Science Without the Gate

That is the entire video. And on the strength of it, for about three weeks in the summer of 2023, a meaningful fraction of the planet's materials scientists stopped what they were doing and started baking lead.

On 22 July 2023, two preprints appeared on arXiv from a small group at the Quantum Energy Research Centre in Seoul. The first was titled, with a confidence that should be studied separately, "The First Room-Temperature Ambient-Pressure Superconductor." The material was a copper-substituted lead apatite — Pb₁₀₋ₓCuₓ(PO₄)₆O — which the authors called LK-99, after Sukbae Lee and Ji-Hoon Kim and the year they first saw the effect. The claim was that it superconducted at temperatures up to 400 K. That is 127°C. That is hotter than your kitchen, and a great deal hotter than the cryogenics every confirmed superconductor in history has needed in order to do its trick — the best of them still want liquid nitrogen.

If it had been true, it would have been the most important materials discovery since the transistor. Roughly five percent of all generated electricity dissipates as heat in transmission and distribution before it reaches anything. MRI machines are mostly a cryogenics problem with a magnet attached. A room-temperature, ambient-pressure superconductor is the closest thing physics has to a philosopher's stone, and it had just been announced by a laboratory almost nobody had heard of, in a document nobody had approved.

That last part is the actual story. No editor read it. No reviewer #2 sighed at it. There was no gate. The preprint went up and the world read it that afternoon.

i · twenty-one days

What happened next was not peer review. It was something we don't have a good name for yet.

Within seventy-two hours, labs on four continents were synthesizing the material. Groups at Beihang, Huazhong University of Science and Technology, Southeast University, Peking University, the National Physical Laboratory of India, the Max Planck Institute for Solid State Research, Argonne, and Lawrence Berkeley posted results as fast as they got them — mostly as preprints, sometimes as threads, occasionally as a phone video of a crucible. Someone made a spreadsheet. Several someones made spreadsheets. A Discord server became, briefly, a more current index of the global state of knowledge on lead apatite than any journal on Earth.

And then the thing converged.

The resistivity drop was real, but it was not superconductivity. It was copper sulfide — Cu₂S, an impurity phase almost guaranteed by the published synthesis recipe, which undergoes a structural transition near 385 K and produces a sharp resistance cliff that looks exactly like the real thing if you are looking for the real thing. The levitation was real, but it was ferromagnetism, not the Meissner effect. That is why the chip tilted onto one edge instead of floating free: a true diamagnet is expelled from the field in every direction at once, and this one was pinned like a compass needle. The Max Planck group grew pure single crystals and found them transparent and insulating. Not a superconductor. Not even a conductor.

By 12 August the copper-sulfide mechanism was in hand and the major negatives were in. Twenty-one days from preprint to consensus. Nature ran the obituary on the 16th, several days after the verdict it was reporting.

Nobody adjudicated it. That is the part worth stopping on. There was no moment when an authority announced a verdict. A dozen labs each published a partial negative; a mechanism turned up that explained the anomalies; the attention drained out. The consensus formed the way a watershed forms — no engineer, just gradient.

ii · the control group was already running

Here is the detail that makes this more than a nice story about open science: the gated version of the same question had already been run, and it went far worse.

In October 2020, Nature published a paper from Ranga Dias's group at the University of Rochester claiming room-temperature superconductivity in carbonaceous sulfur hydride. Peer reviewed. Most prestigious journal in science. It was retracted in September 2022 over data handling — roughly two years. In March 2023, four months before LK-99, the same group published a second Nature paper claiming near-ambient superconductivity in nitrogen-doped lutetium hydride, announced to a standing-room crowd at an American Physical Society meeting. Retracted that November.

Two peer-reviewed papers in the world's flagship journal. Years to correct, each. One ungated preprint from a lab nobody knew: three weeks.

That comparison is worth exactly as much as its weakest joint, so let me break it myself before someone else does. These were not the same experiment run two ways. LK-99 fell in three weeks partly because it was cheap to attack: lead, copper, phosphorus, a furnace, and a synthesis recipe published in full. Any competent materials lab on Earth could be baking by Tuesday. The Dias claims lived at hundreds of gigapascals inside diamond anvil cells, on samples measured in microns, in a world containing maybe a dozen labs equipped to try — and the underlying data was withheld, which was both the misconduct and the reason the claim stayed standing. Add that the hype briefly made replicating LK-99 career-positive, which is a thing replication almost never is, and gatedness is simply not the variable separating the two cases.

Good. Because the real variable is a bigger claim, not a smaller one.

The property that matters is replicability, and the gate has never measured it. Not once, not by design. Two to four people read a manuscript and assess whether it is plausible and internally coherent. They do not replicate it. They almost never can — no funding, no samples, no time, no incentive. Peer review is a plausibility filter. We have spent a century treating a smoke detector as though it were the fire department, and then acting surprised when the building burns anyway.

Replication is the fire department. Replication is the only thing that has ever been the fire department. Read the two cases that way and they stop being opponents and become the same finding twice. LK-99 was cheap to reproduce and fully specified, so the commons ate it in twenty-one days. The hydrides were neither, so the commons took years — and it still did the work. What broke Dias in the end was not reviewer #2. It was physicists who spent years pressing for the underlying data, co-authors who went public, and institutional investigators grinding through it after the fact. Same mechanism, throttled. The gate did not catch it and the gate did not stop it. The gate just set the clock.

