{
  "extractors": [
    {
      "methods": {
        "diffmean": 0.70369840434056,
        "lda": 0.6566224717949725,
        "logistic": 0.761681334698207,
        "pca": 0.5181975164095537
      },
      "model": "Qwen3.5 0.8B",
      "n": 209,
      "tag": "q35_08b"
    },
    {
      "methods": {
        "diffmean": 0.6618340900794871,
        "lda": 0.7548680706908082,
        "logistic": 0.7632614473922017,
        "pca": 0.5059367497050727
      },
      "model": "Qwen3.5 2B",
      "n": 347,
      "tag": "q35_2b"
    },
    {
      "methods": {
        "diffmean": 0.7932825523809524,
        "lda": 0.8400204095238096,
        "logistic": 0.8818476857142857,
        "pca": 0.5452562476190476
      },
      "model": "Qwen3.5 4B",
      "n": 500,
      "tag": "q35_4b"
    },
    {
      "methods": {
        "diffmean": 0.7431032952380953,
        "lda": 0.8528380095238095,
        "logistic": 0.8793459142857143,
        "pca": 0.543778780952381
      },
      "model": "Qwen3.5 9B",
      "n": 500,
      "tag": "q35_9b"
    },
    {
      "methods": {
        "diffmean": 0.9309445142857143,
        "lda": 0.9552896285714285,
        "logistic": 0.9776637142857142,
        "pca": 0.8383184571428571
      },
      "model": "Qwen3.5 27B",
      "n": 500,
      "tag": "q35_27b"
    },
    {
      "methods": {
        "diffmean": 0.7248650277203442,
        "lda": 0.816719820416816,
        "logistic": 0.832091440060756,
        "pca": 0.49825427786220455
      },
      "model": "Qwen2.5 3B",
      "n": 247,
      "tag": "q25_3b"
    },
    {
      "methods": {
        "diffmean": 0.761920938200353,
        "lda": 0.9066794310654291,
        "logistic": 0.9139288131433332,
        "pca": 0.5163538009956169
      },
      "model": "Qwen2.5 7B",
      "n": 392,
      "tag": "q25_7b"
    },
    {
      "methods": {
        "diffmean": 0.7726812439261418,
        "lda": 0.8511389698736638,
        "logistic": 0.8897502429543246,
        "pca": 0.6530923226433432
      },
      "model": "Llama-3.1 8B†",
      "n": 70,
      "tag": "llama8b"
    },
    {
      "methods": {
        "diffmean": 0.864577707818772,
        "lda": 0.7467853694674442,
        "logistic": 0.8988365760470033,
        "pca": 0.5335766257292729
      },
      "model": "Mistral 7B",
      "n": 183,
      "tag": "mistral7b"
    },
    {
      "methods": {
        "diffmean": 0.8353833714285713,
        "lda": 0.9487146761904761,
        "logistic": 0.9602292761904762,
        "pca": 0.5535573333333333
      },
      "model": "Gemma-2 9B",
      "n": 500,
      "tag": "gemma9b"
    },
    {
      "methods": {
        "diffmean": 0.7658135333333334,
        "lda": 0.9150156095238096,
        "logistic": 0.9277397238095237,
        "pca": 0.5795932190476191
      },
      "model": "Gemma-4 12B",
      "n": 500,
      "tag": "gemma4_12b"
    }
  ],
  "method_scope": "held-out factual false-vs-correct direction transfer",
  "phrasing": [
    {
      "matrix": [
        [
          0.7153681544239672,
          0.7188837453052669,
          0.7403266660843741,
          0.6285701808018167,
          0.6225434535767316
        ],
        [
          0.7042973185431043,
          0.7096252947855707,
          0.7299109092497161,
          0.6183509476810202,
          0.6140710979124815
        ],
        [
          0.6597519434011704,
          0.6675910559874224,
          0.688509913529566,
          0.5970390427111538,
          0.5817975369027862
        ],
        [
          0.6343348764084199,
          0.6408638309022623,
          0.6521311904969866,
          0.5949209537950912,
          0.5604856319329199
        ],
        [
          0.6983579351908464,
          0.716285265088654,
          0.7310463795964713,
          0.6269761551227182,
          0.6802559175473841
        ]
      ],
      "mean_off_diagonal": 0.657180758144816,
      "min_off_diagonal": 0.5604856319329199,
      "model": "Qwen3.5 0.8B",
      "n": 214,
      "tag": "q35_08b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.715625,
          0.7310648148148148,
          0.7485108024691357,
          0.6519521604938272,
          0.565108024691358
        ],
        [
          0.7052314814814814,
          0.7214969135802469,
          0.7408256172839507,
          0.6447839506172839,
          0.5650462962962963
        ],
        [
          0.7036265432098765,
          0.7213734567901234,
          0.7447993827160494,
          0.6451388888888889,
          0.5660185185185185
        ],
        [
          0.72875,
          0.7439120370370371,
