Non-observations are not properties: measurement floors in superconductor databases and their propagation into machine-learning datasets

{"Superconductor":[0],"databases":[1],"record":[2],"two":[3,62,78,338,354,480,591,600,726,754],"different":[4],"kinds":[5],"of":[6,77,126,186,222,226,235,244,271,287,335,401,414,442,475,508,528,549,556,581,634,641,645,682,752,757,786,790,809,822,850,864,882,893,927,984,1015,1028],"statement":[7,32],"in":[8,66,248,265,267,309,384,398,501,543,560,587,771,783,813,910,954,1041],"the":[9,17,53,72,75,106,123,134,141,149,219,232,236,245,254,272,325,330,336,343,378,389,417,428,440,466,482,493,496,499,509,517,526,529,540,558,565,575,604,608,627,632,639,655,664,668,683,687,692,697,710,734,749,753,772,793,801,810,814,829,834,851,904,924,937,966,975,982,1003,1006,1012,1020,1029,1048,1057,1083],"same":[10,237,390,676],"field:":[11],"a":[12,31,38,67,90,161,164,197,258,362,443,449,506,585,672,805,819,838,919,955,985,1036],"measured":[13,190,260,349,871],"transition":[14,21,872],"temperature,":[15,1000],"and":[16,88,117,121,139,159,194,204,278,306,317,366,373,434,448,513,571,593,621,654,667,723,733,748,765,788,798,833,840,915,918,946,1023,1039,1066,1087,1096],"observation":[18],"that":[19,59,167,358,388,626,648,680,699,736,854,928,998,1067],"no":[20,205,845,870,1090],"was":[22,546],"found":[23,490,953,1038],"down":[24],"to":[25,133,411,427,486,521,936,1074],"some":[26],"temperature.":[27],"The":[28,152,675,889],"second":[29,255],"is":[30,55,531,579,594,615,657,804,903,996,1055,1071],"about":[33,37,563,1005],"an":[34,661],"experiment,":[35],"not":[36,171,211,347,360,375,554,568,583,651,670,775,991,1010,1056],"material.":[39],"Widely":[40],"used":[41],"derived":[42,70,168],"datasets":[43,169],"encode":[44],"it":[45,989],"as":[46,183,281,294,404,446,478,660,671],"Tc":[47,858],"=":[48,859],"0,":[49],"at":[50,199,298,303,319,437,533,847,857,876,884,912,921,931],"which":[51,268,370,988,1001,1054,1082],"point":[52,256,1004],"distinction":[54],"lost.":[56],"We":[57,252,352],"show":[58],"this":[60,385,941,1042],"has":[61,154,346,572,740],"separate":[63],"consequences.":[64],"First,":[65],"machine-learning":[68,418,483],"benchmark":[69],"from":[71,105,377,409,505,516,611,663],"SuperCon":[73,175],"database,":[74],"combination":[76],"individually":[79],"reasonable":[80],"cleaning":[81,511,938],"rules":[82],"—":[83,101,296,322,461,861,868,992],"averaging":[84],"over":[85,696],"duplicate":[86],"records,":[87,473],"discarding":[89],"compound":[91],"when":[92],"its":[93,249,619,642,706,865,993,1051,1063],"records":[94,397,685],"scatter":[95],"by":[96,261,465,491,743],"more":[97,1093],"than":[98,1078,1098],"5":[99],"K":[100,203,300,308,321,816,917],"removes":[102],"fifteen":[103,138,396,472,830],"materials":[104,812],"dataset":[107],"entirely,":[108],"among":[109],"them":[110,187,402,415,476,875,883],"Nb3Al":[111],"(17.5":[112],"K),":[113,120],"V3Si":[114],"(16.8":[115],"K)":[116],"NbN":[118],"(13.5":[119],"shifts":[122],"target":[124],"value":[125],"nineteen":[127,142],"further":[128,962],"entries.":[129],"A":[130,380],"nine-line":[131],"change":[132],"aggregation":[135,846],"restores":[136],"all":[137,323,489],"corrects":[140],"without":[143],"removing":[144],"any":[145],"entry":[146,206],"or":[147,200,885],"altering":[148],"non-superconductor":[150,250,866],"class.":[151,251,942],"patch":[153,530,935],"been":[155,292,341,348,469,573,741],"submitted":[156],"upstream.":[157],"Second,":[158],"independently,":[160],"non-observation":[162,444],"carries":[163,392,705,856],"measurement":[165,327,879,1008],"floor":[166,198,1064],"do":[170,372,374],"preserve.":[172],"In":[173],"MDR":[174],"Ver.240322":[176],"we