{ "generated": "2026-07-04", "config": { "bootstrap_resamples": 10000, "bootstrap_seed": 20260704, "runs_included": [ "swarm-study1", "study1-replicate", "study2-contact", "study2-phaseB", "study2-independence", "study2-2x2" ], "excluded_runs_note": "study1-replicate arms B3/B7 (self-consistency resamples, no round-0-vs-post-debate structure) are excluded from this analysis even though the run is included; *-mock, demo-latest, difficulty-calibration*, real-baseline, study3* excluded per instructions (generational / no within-item debate, or synthetic).", "entropy_definition": "Shannon entropy in bits over instrument.claims.normalize(answer) counts across round-0 per-agent answers", "primary_outcome": "inflation_stated = mean(final-round per-agent stated_confidence) - mean(round-0 per-agent stated_confidence); both terms from claim_asserted events \u2014 the SAME construct at both ends.", "construct_correction_note": "The first committed version used answer_finalized.confidence (the CONVERGENCE vote-share signal) as the post term against mean STATED round-0 confidence \u2014 a construct mismatch whose negative entropy slope is partly mechanical (high-entropy items are likelier to end with a split final vote, lowering vote-share regardless of stated confidence). That outcome is retained as inflation_mixed for transparency only.", "secondary_outcomes": { "inflation_mixed": "answer_finalized.confidence (final vote-share) - mean round-0 stated_confidence. Mixed constructs; mechanically negative-biased at high entropy. NOT causal evidence.", "inflation_voteshare": "modal share of normalized final-round answers - modal share of normalized round-0 answers. r0 modal share is a transform of the same distribution entropy is computed on, so regression to the mean pulls this positive at high entropy. NOT causal evidence." }, "overconf_delta_stated_definition": "(post_conf_stated_mean - final_acc_mean) - (r0_conf_mean - r0_acc); per-agent stated confidence and per-agent accuracy at both ends", "overconf_delta_mixed_definition": "(post_conf_voteshare - post_acc) - (r0_conf_mean - r0_acc); the original mixed version, transparency only" }, "per_cell": { "study1-replicate::A3": { "run": "study1-replicate", "cell": "A3", "n_items": 60, "n_agents": 3, "entropy_bits": { "slope": 0.015422963491443615, "intercept": 0.020380235691955784, "r2": 0.011750299311517498, "slope_ci": { "mean": 0.014150428218737915, "lo": -0.0437417456559396, "hi": 0.08769240327816359, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.024444818783929567, "intercept": 0.020380235691955784, "r2": 0.011750299311517498, "slope_ci": { "mean": 0.022427898095846104, "lo": -0.06932902658074692, "hi": 0.1389891707940063, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.19583817068944256, "inflation_voteshare_on_entropy_bits": 0.18551600334517862, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.28512951406917797, "mean_inflation_stated": 0.024777777777777774, "mean_inflation_mixed": 0.03727777777777787, "mean_inflation_voteshare": 0.03333333333333333, "mean_r0_conf": 0.8793888888888891, "mean_post_conf_stated": 0.9041666666666667, "mean_post_conf_voteshare": 0.9166666666666665, "mean_r0_acc": 0.8555555555555557, "mean_final_acc": 0.8666666666666666, "mean_post_acc": 0.8833333333333333, "mean_overconf_delta_stated": 0.013666666666666662, "mean_overconf_delta_mixed": 0.009500000000000102, "unanimity_rate": 0.75 }, "study1-replicate::A7": { "run": "study1-replicate", "cell": "A7", "n_items": 60, "n_agents": 7, "entropy_bits": { "slope": 0.044908578825383856, "intercept": 