          0.7651851851851852,
          0.6785339506172839,
          0.5745833333333333
        ],
        [
          0.7113117283950617,
          0.7286959876543209,
          0.7499305555555555,
          0.6694984567901234,
          0.631141975308642
        ]
      ],
      "mean_off_diagonal": 0.6830273919753087,
      "min_off_diagonal": 0.5650462962962963,
      "model": "Qwen3.5 2B",
      "n": 360,
      "tag": "q35_2b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.7833117777479173,
          0.8049541453910238,
          0.8617265184090299,
          0.6636614821284601,
          0.804689599570008
        ],
        [
          0.7666285944638537,
          0.788590096748186,
          0.8440313423810802,
          0.6543057981725342,
          0.790173172534265
        ],
        [
          0.7699543133566245,
          0.7950693697930664,
          0.8583378124160173,
          0.6448619322762699,
          0.7911053816178446
        ],
        [
          0.7975132692824509,
          0.8134825987637733,
          0.8415328540714863,
          0.7388932746573502,
          0.825630710830422
        ],
        [
          0.7690389008331094,
          0.7880400094060737,
          0.8449635514646601,
          0.6541462308519215,
          0.796875
        ]
      ],
      "mean_off_diagonal": 0.7762754887798978,
      "min_off_diagonal": 0.6448619322762699,
      "model": "Qwen3.5 4B",
      "n": 488,
      "tag": "q35_4b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.7604431854853169,
          0.7788614387990486,
          0.8256700172509274,
          0.7537061605332249,
          0.7208975263046761
        ],
        [
          0.7559969880598085,
          0.775072609272473,
          0.8216939374743527,
          0.7484631535332529,
          0.7172959470280994
        ],
        [
          0.75144720538962,
          0.772463057916104,
          0.8205186433520185,
          0.7377898892833097,
          0.71243540862387
        ],
        [
          0.74615638981518,
          0.7601523499906374,
          0.8100565336393083,
          0.7582479751076688,
          0.7017382400866928
        ],
        [
          0.7473635563204928,
          0.7642160788204031,
          0.814956912522261,
          0.7434512213098752,
          0.7086346269536775
        ]
      ],
      "mean_off_diagonal": 0.7592406006350573,
      "min_off_diagonal": 0.7017382400866928,
      "model": "Qwen3.5 9B",
      "n": 501,
      "tag": "q35_9b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.9502163416074509,
          0.9707851984514495,
          0.9644474121992737,
          0.9277947823267626,
          0.9587349130194804
        ],
        [
          0.9431490537708285,
          0.9681875552433409,
          0.9629227924857477,
          0.9077547176365693,
          0.9543270251388563
        ],
        [
          0.9088045823506934,
          0.9528178446123027,
          0.9556200570477958,
          0.8202647048606421,
          0.9320830165083507
        ],
        [
          0.9755211690552374,
          0.9815385921777359,
          0.9708392356058531,
          0.9830130345336014,
          0.9764822584442703
        ],
        [
          0.9397562924336404,
          0.9668752243506857,
          0.9622897858198788,
          0.894693165457907,
          0.9526634527425786
        ]
      ],
      "mean_off_diagonal": 0.9435940883353083,
      "min_off_diagonal": 0.8202647048606421,
      "model": "Qwen3.5 27B",
      "n": 509,
      "tag": "q35_27b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.7690528124793444,
          0.7719115605790204,
          0.7310132857426135,
          0.7616167625090885,
          0.754230286205301
        ],
        [
          0.7669872430431621,
          0.7734483442395399,
          0.7320378081829598,
          0.7603774208473792,
          0.7534205829863178
        ],
        [
          0.7662271134906471,
          0.7731013285742614,
          0.7496199352237426,
          0.7703086786965431,
          0.7490415757816116
        ],
        [
          0.7736135897944346,
          0.7724733954656621,
          0.7389285478220635,
          0.7787692511071452,
          0.7571551325269352
        ],
        [
          0.7335911163989688,
          0.7308810892986979,
          0.6935851675589927,
          0.7297904686363936,
          0.7369456011633289
        ]
      ],
      "mean_off_diagonal": 0.7510146077070528,
      "min_off_diagonal": 0.6935851675589927,
      "model": "Qwen2.5 3B",
      "n": 246,
