metric,pair,mean_a,mean_b,diff,ci_lo,ci_hi,p_raw,p_adj,cohen_d,n Mapped rate,Human vs R3,0.4478595298664756,0.6348352593894342,-0.18697572952295863,-0.19985940580538808,-0.17401066111484312,2.1378438003594844e-105,8.337590821401989e-104,-0.9556672998507099,1000 Mapped rate,Human vs Gemini 3 Pro,0.4478595298664756,0.6626869047619047,-0.21482737489542916,-0.22930368302543042,-0.2010299732019867,4.5399049601144354e-111,1.906760083248063e-109,-1.008307560082463,1000 Mapped rate,Human vs GPT-5.2,0.4478595298664756,0.5258483149695556,-0.07798878510308004,-0.08960609420539357,-0.06617698558925941,1.3009283761454538e-34,3.252320940363635e-33,-0.40717364971026276,1000 Mapped rate,R3 vs Gemini 3 Pro,0.6348352593894342,0.6626869047619047,-0.027851645372470517,-0.042156750890444745,-0.013909670736871833,9.4573756684417e-05,0.000851163810159753,-0.13252480220045362,1000 Mapped rate,R3 vs GPT-5.2,0.6348352593894342,0.5258483149695556,0.1089869444198786,0.09741807385367897,0.1207096787936003,1.091781879051637e-59,2.9478110734394196e-58,0.5787614220794226,1000 Mapped rate,Gemini 3 Pro vs GPT-5.2,0.6626869047619047,0.5258483149695556,0.1368385897923491,0.12356962451189192,0.1499226516272928,9.720210322450509e-70,3.207669406408668e-68,0.6631745501574902,1000 Consequential comment rate,Human vs R3,0.5994751526329508,0.9322189576818236,-0.33274380504887274,-0.3451615869828179,-0.32048111157960546,3.1556486183833327e-156,1.640937281559333e-154,-2.1575539024061468,995 Consequential comment rate,Human vs Gemini 3 Pro,0.5994896027829588,0.7754608294930876,-0.17597122671012871,-0.19361915144081016,-0.15849497023332143,1.1496742565173383e-69,3.6789576208554825e-68,-0.7807115296355536,992 Consequential comment rate,Human vs GPT-5.2,0.5996506809555193,0.7362318458820882,-0.13658116492656897,-0.15066034193399408,-0.12246829198888104,1.2019437657614816e-62,3.4856369207082966e-61,-0.7971727085594834,994 Consequential comment rate,R3 vs Gemini 3 Pro,0.9322189576818236,0.7749653027039962,0.1572536549778274,0.141496538570916,0.17303847389449103,1.5776866780151145e-69,4.890828701846855e-68,0.790987699697854,995 Consequential comment rate,R3 vs GPT-5.2,0.9323549276764438,0.7369423648183335,0.19541256285811023,0.18481707730484495,0.20601499667409465,4.9373427360383975e-129,2.271177658577663e-127,1.4662321685369686,997 Consequential comment rate,Gemini 3 Pro vs GPT-5.2,0.7747389096483664,0.7363148112065021,0.03842409844186421,0.022440913497820375,0.05449677563033706,2.514705839706336e-09,3.2691175916182365e-08,0.18078690072252057,994 Consequential rate (critical only),Human vs R3,0.9196540183311104,0.9477721440366683,-0.02811812570555783,-0.038030145345521564,-0.018375179721388545,5.385657418252093e-07,6.4627889019025115e-06,-0.2332870751395065,986 Consequential rate (critical only),Human vs Gemini 3 Pro,0.9200222350162989,0.9633749133749133,-0.04335267835861464,-0.054635194769110045,-0.03211933719741268,3.2608914792783225e-16,5.217426366845316e-15,-0.3350263597164757,962 Consequential rate (critical only),Human vs GPT-5.2,0.9195591778467513,0.9055044958228549,0.01405468202389654,0.0033443284872832317,0.02477864184136419,0.000582263338924317,0.004658106711394536,0.10587043250473999,985 Consequential rate (critical only),R3 vs Gemini 3 Pro,0.947919317329627,0.9637517146776405,-0.015832397348013703,-0.025267148629069335,-0.006335085671682516,8.167407439557826e-06,8.167407439557826e-05,-0.14104333775988706,972 Consequential rate (critical only),R3 vs GPT-5.2,0.9479722507724724,0.9058668015173368,0.04210544925513563,0.03317535406228254,0.051327453519804105,7.592320226836285e-19,1.2906944385621685e-17,0.3643207754613471,997 Consequential rate (critical only),Gemini 3 Pro vs GPT-5.2,0.9637143837967731,0.9068987601053344,0.05681562369143874,0.04623821148635581,0.06719129065210513,3.8102035029210386e-29,8.382447706426285e-28,0.4539272250432056,971 Specification rate,Human vs R3,0.996540829009431,1.0,-0.003459170990569051,-0.0045709675072271114,-0.002453716655645801,1.6203501809875324e-10,2.4305252714812986e-09,-0.2844975857878191,1000 Specification rate,Human vs Gemini 3 Pro,0.996540829009431,1.0,-0.003459170990569051,-0.004582674389278589,-0.00245944883016532,1.6203501809875324e-10,2.4305252714812986e-09,-0.2844975857878191,1000 Specification rate,Human vs GPT-5.2,0.996540829009431,0.999340750550619,-0.0027999215411880227,-0.003989608341180775,-0.0016741077597711058,5.262490202550481e-06,5.788739222805529e-05,-0.20859428279552164,1000 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-5.2,1.0,0.999340750550619,0.0006592494493810284,0.00021738437001594892,0.0012148402492974859,0.011718685599768632,0.0703121135986118,0.11593724218090665,1000 Specification rate,Gemini 3 Pro vs GPT-5.2,1.0,0.999340750550619,0.0006592494493810284,0.0002153822055137845,0.0012058123765854027,0.011718685599768632,0.0703121135986118,0.11593724218090665,1000 Actionability rate,Human vs R3,0.5771398031508598,0.9921177206150739,-0.4149779174642141,-0.42407439874321307,-0.405985421023306,3.369583159783153e-165,2.1228373906633865e-163,-3.981893830415763,1000 Actionability rate,Human vs Gemini 3 Pro,0.5771398031508598,0.8142694444444445,-0.23712964129358471,-0.25211183839087126,-0.2221281675219664,3.0223205657098504e-114,1.2995978432552357e-112,-1.373769329510085,1000 Actionability rate,Human vs GPT-5.2,0.5771398031508598,0.8256716974876488,-0.24853189433678893,-0.2580781311561694,-0.2390172988087007,2.0057703933662658e-162,1.1232314202851088e-160,-2.2732049469720397,1000 Actionability rate,R3 vs Gemini 3 Pro,0.9921177206150739,0.8142694444444445,0.17784827617062945,0.16573427714545144,0.19046064040009808,1.495969169312463e-110,6.133473594181099e-109,1.2577858290215869,1000 Actionability rate,R3 vs GPT-5.2,0.9921177206150739,0.8256716974876488,0.1664460231274252,0.16238695170428957,0.17054934263838847,1.6232153429806e-161,8.9276843863933e-160,3.588332139415591,1000 Actionability rate,Gemini 3 Pro vs GPT-5.2,0.8142694444444445,0.8256716974876488,-0.011402253043204226,-0.023896037819451578,0.0009581833751862021,0.020769780461854608,0.08307912184741843,-0.07852309842762296,1000 Anchored rate,Human vs R3,0.4261646472194076,0.8004195432015225,-0.37425489598211503,-0.38797143337698564,-0.36011675108530367,2.160844128561123e-154,1.1020305055661727e-152,-2.294866182959001,1000 Anchored rate,Human vs Gemini 3 Pro,0.4261646472194076,0.8730035714285715,-0.4468389242091639,-0.4598702622940942,-0.43388044730568565,1.3947428293371393e-163,8.228982693089122e-162,-2.9100474769927733,1000 Anchored rate,Human vs GPT-5.2,0.4261646472194076,0.8760668272588293,-0.44990218003942184,-0.4612256089199175,-0.4384194656018891,6.208563553349054e-165,3.787223767542923e-163,-3.4730380685034254,1000 Anchored rate,R3 vs Gemini 3 