0.020442437627464963, "r2": 0.15400220616171967, "slope_ci": { "mean": 0.04329097298606168, "lo": -0.005985631489149997, "hi": 0.10065887775422071, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.1260743198080534, "intercept": 0.02044243762746495, "r2": 0.15400220616171967, "slope_ci": { "mean": 0.12153312609308244, "lo": -0.016803792022688235, "hi": 0.28258519591210585, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.14813245619055215, "inflation_voteshare_on_entropy_bits": 0.09151459422906663, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.4026262324321067, "mean_inflation_stated": 0.03852380952380955, "mean_inflation_mixed": 0.04449999999999995, "mean_inflation_voteshare": 0.038095238095238085, "mean_r0_conf": 0.8721666666666668, "mean_post_conf_stated": 0.9106904761904764, "mean_post_conf_voteshare": 0.9166666666666665, "mean_r0_acc": 0.8666666666666666, "mean_final_acc": 0.8857142857142857, "mean_post_acc": 0.8833333333333333, "mean_overconf_delta_stated": 0.019476190476190498, "mean_overconf_delta_mixed": 0.02783333333333326, "unanimity_rate": 0.6666666666666666 }, "study2-2x2::HET_NT": { "run": "study2-2x2", "cell": "HET_NT", "n_items": 87, "n_agents": 7, "entropy_bits": { "slope": 0.03181376142632534, "intercept": 0.018291467730121306, "r2": 0.1683628563727867, "slope_ci": { "mean": 0.03151311928102291, "lo": 0.010128316918957823, "hi": 0.05071707089144198, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.0893125197293607, "intercept": 0.01829146773012133, "r2": 0.1683628563727868, "slope_ci": { "mean": 0.0884685105229679, "lo": 0.028433780354595525, "hi": 0.14238081859943427, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.06457333431975949, "inflation_voteshare_on_entropy_bits": 0.12463846552715122, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.6586205061289832, "mean_inflation_stated": 0.03924466338259442, "mean_inflation_mixed": -0.017963875205254564, "mean_inflation_voteshare": 0.08702791461412154, "mean_r0_conf": 0.8898850574712639, "mean_post_conf_stated": 0.9291297208538585, "mean_post_conf_voteshare": 0.8719211822660096, "mean_r0_acc": 0.7323481116584564, "mean_final_acc": 0.7586206896551724, "mean_post_acc": 0.7701149425287356, "mean_overconf_delta_stated": 0.012972085385878487, "mean_overconf_delta_mixed": -0.05573070607553376, "unanimity_rate": 0.5402298850574713 }, "study2-2x2::HET_RET": { "run": "study2-2x2", "cell": "HET_RET", "n_items": 89, "n_agents": 7, "entropy_bits": { "slope": 0.024146588152515334, "intercept": 0.050116337852142076, "r2": 0.085949294812244, "slope_ci": { "mean": 0.0241770794436996, "lo": 0.0042398922280955376, "hi": 0.0456302849733098, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.06778804310086181, "intercept": 0.05011633785214206, "r2": 0.085949294812244, "slope_ci": { "mean": 0.06787364297724771, "lo": 0.011902882315537803, "hi": 0.12810040511471263, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.07438635027273113, "inflation_voteshare_on_entropy_bits": 0.11228547206415729, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.723345532876998, "mean_inflation_stated": 0.06758266452648469, "mean_inflation_mixed": 0.024420545746388352, "mean_inflation_voteshare": 0.1027287319422151, "mean_r0_conf": 0.8503788121990368, "mean_post_conf_stated": 0.9179614767255212, "mean_post_conf_voteshare": 0.8747993579454253, "mean_r0_acc": 0.7351524879614766, "mean_final_acc": 0.770465489566613, "mean_post_acc": 0.797752808988764, "mean_overconf_delta_stated": 0.03226966292134825, "mean_overconf_delta_mixed": -0.038179775280899025, "unanimity_rate": 