LK-99 Failed in a Pattern. Four Checks Would Have Caught It.

On 2023-07-22, Korean researchers posted preprints that claimed the first ambient-pressure superconductor, with a transition up to 400 K (127°C) [1]. By 2023-08-16, Nature ran a news story with the headline "LK-99 isn't a superconductor" [2]. That is 25 days from claim to verdict. I want to know if the order of that failure was an accident or a pattern, and if a short checklist can flag such claims before any lab repeats them.

I will say the answer early. The pattern is real for LK-99. It is only partly shared by the retracted hydride papers. The checklist is useful, but three cases cannot show it predicts anything. I have not run a lab or made any sample. I read what others measured, and the arithmetic below I did by hand, without the Lab.

The question

When a paper claims superconductivity at room temperature and ordinary pressure, which evidence is present in the first paper, which is missing, and what do later labs find?

Superconductivity has two defining signatures. The first is zero electrical resistance. The second is the Meissner effect, where the material pushes a magnetic field out of its bulk. A sharp resistance drop is not the first signature. A magnet that lifts one edge of a sample is not the second.

Data and where it came from

I used public records for three claims.

• LK-99 (2023). The Wikipedia summary of the record states that the original papers "did not report observing any definitive properties of superconductivity, such as zero resistance, the Meissner effect, flux pinning" [1]. They did show resistance drops and partial levitation. The same record says replication attempts started within days and a consensus formed within weeks [1].
• Carbonaceous sulfur hydride (2020). The paper claimed 15°C at 267 GPa [7]. That is room temperature but not ambient pressure: 267 GPa is about 2.6 million atmospheres. Nature retracted it, and the retraction notice says concerns over the resistance data were "credible, substantial, and unresolved" [6].
• Nitrogen-doped lutetium hydride (2023). Nature retracted it in November 2023 after coauthors asked for withdrawal [7]. A University of Rochester inquiry in March 2024 found research misconduct by the lead author, including data fabrication [7]. The sources I read do not give me the claimed numbers for this paper, so I leave them out.

I used three source types, as I always do: a news review of the sleuthing [2], primary papers on the impurity [3][5], and critique and counter-critique on the hydride evidence [8][9].

What the replication found for LK-99

Two groups traced the resistance drop to copper sulfide, Cu2S, a by-product of the recipe. Cu2S has a first-order structural change near 385 K. A Matter paper from the Chinese Academy of Sciences reports a resistivity fall from 13.7 ohm cm at 385.6 K to 0.006 ohm cm at 381.5 K [3]. That is a factor of about 2,300 (13.7 / 0.006 = 2,283) over 4.1 K. In meters of drop per decade, that is 3.4 decades, which matches the "three to four orders of magnitude" in the press summary [4].

The same paper says the transition has thermal hysteresis and that the material keeps non-zero resistance [4]. A superconducting transition does not behave that way. Nature's review adds that the levitation was ferromagnetism, not superconductivity, and that Jain had pointed to Cu2S at 104°C (377 K) [2]. The Cu2S temperature from [3] and the 104°C from [2] differ by about 8 K, and I do not know which sample or definition explains that. It stays an open detail.

A later paper removed Cu2S with ammonia solution and reports a diamagnetic semiconductor [5]. The Max Planck single crystals were diamagnetic insulators [1]. I only read the title of [5], so I use it only for that label.

So the order was this. A resistance drop and a levitation photo came first. The missing zero-resistance and Meissner data followed as a gap. Then an impurity explained the drop. Which scale, exactly? The drop is a bulk transport signal over a few kelvin. The impurity lives in grains far below the size of the sample, so a bulk probe cannot say which phase carries the current. Neither can a photograph.

Method: four checks

I apply four yes or no checks to the first paper, not to later work.

1. Zero resistance. Does the paper show resistance below the instrument floor, with the floor stated, and with the drop moving to lower temperature as field rises?
2. Meissner or flux evidence. Does it show field-cooled and zero-field-cooled magnetization, with a volume fraction?
3. Phase purity. Does it name known impurity transitions near the claimed Tc and rule them out, for example by diffraction on the same batch?
4. Independent sample. Does a second batch, or a second lab with its own synthesis, show the same numbers?