Pro,0.8004195432015225,0.8730035714285715,-0.07258402822704883,-0.08476605823317923,-0.060786822605416915,6.108810354579414e-27,1.2217620709158828e-25,-0.4900935505608018,1000 Anchored rate,R3 vs GPT-5.2,0.8004195432015225,0.8760668272588293,-0.07564728405730677,-0.08603921614918981,-0.06511440057304539,7.621422164509914e-36,1.9815697627725777e-34,-0.6148444757757505,1000 Anchored rate,Gemini 3 Pro vs GPT-5.2,0.8730035714285715,0.8760668272588293,-0.003063255830257926,-0.012858277785290192,0.006533335470309695,0.3745163071384561,1.0,-0.02782588386575726,1000 Justification rate,Human vs R3,0.5029546000758373,0.9643739169953078,-0.4614193169194707,-0.47123373350406506,-0.45155440243956446,7.174190211092162e-165,4.3045141266552976e-163,-4.056924882492773,1000 Justification rate,Human vs Gemini 3 Pro,0.5029546000758373,0.9114738095238096,-0.4085192094479723,-0.41986746359384186,-0.3972167030760672,1.637852672697918e-163,9.499545501647923e-162,-3.098999195904989,1000 Justification rate,Human vs GPT-5.2,0.5029546000758373,0.8064534598786806,-0.30349885980284336,-0.31532512570325166,-0.2912503927822752,3.287766976588195e-153,1.6438834882940974e-151,-2.1583019792091567,1000 Justification rate,R3 vs Gemini 3 Pro,0.9643739169953078,0.9114738095238096,0.0529001074714983,0.04441165128802357,0.061528489378757184,9.765223179249398e-32,2.3436535630198554e-30,0.5191118067685959,1000 Justification rate,R3 vs GPT-5.2,0.9643739169953078,0.8064534598786806,0.15792045711662728,0.14807393893260312,0.1678294493042438,1.3186450383849281e-126,5.933902672732177e-125,1.3968629913209218,1000 Justification rate,Gemini 3 Pro vs GPT-5.2,0.9114738095238096,0.8064534598786806,0.10502034964512902,0.09405076570611012,0.11611438271583141,7.275360586768245e-61,2.0371009642951086e-59,0.8002496826091958,1000 Claim coverage,Human vs R3,0.764702380952381,0.7748404761904762,-0.01013809523809524,-0.026169166666666663,0.005781011904761903,0.4607852102304355,1.0,-0.049896208550589474,1000 Claim coverage,Human vs Gemini 3 Pro,0.764702380952381,0.5622785714285715,0.20242380952380956,0.18715696428571432,0.21804779761904763,5.129797404921436e-85,1.8980250398209315e-83,0.9651582547112897,1000 Claim coverage,Human vs GPT-5.2,0.764702380952381,0.8537833333333334,-0.08908095238095239,-0.10367392857142857,-0.07442089285714287,5.51983572378717e-30,1.269562216471049e-28,-0.4672570035299487,1000 Claim coverage,R3 vs Gemini 3 Pro,0.7748404761904762,0.5622785714285715,0.21256190476190476,0.1977779166666667,0.2272884523809524,1.7194902161156484e-96,6.534062821239464e-95,1.0800505795644741,1000 Claim coverage,R3 vs GPT-5.2,0.7748404761904762,0.8537833333333334,-0.07894285714285713,-0.09230720238095239,-0.0654920238095238,1.3387030984107446e-27,2.811276506662564e-26,-0.4477009037971251,1000 Claim coverage,Gemini 3 Pro vs GPT-5.2,0.5622785714285715,0.8537833333333334,-0.29150476190476193,-0.30453583333333334,-0.2782617261904762,1.9033508055301536e-133,8.945748785991722e-132,-1.5856789020666318,1000 Comments per paper,Human vs R3,34.962,16.284,18.678,17.998974999999998,19.377025,1.3738112479525345e-162,7.830724113329447e-161,2.3663853908179022,1000 Comments per paper,Human vs Gemini 3 Pro,34.962,5.899,29.063,28.434974999999998,29.709,2.9982362019475756e-165,1.9188711692464484e-163,3.971102320801038,1000 Comments per paper,Human vs GPT-5.2,34.962,21.709,13.253,12.595,13.922025,7.265928374750268e-150,3.560304903627631e-148,1.736097570155535,1000 Comments per paper,R3 vs Gemini 3 