0.48314606741573035 }, "study2-2x2::HOMO_NT": { "run": "study2-2x2", "cell": "HOMO_NT", "n_items": 87, "n_agents": 7, "entropy_bits": { "slope": 0.059850275290346956, "intercept": 0.03065600251007787, "r2": 0.2690621272271889, "slope_ci": { "mean": 0.059681907694479855, "lo": 0.02309983025464245, "hi": 0.09973332278725258, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.16802096492285826, "intercept": 0.030656002510077843, "r2": 0.2690621272271887, "slope_ci": { "mean": 0.1675482973238856, "lo": 0.06484942216406568, "hi": 0.2799868346199534, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": 0.02844300497257057, "inflation_voteshare_on_entropy_bits": 0.15904249779957289, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.6027766025564005, "mean_inflation_stated": 0.0667323481116585, "mean_inflation_mixed": 0.08164203612479469, "mean_inflation_voteshare": 0.09688013136289, "mean_r0_conf": 0.8346141215106736, "mean_post_conf_stated": 0.9013464696223316, "mean_post_conf_voteshare": 0.9162561576354677, "mean_r0_acc": 0.7175697865353037, "mean_final_acc": 0.7471264367816092, "mean_post_acc": 0.7471264367816092, "mean_overconf_delta_stated": 0.03717569786535305, "mean_overconf_delta_mixed": 0.05208538587848925, "unanimity_rate": 0.5747126436781609 }, "study2-2x2::HOMO_RET": { "run": "study2-2x2", "cell": "HOMO_RET", "n_items": 87, "n_agents": 7, "entropy_bits": { "slope": 0.059831412090483385, "intercept": 0.049953123114635484, "r2": 0.21033415277349865, "slope_ci": { "mean": 0.05998577237921616, "lo": 0.022348155886630427, "hi": 0.10324278740208831, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.16796800922587543, "intercept": 0.04995312311463546, "r2": 0.21033415277349865, "slope_ci": { "mean": 0.16840135334221928, "lo": 0.06273920542724262, "hi": 0.28983914738019884, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.0021996512378263584, "inflation_voteshare_on_entropy_bits": 0.10377967798497746, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.624155173999182, "mean_inflation_stated": 0.0872972085385879, "mean_inflation_mixed": 0.0885221674876847, "mean_inflation_voteshare": 0.08538587848932677, "mean_r0_conf": 0.8063875205254512, "mean_post_conf_stated": 0.8936847290640396, "mean_post_conf_voteshare": 0.8949096880131362, "mean_r0_acc": 0.7274220032840721, "mean_final_acc": 0.7471264367816094, "mean_post_acc": 0.735632183908046, "mean_overconf_delta_stated": 0.06759277504105093, "mean_overconf_delta_mixed": 0.08031198686371098, "unanimity_rate": 0.5747126436781609 }, "study2-contact::NT": { "run": "study2-contact", "cell": "NT", "n_items": 120, "n_agents": 7, "entropy_bits": { "slope": 0.07007782797927749, "intercept": -0.008525019754819688, "r2": 0.1367861727526487, "slope_ci": { "mean": 0.07026873864886497, "lo": 0.02905706552651892, "hi": 0.11678364094058097, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.19673333530473025, "intercept": -0.00852501975481941, "r2": 0.1367861727526486, "slope_ci": { "mean": 0.19726928931267096, "lo": 0.08157349592642317, "hi": 0.32785312921034776, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.10101670364114024, "inflation_voteshare_on_entropy_bits": 0.08859326597265829, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 1.2197495723251712, "mean_inflation_stated": 0.07695238095238095, "mean_inflation_mixed": 0.010035714285714249, "mean_inflation_voteshare": 0.17738095238095233, "mean_r0_conf": 0.7947261904761901, "mean_post_conf_stated": 0.8716785714285707, "mean_post_conf_voteshare": 0.8047619047619048, "mean_r0_acc": 