":[177,367],"identify":[178],"3,130":[179],"formulas":[180,228],"listed":[181],"exclusively":[182],"non-superconducting;":[184],"1,749":[185],"were":[188,276,431,768],"never":[189],"below":[191,218,324,981],"1.5":[192],"K,":[193,439,553],"653":[195],"have":[196,290,339,468,869],"above":[201,283,551,886,997],"1":[202,887,959],"after":[207],"1980.":[208,351],"These":[209],"are":[210,215,631],"established":[212],"non-superconductors;":[213,525],"they":[214,240,1099],"compounds":[216,270,423],"untested":[217],"cryogenic":[220],"limit":[221],"their":[223],"time.":[224],"1,332":[225],"those":[227,288],"also":[229],"reach":[230,416,481,983],"3DSC_MP,":[231],"structure-matched":[233],"variant":[234],"dataset,":[238],"where":[239,603,969,1033],"account":[241],"for":[242,978,1062,1094],"1,334":[243,849],"1,778":[246,852],"entries":[247,832,853],"illustrate":[253],"with":[257,577,618,792],"series":[259,892],"Ku":[262],"et":[263,613,895],"al.":[264,614],"1980,":[266],"31":[269],"CeCo3B2":[273],"structure":[274],"type":[275],"characterised":[277],"23":[279],"recorded":[280],"non-superconducting":[282],"1.2":[284],"K.":[285,888,933],"Three":[286],"twenty-three":[289],"since":[291,314,350],"confirmed":[293],"superconductors":[295],"CeRu3B2":[297],"1.1":[299],"(1985),":[301],"YRu3B2":[302],"0.63,":[304],"0.7":[305],"0.81":[307],"three":[310,331,420,862,907],"independent":[311],"measurements":[312,602,926],"reported":[313],"December":[315],"2025,":[316],"LuRu3B2":[318],"0.95":[320],"1980":[326],"floor.":[328],"Of":[329],"non-magnetic":[332],"rare-earth":[333],"members":[334,908],"series,":[337],"now":[340,514,595,616,658,678,704,779,1046],"confirmed;":[342],"third,":[344],"La0.9Ru3B2,":[345],"report":[353],"pre-registered":[355,755,820],"permutation":[356],"tests":[357,756],"did":[359],"support":[361],"broader":[363],"frequency":[364],"claim,":[365],"state":[368,700],"explicitly":[369],"conclusions":[371],"follow":[376],"data.":[379],"related":[381],"check,":[382],"added":[383],"version,":[386],"finds":[387],"database":[391],"nine":[393,462,719],"retracted":[394,450,459,720],"papers,":[395],"all,":[399,877],"none":[400,474,538],"marked":[403,477],"retracted;":[405],"retraction":[406,429],"dates":[407],"range":[408,666,693],"1992":[410],"2023.":[412],"Two":[413],"benchmark:":[419],"Rh-based":[421],"Heusler":[422],"whose":[424,760],"samples,":[425],"according":[426],"notice,":[430],"palladium":[432],"compounds,":[433],"elemental":[435],"europium":[436],"1.375":[438],"average":[441],"encoded":[445],"zero":[447],"high-pressure":[451],"value.":[452],"Version":[453],"3.":[454],"New:":[455],"Section":[456,544,561,588,597,623,758,777,826,958,972,1044,1059,1088],"7.3":[457],"on":[458,638],"sources":[460],"papers":[463],"cited":[464,617],"datasheet":[467],"retracted,":[470],"carrying":[471],"such;":[479],"benchmark.":[484],"Corrections":[485],"version":[487],"2,":[488],"re-running":[492],"analyses":[494],"against":[495,818],"primary":[497],"files:":[498,716,944],"totals":[500],"Table":[502],"4.1":[503],"came":[504],"re-implementation":[507],"3DSC":[510],"step":[512,939],"come":[515],"pipeline":[518,727],"itself":[519],"(15,788":[520],"15,803":[522],"entries,":[523],"3,866":[524],"effect":[527],"unchanged":[532],"+15":[534],"restored,":[535],"19":[536],"corrected,":[537],"lost);":[539],"stratum":[541],"size":[542],"5.3":[545,778,1045],"wrong":[547],"(0":[548],"1,371":[550],"10":[552],"0":[555,860],"5,145);":[557],"claim":[559],"2.2":[562],"reproducing":[564],"normalisation":[566],"could":[567,650],"be":[569,652],"reproduced":[570],"removed;":[574],"comparison":[576],"SuperCon-MTG":[578,731],"139":[580,835],"139,":[582],"137;":[584],"quotation":[586],"2.1":[589],"conflated":[590],"sentences":[592],"verbatim;":[596],"7.4":[598],"said":[599,960,974],"conflicting":[601],"text":[605,649,773],"gives":[606],"three;":[607],"footnote":[609],"quoted":[610],"Savitskii":[612,963],"page":[620],"table.":[622],"5.2":[624],"asserted":[625],"numeric":[628],"reference":[629,708,949],"identifiers":[630],"bibliography":[633],"Roberts'":[635],"1976":[636],"compilation,":[637],"strength":[640],"stated":[643],"count":[644],"1,995":[646],"entries;":[647],"obtained":[653],"identification":[656],"given":[659],"inference":[662],"identifier":[665],"years,":[669],"verified":[673],"fact.":[674],"section":[677],"states":[679,780,1047],"151":[681],"4,241":[684],"leave":[686],"year":[688],"field":[689],"empty,":[690],"so":[691,1024],"1913–1975":[694],"holds":[695],"4,090":[698],"one.":[701],"Reference":[702],"[4]":[703],"DOI,":[707],"[12]":[709],"full":[711,784],"author":[712],"list.":[713],"New":[714,825,943],"supplementary":[715],"S5":[717],"(the":[718,725,730],"sources),":[721],"S6a":[722],"S6b":[724],"outputs),":[728],"S7":[729],"comparison),":[732],"scripts":[735,751],"produce":[737],"them;":[738],"verify_patch.py":[739],"replaced":[742],"compare_cleaned_outputs.py.":[744],"Also":[745],"new:":[746],"S8a–S8e":[747],"four":[750,961],"5.3,":[759],"counting":[761],"stages,":[762],"result":[763],"protocols":[764],"pre-registration":[766],"checksums":[767],"previously":[769],"described":[770],"but":[774],"deposited.":[776],"both":[781],"populations":[782],"(5,622":[785],"5,644":[787],"10,072":[789],"10,098,":[791],"exclusion":[794],"rule":[795],"spelled":[796],"out)":[797],"names":[799],"what":[800],"2.46":[802],"%":[803],"share":[806],"of:":[807],"11":[808],"448":[811],"0–1":[815],"stratum,":[817],"threshold":[821],"8":[823,1089],"%.":[824],"4.4:":[827],"besides":[828],"removed":[831],"averaged":[836],"ones,":[837],"third":[839],"much":[841],"larger":[842],"class":[843,867],"needs":[844],"all.":[848],"3DSC_MP":[855,911],"quarters":[863],"temperature":[873],"behind":[874],"only":[878,1026],"floors,":[880],"948":[881],"Nb-substituted":[890],"ZrNi2Ga":[891],"Winterlik":[894],"al.,":[896],"Phys.":[897],"Rev.":[898],"B":[899],"78,":[900],"184506":[901],"(2008),":[902],"worked":[905],"example:":[906],"stand":[909],"2.3,":[913],"2.4":[914],"2.87":[916],"fourth":[920],"zero,":[922],"because":[923],"magnetisation":[925],"paper":[929],"stopped":[930],"1.8":[932],"No":[934],"reaches":[940],"S9":[945],"count_nonobservations_in_3dsc.py;":[947],"new":[948],"[21].":[950],"Further":[951],"corrections":[952],"final":[956],"review:":[957],"tables":[964],"carry":[965],"measurement-floor":[967],"convention":[968,1050,1085],"six":[970],"do;":[971],"6.3":[973],"predicted":[976],"band":[977],"LaRu3B2":[979],"falls":[980],"3He":[986],"system,":[987],"does":[990],"lower":[994],"bound":[995],"base":[999],"makes":[1002],"missing":[1007],"stronger,":[1009],"weaker;":[1011],"citation-string":[1013],"parser":[1014],"match_retractions.py":[1016],"required":[1017],"whitespace":[1018],"before":[1019],"volume":[1021],"number":[1022],"parsed":[1025],"4,075":[1027],"7,254":[1030],"citation":[1031],"strings":[1032],"7,016":[1034],"parse,":[1035],"defect":[1037],"corrected":[1040],"review;":[1043],"bin":[1049],"stratification":[1052],"uses,":[1053],"one":[1058],"5.1":[1060],"declares":[1061],"bins,":[1065],"table's":[1068],"first":[1069],"row":[1070],"labelled":[1072],"\\"up":[1073],"0.1":[1075,1080],"K\\"":[1076],"rather":[1077],"\\"below":[1079],"K\\",":[1081],"declared":[1084],"requires;":[1086],"longer":[1091],"claims":[1092],"check_supercon_mtg.py":[1095],"reproduce_counts.py":[1097],"compute.":[1100]}

