The proof
Tested against a century of settled physics.
The published record is the answer key. We stand on the field's own literature, test inside a closed environment that runs in both directions, and the physics underneath the platform is independently validated by people who owe us nothing.
every superconductor the published record has settled · 17 papers cited across 5 problem classes, every link verified · independently validated at Stanford and SLAC
The field already wrote the answer key
Superconductivity has been measured, published and archived since 1911. The literature runs to tens of thousands of materials: the field's reference database alone holds around thirty thousand records, the preprint stream adds new measurements daily, and the decisive results sit in journals anyone can open. That century of settled physics is what our tests draw on, and it is where every claim on this site can be checked.
- SuperCon, the reference databaseRoughly thirty thousand records of measured superconductors, curated at Japan's National Institute for Materials Science.
- arXiv · cond-mat.supr-conThe live preprint stream where the field publishes first. New measurements arrive here daily.
- Physical Review LettersOne of the journals of record, and where the independent measurement cited below was published.
The problems we work on have a literature
Every brief the platform takes lands in a field somebody has already measured: decades of published work on films, interfaces, junctions and wires. These are the papers we check ourselves against, grouped by problem class, and every link was opened and verified before it shipped. Reading them is the fastest way to see the standard our answers have to meet.
Nitride films
Superconducting nitride films, NbN and NbTiN, are the working layer of single-photon detectors and superconducting electronics, and film quality decides everything downstream.
- Picosecond superconducting single-photon optical detectorGol'tsman, G. N. et al. · Appl. Phys. Lett. 79, 705 (2001)
- Superconducting nanowire single-photon detectors: physics and applicationsNatarajan, C. M., Tanner, M. G. & Hadfield, R. H. · Supercond. Sci. Technol. 25, 063001 (2012)
- Superconducting properties and chemical composition of NbTiN thin films with different thicknessZhang, L., Peng, W., You, L. X. & Wang, Z. · Appl. Phys. Lett. 107, 122603 (2015)
Superconductor on semiconductor
Growing a superconductor against a semiconductor without ruining either is its own field, and the cleanest results are epitaxial.
- Epitaxy of semiconductor–superconductor nanowiresKrogstrup, P. et al. · Nat. Mater. 14, 400 (2015)
- Hard gap in epitaxial semiconductor–superconductor nanowiresChang, W. et al. · Nat. Nanotechnol. 10, 232 (2015)
- Two-dimensional epitaxial superconductor-semiconductor heterostructures: a platform for topological superconducting networksShabani, J. et al. · Phys. Rev. B 93, 155402 (2016)
Junctions
Two superconductors and the barrier between them: the working element of qubits and SQUIDs, and the place devices fail.
- Possible new effects in superconductive tunnellingJosephson, B. D. · Phys. Lett. 1, 251 (1962)
- Grain boundaries in high-Tc superconductorsHilgenkamp, H. & Mannhart, J. · Rev. Mod. Phys. 74, 485 (2002)
- Superconducting quantum bitsClarke, J. & Wilhelm, F. K. · Nature 453, 1031 (2008)
- Materials in superconducting quantum bitsOliver, W. D. & Welander, P. B. · MRS Bull. 38, 816 (2013)
Making known materials usable
The gap between a superconductor and a conductor anybody can wind is decades of applied work on wires, tapes and strain.
- High-Tc superconducting materials for electric power applicationsLarbalestier, D., Gurevich, A., Feldmann, D. M. & Polyanskii, A. · Nature 414, 368 (2001)
- A review of the properties of Nb₃Sn and their variation with A15 composition, morphology and strain stateGodeke, A. · Supercond. Sci. Technol. 19, R68 (2006)
- Materials science challenges for high-temperature superconducting wireFoltyn, S. R. et al. · Nat. Mater. 6, 631 (2007)
- Isotropic round-wire multifilament cuprate superconductor for generation of magnetic fields above 30 TLarbalestier, D. C. et al. · Nat. Mater. 13, 375 (2014)
Machine learning before us
Transition temperatures have been predicted from the SuperCon database before, and the large structure-discovery models are real. We cite the efforts this platform is most often compared to; the FAQ says where ours differs.
- Machine learning modeling of superconducting critical temperatureStanev, V. et al. · npj Comput. Mater. 4, 29 (2018)
- A data-driven statistical model for predicting the critical temperature of a superconductorHamidieh, K. · Comput. Mater. Sci. 154, 346 (2018)
- Scaling deep learning for materials discoveryMerchant, A. et al. · Nature 624, 80 (2023)
The closed testing environment
Validation runs in both directions. Forward, our number goes on the record before the instrument runs, so the prediction is committed while it can still be wrong. Backward, inside a closed environment: superconductors the field settled decades ago go in with part of what is known withheld from the platform's inputs, and the platform recovers in minutes what the literature took decades to establish. What comes back is scored against the published record and nothing else.
There is no shortlist. Every superconductor the published record has settled is in scope, the archive it draws on is more than a century deep, and the suite reruns in full each time the models improve. What it establishes is agreement with the published record, stated exactly that way on purpose: the stronger-sounding versions of that sentence are the ones this page refuses below.
How the models learn
The loop has a fixed shape. The published record above goes in, the closed environment scores what comes out against it, and every partner-lab result, hit or miss, retrains the next version. The shape is public and the contents are not: which variables the framework isolates, how the criterion ranks, and what the training set holds are the part we license, and they stay closed on purpose.
The number we will not print
Most platforms in this field quote an accuracy figure. We had one, and we retired it in July 2026 rather than defend something we could not stand behind. A single number in kelvin depends on which transition criterion the original measurement used, and those differ across the papers the record is drawn from. Quoted tightly across materials running from a few kelvin to 135 K, it reads as overfitting to the exact audience qualified to catch it.
So the claim is agreement with the published record. It is the weaker sentence and the honest one, and it is the one a physicist can go and verify this afternoon. The number comes back when there is a verified table to put behind it, and not one day sooner.
Independent validation
The effect the platform relies on has been measured independently at Stanford and SLAC and published in Physical Review Letters, by people who have never worked with us and owe us nothing. We did not commission the work and had no hand in it. One distinction stays sharp on purpose: an independent measurement of the physics underneath the platform is not an independent measurement of a candidate, and we cite it as exactly what it is.
The short version of all of this is in the FAQ