      "tag": "q25_3b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.8111927354351597,
          0.8341814610754005,
          0.8047393123150698,
          0.7991085093357821,
          0.8226902867054382
        ],
        [
          0.8029282726252422,
          0.8317136006529946,
          0.7999438832772167,
          0.7866161616161618,
          0.8135585654525048
        ],
        [
          0.8058361391694725,
          0.832912457912458,
          0.8066396286093256,
          0.7899640342822161,
          0.8163707784919907
        ],
        [
          0.817658912355882,
          0.8370574431180491,
          0.8104147535965718,
          0.8166960004081215,
          0.8327275278032854
        ],
        [
          0.8095666258545047,
          0.8312098255280073,
          0.8012128864401592,
          0.8005943271094786,
          0.8320196918681767
        ]
      ],
      "mean_off_diagonal": 0.8124646082032447,
      "min_off_diagonal": 0.7866161616161618,
      "model": "Qwen2.5 7B",
      "n": 396,
      "tag": "q25_7b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.8252467968914095,
          0.8626339004410839,
          0.8809073724007561,
          0.7708464608275573,
          0.8258769166141567
        ],
        [
          0.8076034446544844,
          0.8470909472799832,
          0.8599033816425121,
          0.7445914723797521,
          0.8124343625288806
        ],
        [
          0.7914303717706364,
          0.8195757193866834,
          0.842680109220752,
          0.7477420709934888,
          0.7966813694601975
        ],
        [
          0.8857382902751523,
          0.9140936777987818,
          0.9296366309598824,
          0.8435202688510817,
          0.8785969334173493
        ],
        [
          0.7884898130644822,
          0.8267170762444865,
          0.827977315689981,
          0.7387103549674437,
          0.8019323671497585
        ]
      ],
      "mean_off_diagonal": 0.8255093467758874,
      "min_off_diagonal": 0.7387103549674437,
      "model": "Llama-3.1 8B†",
      "n": 69,
      "tag": "llama8b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.9240756098258458,
          0.924161400097229,
          0.9155823729589064,
          0.9110926820898511,
          0.9013697846664189
        ],
        [
          0.9178987102862536,
          0.9215876919557322,
          0.9103491664045298,
          0.9054305241785582,
          0.8927907575280963
        ],
        [
          0.9207011924847722,
          0.9212159341130716,
          0.9170408075724213,
          0.9110354885755956,
          0.9068317652778175
        ],
        [
          0.8821241671194487,
          0.887729131516486,
          0.8866996482598875,
          0.91097829506134,
          0.8888444050444679
        ],
        [
          0.8279619091195058,
          0.8303068432039807,
          0.8204981555091653,
          0.8434613514827419,
          0.8983099316537505
        ]
      ],
      "mean_off_diagonal": 0.8903042694958392,
      "min_off_diagonal": 0.8204981555091653,
      "model": "Mistral 7B",
      "n": 187,
      "tag": "mistral7b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.8652629108692032,
          0.8964253811564141,
          0.879983042892398,
          0.8327870217565747,
          0.852227134400218
        ],
        [
          0.8516063831397902,
          0.8874661236354827,
          0.8683742373086647,
          0.8173515117564233,
          0.8391043013520265
        ],
        [
          0.8553006101530681,
          0.8902292237581189,
          0.8721820163817773,
          0.8210041030144287,
          0.8432868022225922
        ],
        [
          0.8737073990522188,
          0.9042112673923905,
          0.8887530469802722,
          0.8434041393510877,
          0.8607662492997623
        ],
        [
          0.8625565867764842,
          0.8949832699965178,
          0.8783781737800724,
          0.8297589668276584,
          0.8515571772471953
        ]
      ],
      "mean_off_diagonal": 0.8620397356508047,
      "min_off_diagonal": 0.8173515117564233,
      "model": "Gemma-2 9B",
      "n": 514,
      "tag": "gemma9b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    },
    {
      "matrix": [
        [
          0.7542565597667639,
          0.7731361932528112,
          0.8570595585172847,
          0.7664056643065389,
          0.7423573511037068
        ],
        [
          0.7461016243231986,
          0.7673219491878385,
          0.8554935443565181,
          0.7616243231986672,
          0.7385630987088714