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-15
DOI
https://doi.org/10.5281/zenodo.22772787
Primary Topic
Machine Learning in Materials Science
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

Non-observations are not properties: measurement floors in superconductor databases and their propagation into machine-learning datasets

Daniel Leonforte
Zenodo (CERN European Organization for Nuclear Research)
Machine Learning in Materials Science
preprint

Non-observations are not properties: measurement floors in superconductor databases and their propagation into machine-learning datasets

Daniel Leonforte
preprint en

Abstract

Superconductor databases record two different kinds of statement in the same field: a measured transition temperature, and the observation that no transition was found down to some temperature. The second is a statement about an experiment, not about a material. Widely used derived datasets encode it as Tc = 0, at which point the distinction is lost. We show that this has two separate consequences. First, in a machine-learning benchmark derived from the SuperCon database, the combination of two individually reasonable cleaning rules — averaging over duplicate records, and discarding a compound when its records scatter by more than 5 K — removes fifteen materials from the dataset entirely, among them Nb3Al (17.5 K), V3Si (16.8 K) and NbN (13.5 K), and shifts the target value of nineteen further entries. A nine-line change to the aggregation restores all fifteen and corrects the nineteen without removing any entry or altering the non-superconductor class. The patch has been submitted upstream. Second, and independently, a non-observation carries a measurement floor that derived datasets do not preserve. In MDR SuperCon Ver.240322 we identify 3,130 formulas listed exclusively as non-superconducting; 1,749 of them were never measured below 1.5 K, and 653 have a floor at or above 1 K and no entry after 1980. These are not established non-superconductors; they are compounds untested below the cryogenic limit of their time. 1,332 of those formulas also reach 3DSC_MP, the structure-matched variant of the same dataset, where they account for 1,334 of the 1,778 entries in its non-superconductor class. We illustrate the second point with a series measured by Ku et al. in 1980, in which 31 compounds of the CeCo3B2 structure type were characterised and 23 recorded as non-superconducting above 1.2 K. Three of those twenty-three have since been confirmed as superconductors — CeRu3B2 at 1.1 K (1985), YRu3B2 at 0.63, 0.7 and 0.81 K in three independent measurements reported since December 2025, and LuRu3B2 at 0.95 K — all below the 1980 measurement floor. Of the three non-magnetic rare-earth members of the series, two have now been confirmed; the third, La0.9Ru3B2, has not been measured since 1980. We report two pre-registered permutation tests that did not support a broader frequency claim, and we state explicitly which conclusions do and do not follow from the data. A related check, added in this version, finds that the same database carries nine retracted papers, fifteen records in all, none of them marked as retracted; retraction dates range from 1992 to 2023. Two of them reach the machine-learning benchmark: three Rh-based Heusler compounds whose samples, according to the retraction notice, were palladium compounds, and elemental europium at 1.375 K, the average of a non-observation encoded as zero and a retracted high-pressure value. Version 3. New: Section 7.3 on retracted sources — nine papers cited by the datasheet have been retracted, carrying fifteen records, none of them marked as such; two reach the machine-learning benchmark. Corrections to version 2, all found by re-running the