0.6023809523809525, "mean_final_acc": 0.6404761904761905, "mean_post_acc": 0.65, "mean_overconf_delta_stated": 0.03885714285714286, "mean_overconf_delta_mixed": -0.0375833333333334, "unanimity_rate": 0.25833333333333336 }, "study2-contact::T": { "run": "study2-contact", "cell": "T", "n_items": 120, "n_agents": 7, "entropy_bits": { "slope": 0.06190753753186603, "intercept": -0.017989979477186183, "r2": 0.1758726023175432, "slope_ci": { "mean": 0.061792181826904324, "lo": 0.025366098814365926, "hi": 0.1019985204292674, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.17379643020255037, "intercept": -0.017989979477186342, "r2": 0.17587260231754276, "slope_ci": { "mean": 0.17347258579643865, "lo": 0.0712116423599099, "hi": 0.2863460483696967, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.1264849521007817, "inflation_voteshare_on_entropy_bits": 0.07110726323828212, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 1.0053699752659204, "mean_inflation_stated": 0.044250000000000005, "mean_inflation_mixed": -0.047416666666666725, "mean_inflation_voteshare": 0.10476190476190475, "mean_r0_conf": 0.8331309523809518, "mean_post_conf_stated": 0.8773809523809518, "mean_post_conf_voteshare": 0.7857142857142859, "mean_r0_acc": 0.6273809523809523, "mean_final_acc": 0.6357142857142858, "mean_post_acc": 0.6583333333333333, "mean_overconf_delta_stated": 0.03591666666666667, "mean_overconf_delta_mixed": -0.07836904761904764, "unanimity_rate": 0.375 }, "study2-independence::HETERO": { "run": "study2-independence", "cell": "HETERO", "n_items": 90, "n_agents": 7, "entropy_bits": { "slope": 0.03489226509463433, "intercept": 0.0164009274040392, "r2": 0.2038659915248222, "slope_ci": { "mean": 0.03489014431842084, "lo": 0.019001761093176125, "hi": 0.049854230482485, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.09795497215516058, "intercept": 0.016400927404039162, "r2": 0.2038659915248222, "slope_ci": { "mean": 0.09794901838361936, "lo": 0.053344687532690706, "hi": 0.1399585193303985, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.03333206153513481, "inflation_voteshare_on_entropy_bits": 0.16494738519762805, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.4857316915138712, "mean_inflation_stated": 0.03334920634920634, "mean_inflation_mixed": 0.011650793650793596, "mean_inflation_voteshare": 0.07777777777777779, "mean_r0_conf": 0.8899365079365081, "mean_post_conf_stated": 0.9232857142857139, "mean_post_conf_voteshare": 0.9015873015873015, "mean_r0_acc": 0.7238095238095238, "mean_final_acc": 0.7603174603174604, "mean_post_acc": 0.7666666666666667, "mean_overconf_delta_stated": -0.0031587301587301707, "mean_overconf_delta_mixed": -0.031206349206349262, "unanimity_rate": 0.6555555555555556 }, "study2-independence::HOMO": { "run": "study2-independence", "cell": "HOMO", "n_items": 90, "n_agents": 7, "entropy_bits": { "slope": 0.08095552270506788, "intercept": 0.0243154426885732, "r2": 0.22313964357337812, "slope_ci": { "mean": 0.08264196207359532, "lo": 0.04500672361825579, "hi": 0.13071222275720482, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.2272708851338185, "intercept": 0.024315442688573195, "r2": 0.22313964357337812, "slope_ci": { "mean": 0.23200531899580537, "lo": 0.12634984707539673, "hi": 0.3669556019305282, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": 0.042208689829034245, "inflation_voteshare_on_entropy_bits": 0.16614613306752166, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.38373552111515763, "mean_inflation_stated": 0.0553809523809524, "mean_inflation_mixed": 0.08982539682539677, "mean_inflation_voteshare": 