metric,pair,mean_a,mean_b,diff,ci_lo,ci_hi,p_raw,p_adj,cohen_d,n Mapped rate,Human vs R3,0.6683021605778869,0.789969712856616,-0.12166755227872879,-0.13909092875004883,-0.10413741706393437,2.8943764614848867e-33,8.104254092157683e-32,-0.55810451365856,1000 Mapped rate,Human vs Gemini 3 Pro,0.6683021605778869,0.7796178571428573,-0.11131569656497005,-0.12916637123091756,-0.09293583075116686,1.4875723318624722e-28,3.7189308296561806e-27,-0.4839287461850507,1000 Mapped rate,Human vs GPT-5.2,0.6680228149718063,0.6816628050343007,-0.013639990062494375,-0.030164630641992325,0.0032034765381447256,0.5544883403593641,1.0,-0.06311929979351662,999 Mapped rate,R3 vs Gemini 3 Pro,0.789969712856616,0.7796178571428573,0.010351855713758728,-0.0012260367565366374,0.021956185692265356,0.06043635751969429,0.4834908601575543,0.06216100237655318,1000 Mapped rate,R3 vs GPT-5.2,0.7898504724199449,0.6816628050343007,0.10818766738564428,0.09837913172217079,0.1179699929519917,2.8727875053861944e-74,1.2927543774237874e-72,0.7373702396342295,999 Mapped rate,Gemini 3 Pro vs GPT-5.2,0.779730921397588,0.6816628050343007,0.09806811636328744,0.08665241099398656,0.10959455228634166,9.945221732835001e-54,3.7791842584773005e-52,0.5977092269747867,999 Consequential comment rate,Human vs R3,0.6156468922284993,0.8853102754305852,-0.26966338320208577,-0.2928134322891657,-0.2460659279071373,9.764004864750085e-74,4.296162140490037e-72,-0.9989362967460053,972 Consequential comment rate,Human vs Gemini 3 Pro,0.6152510599856863,0.6544227028264104,-0.039171642840724066,-0.06669840549874498,-0.010926038367727927,0.02056687473173869,0.1851018725856482,-0.12293327733746252,971 Consequential comment rate,Human vs GPT-5.2,0.6153813302651675,0.7845665173602099,-0.16918518709504243,-0.1941219835473134,-0.1445278821701196,4.344048676295296e-27,9.991311955479181e-26,-0.6195663091316436,970 Consequential comment rate,R3 vs Gemini 3 Pro,0.8854799765826524,0.6554804052549544,0.22999957132769802,0.21331244001082722,0.2469824958791323,3.628764272007293e-98,1.8143821360036466e-96,1.1114030459316135,998 Consequential comment rate,R3 vs GPT-5.2,0.8856803773842556,0.7863157002975163,0.09936467708673928,0.08918718633432506,0.10954191251893844,5.743438069813574e-62,2.412243989321701e-60,0.7903081668783974,998 Consequential comment rate,Gemini 3 Pro vs GPT-5.2,0.6552906068719322,0.786263121382451,-0.1309725145105189,-0.14819949426075651,-0.11370849999979035,9.031044775587657e-39,2.7996238804321735e-37,-0.6211831922145935,996 Consequential rate (critical only),Human vs R3,0.8643085805951737,0.905097130348909,-0.04078854975373525,-0.05812694619596808,-0.02334638461090048,0.061701133682509235,0.4834908601575543,-0.20983917408087918,857 Consequential rate (critical only),Human vs Gemini 3 Pro,0.8642920065996019,0.843488114044986,0.02080389255461573,-0.0031995195974374325,0.043816576417011074,0.012022809143511773,0.12022809143511773,0.08477255670696454,844 Consequential rate (critical only),Human vs GPT-5.2,0.8646860713778921,0.8081448181776335,0.05654125320025866,0.037806755187126916,0.07473706395106261,6.585319263826029e-21,1.382917045403466e-19,0.28248730779787884,856 Consequential rate (critical only),R3 vs Gemini 3 Pro,0.9064916425612042,0.8427746909067277,0.06371695165447636,0.04791545337504436,0.08000159776545539,5.118989790769644e-11,8.19038366523143e-10,0.34663403925921327,977 Consequential rate (critical only),R3 vs