that made the floor look like a photograph of a floor. His shoes made a sound on the linoleum that he had heard a thousand times and never noticed until now.

Behind him, the fridge hummed. The upload was done. The data was in the archive, in the folder everyone could see, including Minjun, including the committee, including whoever would review the paper if there was a paper. It had not required Minjun’s permission.

He pushed through the fire door and stepped into the cold air of the courtyard. The sky was the gray that comes just before dawn, not dark anymore, not light yet. His breath made a small cloud that disappeared. He walked across the courtyard toward the gate, and the gate was open, and beyond it the street was empty, and the empty street was the first thing he saw that was not the lab.

The folder of raw data sat in his bag. He did not look back.