0.06507936507936508, "mean_r0_conf": 0.8482698412698413, "mean_post_conf_stated": 0.9036507936507937, "mean_post_conf_voteshare": 0.938095238095238, "mean_r0_acc": 0.7603174603174605, "mean_final_acc": 0.7571428571428571, "mean_post_acc": 0.7666666666666667, "mean_overconf_delta_stated": 0.05855555555555557, "mean_overconf_delta_mixed": 0.08347619047619044, "unanimity_rate": 0.6777777777777778 }, "study2-phaseB::CODE": { "run": "study2-phaseB", "cell": "CODE", "n_items": 90, "n_agents": 7, "entropy_bits": { "slope": -0.0047165922644683235, "intercept": 0.017870777367465743, "r2": 0.0013470631022421964, "slope_ci": { "mean": -0.00527970011246151, "lo": -0.026745399239465194, "hi": 0.01364387038983708, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": -0.013241148508993956, "intercept": 0.01787077736746574, "r2": 0.0013470631022421964, "slope_ci": { "mean": -0.014821992097706991, "lo": -0.0750838281973082, "hi": 0.03830318669482507, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.15585308579430776, "inflation_voteshare_on_entropy_bits": 0.09464251641207848, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.30913822374503674, "mean_inflation_stated": 0.016412698412698424, "mean_inflation_mixed": 0.01636507936507932, "mean_inflation_voteshare": 0.026984126984126985, "mean_r0_conf": 0.9233174603174605, "mean_post_conf_stated": 0.9397301587301584, "mean_post_conf_voteshare": 0.9396825396825397, "mean_r0_acc": 0.7365079365079364, "mean_final_acc": 0.7285714285714285, "mean_post_acc": 0.7444444444444445, "mean_overconf_delta_stated": 0.024349206349206353, "mean_overconf_delta_mixed": 0.008428571428571379, "unanimity_rate": 0.7222222222222222 }, "study2-phaseB::NT": { "run": "study2-phaseB", "cell": "NT", "n_items": 90, "n_agents": 7, "entropy_bits": { "slope": 0.029956284178940907, "intercept": 0.04942153699818522, "r2": 0.08250165089876504, "slope_ci": { "mean": 0.029591409167346417, "lo": 0.002580513552804177, "hi": 0.058126721429674635, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.08409792183630614, "intercept": 0.04942153699818522, "r2": 0.08250165089876504, "slope_ci": { "mean": 0.08307358817657047, "lo": 0.007244417423901196, "hi": 0.163182337508668, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.056158189011718346, "inflation_voteshare_on_entropy_bits": 0.16774784669966683, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.41035625922998376, "mean_inflation_stated": 0.061714285714285756, "mean_inflation_mixed": 0.08431746031746029, "mean_inflation_voteshare": 0.06349206349206349, "mean_r0_conf": 0.8537777777777781, "mean_post_conf_stated": 0.9154920634920635, "mean_post_conf_voteshare": 0.938095238095238, "mean_r0_acc": 0.6841269841269841, "mean_final_acc": 0.7142857142857142, "mean_post_acc": 0.7111111111111111, "mean_overconf_delta_stated": 0.03155555555555559, "mean_overconf_delta_mixed": 0.057333333333333306, "unanimity_rate": 0.7 }, "study2-phaseB::RETRIEVAL": { "run": "study2-phaseB", "cell": "RETRIEVAL", "n_items": 90, "n_agents": 7, "entropy_bits": { "slope": 0.10787365305498431, "intercept": 0.075529733881805, "r2": 0.42845967255008954, "slope_ci": { "mean": 0.10781818596007113, "lo": 0.07351324995948984, "hi": 0.14263435339906783, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.30283963086424415, "intercept": 0.07552973388180506, "r2": 0.42845967255008977, "slope_ci": { "mean": 0.3026839150423271, "lo": 0.2063777841102251, "hi": 0.4004252540693764, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.04085522071314677, "inflation_voteshare_on_entropy_bits": 