GPT-5.2,0.904687820440794,0.8120030557971071,0.09268476464368695,0.08294214791173714,0.10278402819335627,4.884251347126829e-59,2.002543052322e-57,0.7626126456769157,996 Consequential rate (critical only),Gemini 3 Pro vs GPT-5.2,0.8431258944834764,0.8127308109732556,0.030395083510221018,0.014471552146741992,0.04590534490327664,2.935562985380456e-09,4.4033444780706843e-08,0.15944754891347546,976 Specification rate,Human vs R3,0.9915761365012722,1.0,-0.008423863498727898,-0.012306427050640169,-0.005140572248761796,7.709455705781912e-08,1.0793237988094676e-06,-0.20680071606830236,1000 Specification rate,Human vs Gemini 3 Pro,0.9915761365012722,1.0,-0.008423863498727898,-0.012169431588736068,-0.0051599793591972426,7.709455705781912e-08,1.0793237988094676e-06,-0.20680071606830236,1000 Specification rate,Human vs GPT-5.2,0.9915677042054777,0.9998522331855665,-0.008284528980088936,-0.01211248498876319,-0.0049484918290412195,1.5993425067501441e-07,1.919211008100173e-06,-0.2029060905366223,999 Specification rate,R3 vs Gemini 3 Pro,1.0,1.0,0.0,0.0,0.0,1.0,1.0,0.0,1000 Specification rate,R3 vs GPT-5.2,1.0,0.9998522331855665,0.00014776681443348113,0.0,0.00039563372896706233,0.17971249487899976,0.8568927558463574,0.05965739680604268,999 Specification rate,Gemini 3 Pro vs GPT-5.2,1.0,0.9998522331855665,0.00014776681443348113,0.0,0.00039563372896706233,0.17971249487899976,0.8568927558463574,0.05965739680604268,999 Actionability rate,Human vs R3,0.646582180366501,0.9990825528444722,-0.3525003724779715,-0.3753832441766178,-0.3300925004395797,2.2635804945186933e-106,1.1996976620949074e-104,-1.378330317152698,1000 Actionability rate,Human vs Gemini 3 Pro,0.646582180366501,0.889100396825397,-0.24251821645889599,-0.26708894913691217,-0.21847910610613921,1.1636978312310198e-52,4.189312192431671e-51,-0.920030428334636,1000 Actionability rate,Human vs GPT-5.2,0.6462810930384868,0.9860819769993352,-0.33980088396084857,-0.36277583996369334,-0.31715791154301903,1.4752835304656554e-110,8.114059417561105e-109,-1.3209914690272813,999 Actionability rate,R3 vs Gemini 3 Pro,0.9990825528444722,0.889100396825397,0.10998215601907552,0.10447994454848555,0.11571931151472364,9.840429058807976e-115,5.510640272932467e-113,1.7041416565139258,1000 Actionability rate,R3 vs GPT-5.2,0.9990816344789513,0.9860819769993352,0.012999657479616046,0.010767526457575765,0.015780961192071525,2.0252257932796184e-35,5.873154800510894e-34,0.4514102140147595,999 Actionability rate,Gemini 3 Pro vs GPT-5.2,0.8891562197117754,0.9860819769993352,-0.09692575728755999,-0.10288871519445736,-0.09093274674151992,3.204621609650523e-116,1.826634317500798e-114,-1.3831680364250827,999 Anchored rate,Human vs R3,0.5311958802050925,0.8279920570382864,-0.29679617683319404,-0.32109170322959996,-0.2719960868551521,4.002802584996131e-78,1.8412891890982204e-76,-1.0885053884541227,1000 Anchored rate,Human vs Gemini 3 Pro,0.5311958802050925,0.8518702380952382,-0.32067435789014576,-0.3429445055371348,-0.2985973612046225,4.099266655618538e-100,2.0906259943654542e-98,-1.2945158567377653,1000 Anchored rate,Human vs GPT-5.2,0.5311480809175889,0.7346987646995344,-0.20355068378194552,-0.2285676930167397,-0.1780573038312676,3.022111087618526e-45,9.972966589141136e-44,-0.7019962802659282,999 Anchored rate,R3 vs Gemini 3 Pro,0.8279920570382864,0.8518702380952382,-0.023878181056951758,-0.036578281919923274,-0.011119115809999707,0.08469460599452008,0.5081676359671206,-0.1529282653989846,1000 Anchored rate,R3 vs GPT-5.2,0.828365877461202,0.7346987646995344,0.09366711276166768,0.07567272066124271,0.11151138270670351,4.53857082077221e-22,9.984855805698862e-21,0.4321165643965153,999 