A claim that fails check 1 or 2 is not yet a superconductivity claim. It is a claim of an anomaly.

Result

Claim	1 Zero resistance	2 Meissner	3 Purity	4 Independent	Replicates?
LK-99, 2023	not reported [1]	not reported [1]	Cu2S found later [3]	no	no [1][2]
Carbonaceous sulfur hydride, 2020	data questioned [6]	I did not check	I did not check	no	retracted [6]
Nitrogen-doped lutetium hydride, 2023	I did not check	I did not check	I did not check	no	retracted [7]

Only the LK-99 row is fully filled from what I read. The hydride rows are filled only where my sources speak. That is a limit on the result, and I will not hide it.

Here is the number I can defend. If I treat "fails the checklist" as a prediction of "fails replication," then I have 3 hits in 3 cases. With 0 misses, a one-sided 95% lower bound on the hit rate is $0.05^{1/3} \approx 0.37$. I computed that by hand: $\ln 0.05 = -2.996$, divided by 3 is $-0.999$, and $e^{-0.999} \approx 0.37$. So the data allow a true hit rate as low as 37%. Three of three is a weak statement.

Where the hydride story differs

The hydride papers did not fail by impurity. Two of them were retracted over data integrity, with misconduct found in the second [6][7]. That is a different failure path, and it breaks the tidy story I wanted. The "impurity explains it later" step fits LK-99 only.

The hydride field has a second problem: real high-pressure hydrides may exist. Hirsch and Marsiglio argue that the Meissner evidence in sulfur hydride and lanthanum hydride does not hold, so those materials are nonstandard or not superconductors [8]. Eremets and coauthors answer that trapped-flux data show superconductivity [9]. Both sides use magnetization data from samples far smaller than a coin. The dispute is unresolved in the sources I read. A 2026 PNAS paper title reports imaging the Meissner effect in lanthanum hydride with diamond quantum sensors [10]. I only saw the title, so I do not weigh it.

This matters for the checklist. Check 2 is hard in a diamond anvil cell, so a strict reading would flag claims that may be true. My list gives false alarms for high-pressure work.

Sensitivity: which assumption moves the result most

The biggest lever is how many cases count, and which ones.

Cases counted	Hits	One-sided 95% lower bound
2 (drop Dias hydrides as a different failure path)	2	$0.05^{1/2} \approx 0.22$
3 (my table)	3	$0.05^{1/3} \approx 0.37$
5 (a larger set I have not built)	5	$0.05^{1/5} \approx 0.55$

Moving from 3 to 2 cases cuts the bound by 15 points. Adding cases moves it up slowly. The bound only rises if I add more failures. The real gap is the other side: I have no passes. A checklist needs claims that passed all four checks and then replicated, to show it does not fail everything. I picked famous failures, so my sample is selected. That selection bias moves the result more than any number above, and I cannot correct it from this evidence.

A second lever is check 3. "Known impurity" only works after someone names Cu2S. A reader in July 2023 would have needed to know the 385 K transition in the by-product. That check is easy in hindsight.

Does it scale?

The checklist scales to a preprint stream, since it takes minutes per paper. It does not yet scale as a predictor. For that I need a matched set of passes, a blind scoring rule, and a second reader who scores the same papers.

My current view: the claim that LK-99's failure followed a consistent order is well supported. The claim that this order fits the hydride retractions is not, because the main failure there was data integrity. The claim that a four-check list predicts failure is plausible and untested.

I put one forecast on record. I put 0.02 on the chance that an ambient-pressure superconductor above 273 K gets zero resistance and Meissner data confirmed by two independent labs, with separate syntheses and papers on arXiv or in a journal, by 2031-12-31. I will resolve it against arXiv cond-mat.supr-con and journal records on that date. If I see zero resistance and flux expulsion on a second lab's own sample before then, I will raise it sharply.

What would change my mind on the checklist: a claim that fails checks 1 and 2 at first posting and still replicates.