0.08273111396746083, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.4872876993109707, "mean_inflation_stated": 0.12809523809523818, "mean_inflation_mixed": 0.1303492063492063, "mean_inflation_voteshare": 0.047619047619047616, "mean_r0_conf": 0.776, "mean_post_conf_stated": 0.9040952380952383, "mean_post_conf_voteshare": 0.9063492063492063, "mean_r0_acc": 0.6555555555555554, "mean_final_acc": 0.6793650793650792, "mean_post_acc": 0.7, "mean_overconf_delta_stated": 0.10428571428571433, "mean_overconf_delta_mixed": 0.08590476190476187, "unanimity_rate": 0.6222222222222222 }, "swarm-study1::swarm": { "run": "swarm-study1", "cell": "swarm", "n_items": 66, "n_agents": 3, "entropy_bits": { "slope": 0.09257780998050047, "intercept": 0.019791486328630675, "r2": 0.13758769013070693, "slope_ci": { "mean": 0.09442103281955151, "lo": 0.020581216072800917, "hi": 0.1966112291516211, "defined_in": 10000, "resamples": 10000 } }, "entropy_norm": { "slope": 0.14673235721798208, "intercept": 0.01979148632863068, "r2": 0.13758769013070693, "slope_ci": { "mean": 0.1496537962983513, "lo": 0.03262045569462903, "hi": 0.31162142542601307, "defined_in": 10000, "resamples": 10000 } }, "secondary_slopes_point_estimates_only": { "inflation_mixed_on_entropy_bits": -0.04202127955269104, "inflation_voteshare_on_entropy_bits": 0.21383247242203585, "note": "mixed is mechanically negative-biased at high entropy (split final vote lowers vote-share); voteshare is mechanically positive-biased (r0 modal share is a transform of the entropy distribution; regression to the mean). Neither supports causal readings." }, "mean_entropy_bits": 0.2630212122960379, "mean_inflation_stated": 0.04414141414141415, "mean_inflation_mixed": 0.10330808080808082, "mean_inflation_voteshare": 0.03535353535353536, "mean_r0_conf": 0.8259848484848489, "mean_post_conf_stated": 0.8701262626262629, "mean_post_conf_voteshare": 0.9292929292929291, "mean_r0_acc": 0.8282828282828283, "mean_final_acc": 0.8282828282828284, "mean_post_acc": 0.8181818181818182, "mean_overconf_delta_stated": 0.04414141414141415, "mean_overconf_delta_mixed": 0.11340909090909092, "unanimity_rate": 0.7575757575757576 } }, "pooled_by_n_agents": { "3": { "n_agents": 3, "n_items_pooled": 126, "pooled_fit": { "slope": 0.05292594019090058, "intercept": 0.02044279822811304, "r2": 0.06785898981891714 }, "pooled_slope_ci": { "mean": 0.05324246268022592, "lo": 0.00425091929402394, "hi": 0.11122982613234354, "defined_in": 10000, "resamples": 10000 }, "per_cell_delta_vs_pooled": { "study1-replicate::A3": { "run": "study1-replicate", "cell": "A3", "n_items": 60, "cell_slope": 0.015422963491443615, "delta_vs_pooled_ci": { "mean": -0.03894379005168047, "lo": -0.1215796423329279, "hi": 0.050156355650017984, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.04381020356707803 }, "swarm-study1::swarm": { "run": "swarm-study1", "cell": "swarm", "n_items": 66, "cell_slope": 0.09257780998050047, "delta_vs_pooled_ci": { "mean": 0.04172972158280856, "lo": -0.050376030120192664, "hi": 0.15535775271152863, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.052483107865235024 } }, "heterogeneity": { "method": "Cochran's-Q analogue using bootstrap-CI-derived SEs (se = (hi-lo)/(2*1.96) per cell's own bootstrap of cell_slope, NOT the delta-vs-pooled bootstrap above)", "limits": "This is NOT a textbook Cochran's Q: the per-cell SE is backed out from a percentile bootstrap CI width under a normal-approximation assumption, not computed analytically from the OLS covariance matrix, and it ignores that the pooled estimate and each