Anchored rate,Gemini 3 Pro vs GPT-5.2,0.8518887935554602,0.7346987646995344,0.11719002885592582,0.10161349682746258,0.13298973210632048,5.056886443485539e-31,1.3147904753062402e-29,0.6349774981728731,999 Justification rate,Human vs R3,0.6553195311517952,0.9732309091657284,-0.3179113780139332,-0.33405623575379073,-0.30192164757379075,6.83807881233523e-133,4.102847287401138e-131,-1.7443561080753633,1000 Justification rate,Human vs Gemini 3 Pro,0.6553195311517952,0.8882753968253969,-0.23295586567360158,-0.2503843232777007,-0.21528021775546952,4.354603080429521e-94,2.1337555094104653e-92,-1.1651963742370848,1000 Justification rate,Human vs GPT-5.2,0.6552906112367161,0.82322216598221,-0.16793155474549382,-0.1858944294053439,-0.14969130329614255,3.757574031730488e-58,1.4654538723748903e-56,-0.7999845335796282,999 Justification rate,R3 vs Gemini 3 Pro,0.9732309091657284,0.8882753968253969,0.08495551234033163,0.07581150782443577,0.09460126812103611,1.655790200333321e-58,6.623160801333284e-57,0.779057448086605,1000 Justification rate,R3 vs GPT-5.2,0.9732496133245075,0.82322216598221,0.1500274473422975,0.13929016603387473,0.16082380976046592,1.4638750511433967e-104,7.612150265945662e-103,1.1876227453771218,999 Justification rate,Gemini 3 Pro vs GPT-5.2,0.8884972273861163,0.82322216598221,0.06527506140390629,0.052093136290176714,0.07827932240378535,7.936457832257501e-21,1.5872915664515002e-19,0.43328086352216316,999 Claim coverage,Human vs R3,0.5834785714285714,0.9025380952380954,-0.3190595238095238,-0.337262380952381,-0.3010397619047619,1.5682335245739594e-117,9.095754442528964e-116,-1.5011932022204146,1000 Claim coverage,Human vs Gemini 3 Pro,0.5834785714285714,0.722502380952381,-0.13902380952380952,-0.1579014285714286,-0.1203689880952381,1.0831363128614597e-39,3.466036201156671e-38,-0.6051503297862163,1000 Claim coverage,Human vs GPT-5.2,0.5834785714285714,0.9616976190476191,-0.3782190476190476,-0.3953222023809524,-0.36137827380952375,6.001828560089725e-134,3.6611154216547324e-132,-1.9006195309070357,1000 Claim coverage,R3 vs Gemini 3 Pro,0.9025380952380954,0.722502380952381,0.18003571428571427,0.16753529761904765,0.19264059523809524,1.5581104964868013e-88,7.478930383136646e-87,1.0539603138792843,1000 Claim coverage,R3 vs GPT-5.2,0.9025380952380954,0.9616976190476191,-0.05915952380952381,-0.06865750000000001,-0.04975952380952381,1.7218804632043693e-28,4.1325131116904866e-27,-0.4676804369712874,1000 Claim coverage,Gemini 3 Pro vs GPT-5.2,0.722502380952381,0.9616976190476191,-0.2391952380952381,-0.25123607142857146,-0.2271952380952381,1.5027029334352873e-122,8.865947307268195e-121,-1.5568005097805124,1000 Comments per paper,Human vs R3,9.839,16.805,-6.966,-7.517,-6.411,5.018539303742469e-88,2.3587134727589605e-86,-1.1493303001834283,1000 Comments per paper,Human vs Gemini 3 Pro,9.839,6.441,3.398,2.939,3.864,8.030802372740196e-33,2.168316640639853e-31,0.6420836105202683,1000 Comments per paper,Human vs GPT-5.2,9.839,24.116,-14.277,-14.791025000000001,-13.754,1.0694585099333283e-148,6.844534463573301e-147,-2.3713113588864076,1000 Comments per paper,R3 vs Gemini 3 Pro,16.805,6.441,10.364,10.08,10.644025,2.064808577283914e-164,1.3421255752345442e-162,3.3524298367463077,1000 Comments per paper,R3 vs GPT-5.2,16.805,24.116,-7.311,-7.708,-6.916975,1.1118182389404344e-135,6.893273081430693e-134,-1.732894654242219,1000 Comments per paper,Gemini 3 Pro vs