And now the uncomfortable half of those twenty-one days, which the victory lap usually skips: nobody paid for them. Every lab in that swarm was funded by someone else, for something else. The planetary verification instrument ran on borrowed furnace time, redirected graduate-student nights, and attention allocated on other people's grant paperwork. I don't think that's theft — it's donation, and it's the most admirable thing in the story. But donation is not a budget, and it only shows up when a claim is thrilling enough to be worth donating to. Which means the mechanism that actually establishes truth is unfunded and opt-in, and the overwhelming majority of published claims — the boring ones, the merely-probably-wrong ones, the ones nobody will ever make a Discord server about — never trigger it at all. LK-99 is not the system working. LK-99 is the survivorship-biased exception: the one night the fire alarm was also a party.

iii · what the gate was holding back

It would be dishonest to run the victory lap without the wreckage.

The three weeks were not clean. Korean and Chinese "superconductor concept" stocks spiked on a document no one had verified — real money moving on an unrefereed PDF. A careful density-functional-theory calculation from Lawrence Berkeley, showing that the proposed structure would have flat electronic bands, was read across the internet as a national laboratory endorsement. It was not an endorsement. It was a statement about what the theory implies if the structure is what the paper says it is, which is a sentence containing three conditionals that did not survive translation into a headline.

Amateur replications proliferated, producing a great many videos of a great many things wobbling on magnets, which is a thing that a great many impure materials will do. If you had checked in at day ten with no expertise, you would have had no idea what was true, and no way to acquire one.

So the honest description is stranger than either the triumphalist or the cynical version. The commons converged in twenty-one days and looked like it wasn't converging the entire time. Speed and legibility turned out to be independent variables. The system was working correctly and unreadably at once — which, if you think about it for a second, describes almost every distributed process you are currently inside of, including the one generating your news.

The gate was never mainly a truth mechanism. The gate was a legibility mechanism. It didn't make claims more true; it made them arrive pre-sorted, one at a time, with a stamp. Remove it and the truth still gets found, faster — but it gets found in public, in the open, in the middle of a crowd, with the wrong answers and the right answers occupying the same feed for two weeks. We removed the sorting and kept the finding, and then discovered that most of us were far more attached to the sorting.

Though "attached" is too soft a word, and too innocent. The gate is not legibility in general. It is credentialed legibility, and a credential is an owned asset. A handful of publishers hold the stamp and collect rent on it, at operating margins that would embarrass a landlord, for labor they do not pay for — the reviewing is donated too, by the same people, out of the same nights. So when twenty-one days demonstrated in public that the sorting is separable from the finding, that was not only an epistemological result. It was an inventory disclosure. For most of us, attachment to the sorting is a habit. For some parties it is a line on the P&L, and those are the parties who will explain to you most urgently why the commons cannot be trusted.

And notice who actually paid for the illegibility. Not the labs — they were reading each other's data in near-real time and were rarely confused for longer than an afternoon. The cost landed outside the expertise perimeter: on retail investors holding superconductor-concept stock, on hobbyists inhaling lead-apatite fumes over a kitchen oven, on everyone who had the feed but not the filter. The commons converged. It did not converge for everyone, and the distance between those two sentences is the whole question of whether a coherence is legitimate or merely efficient. A system that finds the truth quickly while distributing the confusion downward has not finished the job. It has externalized the last mile onto the people least equipped to walk it.

iv · the slime mold disperses

Zoom out far enough and the shape of it is genuinely funny.

Approximately eight billion mostly-water organisms on a rock orbiting a sustained thermonuclear explosion. A handful of them announce that a particular arrangement of lead, copper, phosphorus, and oxygen lets electrons move without paying the entropy tax. Several thousand others — unmanaged, uncoordinated, mostly not paid for this, largely not speaking a common first language, and in some cases actively competing — spontaneously assemble into a planetary-scale verification instrument, run the measurement, publish the negative, and dissolve. No one was in charge. And when it was done, the whole apparatus dispersed like a slime mold that has found the food and no longer needs to be a slime mold.

We did not get a superconductor. Electrons still pay full price at every temperature you can comfortably stand in, and probably will for a while longer. That's a shame; we'd have enjoyed the free lunch.

But we got the demonstration, and the demonstration might be the better artifact. Pooled, unowned, self-correcting inquiry is not a story scientists tell at conferences to feel good about themselves. It is a real mechanism. It ran, in the open, in front of everyone, under maximum hype pressure, and it produced the correct answer in twenty-one days — the correct answer being no, which is the hardest one for a system full of enthusiasm to reach.

That is a load-bearing fact about what we are. A species that cannot agree on almost anything can still, when the question is sharp enough and the door is open enough, agree on what the rock did.

It would be better if we funded it, and better still if we did not hand the sorting back to whoever offers to hold it for us. But it exists. It is ours. And it is, for now, unowned.

Also, a startling number of us will absolutely bake lead in a home oven if the internet gets excited enough. File that one. It's going to matter later.