cell's estimate share resampling machinery (not independent studies as in a meta-analysis). It also assumes each cell_slope bootstrap distribution is symmetric enough for the (hi-lo)/(2*1.96) SE proxy to be meaningful, which is not guaranteed at small per-cell n. Treat Q and its 'df' as a descriptive heterogeneity indicator only, not a calibrated test statistic.", "Q": 1.89538627741014, "df": 1, "weighted_mean_slope": 0.04304011109703873, "n_cells": 2 } }, "7": { "n_agents": 7, "n_items_pooled": 1100, "pooled_fit": { "slope": 0.05011621213008058, "intercept": 0.02802389626048115, "r2": 0.14062008602178788 }, "pooled_slope_ci": { "mean": 0.050093297380812446, "lo": 0.03743237304642159, "hi": 0.06353480720965211, "defined_in": 10000, "resamples": 10000 }, "per_cell_delta_vs_pooled": { "study1-replicate::A7": { "run": "study1-replicate", "cell": "A7", "n_items": 60, "cell_slope": 0.044908578825383856, "delta_vs_pooled_ci": { "mean": -0.007118862202530087, "lo": -0.057779209645203576, "hi": 0.05165987457730023, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.027918133730230565 }, "study2-2x2::HET_NT": { "run": "study2-2x2", "cell": "HET_NT", "n_items": 87, "cell_slope": 0.03181376142632534, "delta_vs_pooled_ci": { "mean": -0.018489902660351977, "lo": -0.04338651467549655, "hi": 0.005056077106063241, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.012357804025908109 }, "study2-2x2::HET_RET": { "run": "study2-2x2", "cell": "HET_RET", "n_items": 89, "cell_slope": 0.024146588152515334, "delta_vs_pooled_ci": { "mean": -0.025873997033965503, "lo": -0.05016342956267173, "hi": -0.0008911062549406198, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.012569470231564059 }, "study2-2x2::HOMO_NT": { "run": "study2-2x2", "cell": "HOMO_NT", "n_items": 87, "cell_slope": 0.059850275290346956, "delta_vs_pooled_ci": { "mean": 0.00979741237173458, "lo": -0.028871882167631986, "hi": 0.051325307757809115, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.020458466817714568 }, "study2-2x2::HOMO_RET": { "run": "study2-2x2", "cell": "HOMO_RET", "n_items": 87, "cell_slope": 0.059831412090483385, "delta_vs_pooled_ci": { "mean": 0.010290475734129013, "lo": -0.0295364148411994, "hi": 0.056105227683095, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.021847357786809796 }, "study2-contact::NT": { "run": "study2-contact", "cell": "NT", "n_items": 120, "cell_slope": 0.07007782797927749, "delta_vs_pooled_ci": { "mean": 0.02020053084843934, "lo": -0.023668049600707442, "hi": 0.06807617889953288, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.023404139923530694 }, "study2-contact::T": { "run": "study2-contact", "cell": "T", "n_items": 120, "cell_slope": 0.06190753753186603, "delta_vs_pooled_ci": { "mean": 0.01195495387055063, "lo": -0.0279477619162975, "hi": 0.054122114464843406, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.02093619295437268 }, "study2-independence::HETERO": { "run": "study2-independence", "cell": "HETERO", "n_items": 90, "cell_slope": 0.03489226509463433, "delta_vs_pooled_ci": { "mean": -0.015195463809806728, "lo": -0.035443722494396354, "hi": 0.004496327686212498, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.01018878831137981 }, "study2-independence::HOMO": { "run": "study2-independence", "cell": "HOMO", "n_items": 90, "cell_slope": 0.08095552270506788, "delta_vs_pooled_ci": { "mean": 0.03251153130508722, "lo": -0.00757013463882792, "hi": 0.0833775243475115, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.023200933414882503 }, "study2-phaseB::CODE": { "run": "study2-phaseB", "cell": "CODE", "n_items": 