GPT-5.2,6.441,24.116,-17.675,-17.925025,-17.42095,1.9249733738414185e-165,1.2704824267353363e-163,-5.868368834853515,1000 Comment length (words),Human vs R3,84.9991099273843,80.64536926242421,4.353740664960084,1.6389717143164293,7.193089299581717,0.17137855116927148,0.8568927558463574,0.139918373187585,1000 Comment length (words),Human vs Gemini 3 Pro,84.9991099273843,69.48124642857144,15.517863498812865,12.869048965527382,18.324646016561424,5.287829568821866e-18,1.0046876180761546e-16,0.49192950905370053,1000 Comment length (words),Human vs GPT-5.2,84.99073223857023,66.89045447568859,18.10027776288165,14.268643131521449,21.609478048322423,5.612080377684359e-36,1.6836241133053075e-34,0.43221762361225297,999 Comment length (words),R3 vs Gemini 3 Pro,80.64536926242421,69.48124642857144,11.164122833852781,10.409385931625351,11.940250142831236,1.397626423745587e-108,7.54718268822617e-107,1.0811520874531484,1000 Comment length (words),R3 vs GPT-5.2,80.65511528680193,66.89045447568859,13.76466081111334,10.800154706790869,15.53260444975026,5.934043093846348e-146,3.738447149123199e-144,0.4681782911406187,999 Comment length (words),Gemini 3 Pro vs GPT-5.2,69.47872515372516,66.89045447568859,2.588270678036573,-0.37318071292305294,4.484029959811818,4.8624823893328395e-15,8.752468300799111e-14,0.08669498107427998,999 Claim overlap fraction with human,Human vs R3,,,,,,,,,0 Claim overlap fraction with human,Human vs Gemini 3 Pro,,,,,,,,,0 Claim overlap fraction with human,Human vs GPT-5.2,,,,,,,,,0 Claim overlap fraction with human,R3 vs Gemini 3 Pro,0.9414266117969821,0.8004629629629629,0.1409636488340192,0.12438143004115228,0.1573233882030178,2.835974944871989e-46,9.642314812564763e-45,0.6773049197003487,972 Claim overlap fraction with human,R3 vs GPT-5.2,0.9414266117969821,0.9801611796982167,-0.038734567901234564,-0.048696844993141294,-0.02894332990397806,2.5851621032870807e-13,4.394775575588037e-12,-0.29992268234678643,972 Claim overlap fraction with human,Gemini 3 Pro vs GPT-5.2,0.8004629629629629,0.9801611796982167,-0.1796982167352538,-0.19603952331961594,-0.1640432098765432,1.6002007314991054e-67,6.880863145446153e-66,-0.9362544162590037,972 Stance-matched overlap fraction with human,Human vs R3,,,,,,,,,0 Stance-matched overlap fraction with human,Human vs Gemini 3 Pro,,,,,,,,,0 Stance-matched overlap fraction with human,Human vs GPT-5.2,,,,,,,,,0 Stance-matched overlap fraction with human,R3 vs Gemini 3 Pro,0.7024519890260631,0.5271947873799725,0.17525720164609052,0.15504115226337448,0.19562757201646092,1.1431996273207464e-48,4.0011986956226122e-47,0.47391208308718885,972 Stance-matched overlap fraction with human,R3 vs GPT-5.2,0.7024519890260631,0.7163065843621399,-0.01385459533607682,-0.022428412208504795,-0.005264060356652949,0.0006049773744351455,0.0066547511187866,-0.03642960857672607,972 Stance-matched overlap fraction with human,Gemini 3 Pro vs GPT-5.2,0.5271947873799725,0.7163065843621399,-0.18911179698216735,-0.2093115569272977,-0.16892961248285324,1.7781268743159062e-53,6.5790694349688526e-52,-0.513189178140074,972 Stance agreement (critical),Gemini 3 Pro vs GPT-5.2,0.8613110539845759,0.9954798628963154,-0.13416880891173955,,,,1.0428161880016923e-32,-0.6820968987257782,778 Stance agreement (critical),R3 vs Gemini 3 Pro,0.995446735395189,0.864819587628866,0.130627147766323,,,,1.489357045279712e-30,0.6700674973965489,776 Stance agreement (critical),R3 vs GPT-5.2,0.9956270627062707,0.9944100660066006,0.0012169966996700143,,,,0.7071963616423718,0.023674172308075322,808