Pro,16.284,5.899,10.385,10.11,10.664,2.106288699442109e-165,1.3690876546373709e-163,3.2999071689996358,1000 Comments per paper,R3 vs GPT-5.2,16.284,21.709,-5.425,-5.747,-5.097,6.003723807479436e-118,2.641638475290952e-116,-1.4189883189191697,1000 Comments per paper,Gemini 3 Pro vs GPT-5.2,5.899,21.709,-15.81,-16.018,-15.603975,1.4731628288148196e-165,9.72287467017781e-164,-6.520800593992925,1000 Comment length (words),Human vs R3,39.660525339198024,78.2965078015314,-38.63598246233339,-39.37859812674344,-37.88635800030314,4.65493809022252e-165,2.886061615937963e-163,-4.3969777987603065,1000 Comment length (words),Human vs Gemini 3 Pro,39.660525339198024,68.76081309523809,-29.100287756040068,-30.037897376919496,-28.125442328486706,1.1977674628468024e-160,6.348167553088053e-159,-2.609385005920691,1000 Comment length (words),Human vs GPT-5.2,39.660525339198024,64.18198102683631,-24.521455687638284,-25.31913922127746,-23.728004690798425,8.126524774824038e-161,4.3883233784049805e-159,-2.5878387914994407,1000 Comment length (words),R3 vs Gemini 3 Pro,78.2965078015314,68.76081309523809,9.535694706293318,8.658135133583217,10.392855308232667,3.234882270258317e-75,1.132208794590411e-73,0.9339512856430713,1000 Comment length (words),R3 vs GPT-5.2,78.2965078015314,64.18198102683631,14.114526774695102,13.418034711861472,14.796391868184612,1.1105190187259838e-137,5.330491289884722e-136,1.691090980277346,1000 Comment length (words),Gemini 3 Pro vs GPT-5.2,68.76081309523809,64.18198102683631,4.578832068401784,3.643285075650944,5.4930680766061295,2.3740807799287133e-21,4.273345403871684e-20,0.4236339810092716,1000 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,0.8326448651749856,0.645426945878753,0.18721791929623255,0.17039784136546185,0.20393478676611207,3.1332921219233894e-71,1.0653193214539524e-69,0.8566155079801938,996 Claim overlap fraction with human,R3 vs GPT-5.2,0.8326448651749856,0.9064304838401223,-0.07378561866513673,-0.08713186077643909,-0.060414873780837645,1.14189993447427e-23,2.1696098755011133e-22,-0.416833777844877,996 Claim overlap fraction with human,Gemini 3 Pro vs GPT-5.2,0.645426945878753,0.9064304838401223,-0.2610035379613693,-0.27652532750047815,-0.2456827309236948,1.0206556132334525e-107,4.08262245293381e-106,-1.2893428413612993,996 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,0.6481449607955633,0.45734844138458597,0.19079651941097725,0.1732090863453815,0.20850066934404285,3.3126221421233785e-67,9.937866426370135e-66,0.7195957447851364,996 Stance-matched overlap fraction with human,R3 vs GPT-5.2,0.6481449607955633,0.671163224325875,-0.023018263530311726,-0.036990880187416336,-0.009086166092943205,0.002253504220632824,0.015774529544429768,-0.08759705307578994,996 Stance-matched overlap fraction with human,Gemini 3 Pro vs GPT-5.2,0.45734844138458597,0.671163224325875,-0.21381478294128897,-0.23068984270414994,-0.19711674077261426,1.02572939911469e-82,3.692625836812884e-81,-0.82194312932501,996 Stance agreement (critical),R3 vs Gemini 3 Pro,0.9901382823871907,0.8502183406113537,0.139919941775837,,,,4.224371423368403e-38,0.6727110184919292,916 Stance agreement (critical),R3 vs GPT-5.2,0.9904711673699016,0.9301336146272855,0.060337552742616096,,,,1.3760950665946337e-23,0.49081190130973296,948 Stance agreement (critical),Gemini 3 Pro vs GPT-5.2,0.8470503076366268,0.9309808179515019,-0.08393051031487508,,,,9.44640980025022e-15,-0.36003388051037594,921