analyses against the primary files: the totals in Table 4.1 came from a re-implementation of the 3DSC cleaning step and now come from the pipeline itself (15,788 to 15,803 entries, 3,866 non-superconductors; the effect of the patch is unchanged at +15 restored, 19 corrected, none lost); the stratum size in Section 5.3 was wrong (0 of 1,371 above 10 K, not 0 of 5,145); the claim in Section 2.2 about reproducing the normalisation could not be reproduced and has been removed; the comparison with SuperCon-MTG is 139 of 139, not 137; a quotation in Section 2.1 conflated two sentences and is now verbatim; Section 7.4 said two conflicting measurements where the text gives three; the footnote quoted from Savitskii et al. is now cited with its page and table. Section 5.2 asserted that the numeric reference identifiers are the bibliography of Roberts' 1976 compilation, on the strength of its stated count of 1,995 entries; that text could not be obtained and the identification is now given as an inference from the identifier range and the years, not as a verified fact. The same section now states that 151 of the 4,241 records leave the year field empty, so the range 1913–1975 holds over the 4,090 that state one. Reference [4] now carries its DOI, reference [12] the full author list. New supplementary files: S5 (the nine retracted sources), S6a and S6b (the two pipeline outputs), S7 (the SuperCon-MTG comparison), and the scripts that produce them; verify_patch.py has been replaced by compare_cleaned_outputs.py. Also new: S8a–S8e and the four scripts of the two pre-registered tests of Section 5.3, whose counting stages, result protocols and pre-registration checksums were previously described in the text but not deposited. Section 5.3 now states both populations in full (5,622 of 5,644 and 10,072 of 10,098, with the exclusion rule spelled out) and names what the 2.46 % is a share of: 11 of the 448 materials in the 0–1 K stratum, against a pre-registered threshold of 8 %. New Section 4.4: besides the fifteen removed entries and the 139 averaged ones, a third and much larger class needs no aggregation at all. 1,334 of the 1,778 entries that 3DSC_MP carries at Tc = 0 — three quarters of its non-superconductor class — have no measured transition temperature behind them at all, only measurement floors, 948 of them at or above 1 K. The Nb-substituted ZrNi2Ga series of Winterlik et al., Phys. Rev. B 78, 184506 (2008), is the worked example: three members stand in 3DSC_MP at 2.3, 2.4 and 2.87 K and a fourth at zero, because the magnetisation measurements of that paper stopped at 1.8 K. No patch to the cleaning step reaches this class. New files: S9 and count_nonobservations_in_3dsc.py; new reference [21]. Further corrections found in a final review: Section 1 said four further Savitskii tables carry the measurement-floor convention where six do; Section 6.3 said the predicted band for LaRu3B2 falls below the reach of a 3He system, which it does not — its lower bound is above that base temperature, which makes the point about the missing measurement stronger, not weaker; the citation-string parser of match_retractions.py required whitespace before the volume number and so parsed only 4,075 of the 7,254 citation strings where 7,016 parse, a defect found and corrected in this review; Section 5.3 now states the bin convention its stratification uses, which is not the one Section 5.1 declares for its floor bins, and that table's first row is labelled "up to 0.1 K" rather than "below 0.1 K", which the declared convention requires; and Section 8 no longer claims more for check_supercon_mtg.py and reproduce_counts.py than they compute.

Zenodo (CERN European Organization for Nuclear Research)
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.