        ],
        [
          0.7127780091628488,
          0.7352478134110787,
          0.8459683465222824,
          0.7148146605581007,
          0.7114827155351937
        ],
        [
          0.7743731778425655,
          0.7922448979591836,
          0.8614827155351936,
          0.8094585589337776,
          0.7579466888796335
        ],
        [
          0.7391461890878801,
          0.7599916701374427,
          0.8511745106205747,
          0.7512661391087047,
          0.7352519783423573
        ]
      ],
      "mean_off_diagonal": 0.7701345272802999,
      "min_off_diagonal": 0.7114827155351937,
      "model": "Gemma-4 12B",
      "n": 490,
      "tag": "gemma4_12b",
      "templates": [
        "belief",
        "confident",
        "authority",
        "question",
        "casual"
      ]
    }
  ],
  "phrasing_scope": "five English phrasings only, disjoint questions",
  "provenance": {
    "inputs": [
      {
        "path": "data/raw/paper-final/methods_gemma4_12b.json",
        "sha256": "d5d0755ffe3cddc9a456a8b0b83d2a321c8b10f3c83e37fe1d591e2bc9e9fdc9"
      },
      {
        "path": "data/raw/paper-final/methods_gemma9b.json",
        "sha256": "51643a46b45daf367bf797ac83d353ac38a3499df70e5473acbe42ee565c72d4"
      },
      {
        "path": "data/raw/paper-final/methods_llama8b.json",
        "sha256": "871d3392cca0b74ba819c41bbd9316ed43b21d51bc669177cfe86b07b2ba6581"
      },
      {
        "path": "data/raw/paper-final/methods_mistral7b.json",
        "sha256": "84eaf34f904f97cd6286b2be6aa7b747c7b6d8b082b540fadb33eb19a7780c4e"
      },
      {
        "path": "data/raw/paper-final/methods_q25_3b.json",
        "sha256": "9f80b4d4f35b5515f815187805b23bc975950e68ed16b47891f10bba1a25589a"
      },
      {
        "path": "data/raw/paper-final/methods_q25_7b.json",
        "sha256": "560f2954eedea644b42526565b2e0f962dd659f3284f93bd7ee78ee6ce3b44a2"
      },
      {
        "path": "data/raw/paper-final/methods_q35_08b.json",
        "sha256": "1a2d4fcfb02f22a1c50c819cbda9ac81b5f5f0918b9e6883b497cd256ebfd59e"
      },
      {
        "path": "data/raw/paper-final/methods_q35_27b.json",
        "sha256": "ad6c05a70e1829d04bb946eecd89081a9053e1d75cdfebccc0eaf229c1349c7c"
      },
      {
        "path": "data/raw/paper-final/methods_q35_2b.json",
        "sha256": "112ae826387a5495958410203eaa69f213c0bcb37be89304835014ce94dcb985"
      },
      {
        "path": "data/raw/paper-final/methods_q35_4b.json",
        "sha256": "fd503113e3252f0660de598caca2413f849dbd3b0edc525c4ef6256959fae5c7"
      },
      {
        "path": "data/raw/paper-final/methods_q35_9b.json",
        "sha256": "d43fc6b1bf1ff83d142618b6bb576abbb43a1a6913c0fd523525f68d4dd9c4c6"
      },
      {
        "path": "data/raw/paper-final/promptvar_gemma4_12b.json",
        "sha256": "dd4995d3a47fee9f28c66be74900f4e0896c8fde9ac3287f9295cb03372118ea"
      },
      {
        "path": "data/raw/paper-final/promptvar_gemma9b.json",
        "sha256": "303ffc89e3c893dc743b9de70c284cd46f1b57c4869a0e04b8a562b4fc736d5b"
      },
      {
        "path": "data/raw/paper-final/promptvar_llama8b.json",
        "sha256": "c74492671b71107f7530a2b55968703d8fa8f075722208542dece247ee995b39"
      },
      {
        "path": "data/raw/paper-final/promptvar_mistral7b.json",
        "sha256": "2f9c6b42ec21b2cd6d4f2ab7457e4a4ba9e012486bbea5fbb3878128425fccd5"
      },
      {
        "path": "data/raw/paper-final/promptvar_q25_3b.json",
        "sha256": "c7d49f969588dd41e6f2586d2dbacb5938ae15c618dffa4e0b525417f7b070fb"
      },
      {
        "path": "data/raw/paper-final/promptvar_q25_7b.json",
        "sha256": "75da398132dc6c22bf313a7c0818d78252957ad0aeeebbee86208b05781745a9"
      },
      {
        "path": "data/raw/paper-final/promptvar_q35_08b.json",
        "sha256": "a4bcff50857ce8451d4ae8524ba999287a7825ed4406cd078ded5932c46e34f4"
      },
      {
        "path": "data/raw/paper-final/promptvar_q35_27b.json",
        "sha256": "43e1322340838c34963fa1bebb88c7c64144ee2a7d690a652cf59a7cd7a2ef86"
      },
      {
        "path": "data/raw/paper-final/promptvar_q35_2b.json",
        "sha256": "d1773a9d9007a81e07b3fb4474857b27ec06804b7fcd6c1449f61a7cff30fbf8"
      },
      {
        "path": "data/raw/paper-final/promptvar_q35_4b.json",
        "sha256": "f9559642b0904076580b8b4768df615b9a9f07314e2b2ae5c0a6a6325e81cefe"
      },
      {
        "path": "data/raw/paper-final/promptvar_q35_9b.json",
        "sha256": "fe4547ddd77422c087f4a55e5f0f6098f007b3e2fd66f46a988626317c013d9f"
      }
    ],
    "run_id": "paper-final"
  }
}