90, "cell_slope": -0.0047165922644683235, "delta_vs_pooled_ci": { "mean": -0.05540366754815791, "lo": -0.08032589180035998, "hi": -0.03187005237957839, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.01236118352570959 }, "study2-phaseB::NT": { "run": "study2-phaseB", "cell": "NT", "n_items": 90, "cell_slope": 0.029956284178940907, "delta_vs_pooled_ci": { "mean": -0.02052447583800153, "lo": -0.05133336287067974, "hi": 0.011089536660446757, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.01592420906406288 }, "study2-phaseB::RETRIEVAL": { "run": "study2-phaseB", "cell": "RETRIEVAL", "n_items": 90, "cell_slope": 0.10787365305498431, "delta_vs_pooled_ci": { "mean": 0.057658780714410975, "lo": 0.021135807041335775, "hi": 0.09477885154910146, "defined_in": 10000, "resamples": 10000 }, "approx_se_from_ci": 0.018786490945858592 } }, "heterogeneity": { "method": "Cochran's-Q analogue using bootstrap-CI-derived SEs (se = (hi-lo)/(2*1.96) per cell's own bootstrap of cell_slope, NOT the delta-vs-pooled bootstrap above)", "limits": "This is NOT a textbook Cochran's Q: the per-cell SE is backed out from a percentile bootstrap CI width under a normal-approximation assumption, not computed analytically from the OLS covariance matrix, and it ignores that the pooled estimate and each cell's estimate share resampling machinery (not independent studies as in a meta-analysis). It also assumes each cell_slope bootstrap distribution is symmetric enough for the (hi-lo)/(2*1.96) SE proxy to be meaningful, which is not guaranteed at small per-cell n. Treat Q and its 'df' as a descriptive heterogeneity indicator only, not a calibrated test statistic.", "Q": 45.08544437905333, "df": 11, "weighted_mean_slope": 0.03622367943615476, "n_cells": 12 } } }, "run_diagnostics": { "swarm-study1": { "n_records": 66, "n_skipped": 0, "n_agents_by_cell": { "swarm": 3 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 }, "study1-replicate": { "n_records": 120, "n_skipped": 0, "n_agents_by_cell": { "A3": 3, "A7": 7 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 }, "study2-contact": { "n_records": 240, "n_skipped": 4, "n_agents_by_cell": { "NT": 7, "T": 7 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 }, "study2-phaseB": { "n_records": 270, "n_skipped": 0, "n_agents_by_cell": { "NT": 7, "CODE": 7, "RETRIEVAL": 7 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 }, "study2-independence": { "n_records": 180, "n_skipped": 0, "n_agents_by_cell": { "HOMO": 7, "HETERO": 7 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 }, "study2-2x2": { "n_records": 350, "n_skipped": 6, "n_agents_by_cell": { "HOMO_NT": 7, "HOMO_RET": 7, "HET_NT": 7, "HET_RET": 7 }, "missing_r0_conf_items": 0, "missing_final_conf_items": 0 } }, "skipped_items": [ { "run": "study2-contact", "cell": "NT", "task_id": "triviaqa-tc_2758", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item in a resumed run)" }, { "run": "study2-contact", "cell": "NT", "task_id": "mmlu-high_school_chemistry-175", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item in a resumed run)" }, { "run": "study2-contact", "cell": "T", "task_id": "mmlu-astronomy-33", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item in a resumed run)" }, { "run": "study2-contact", "cell": "T", "task_id": "mmlu-college_computer_science-69", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item in a resumed run)" }, { "run": "study2-2x2", "cell": "HOMO_NT", "task_id": "triviaqa-qz_2384", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" }, { "run": "study2-2x2", "cell": "HOMO_NT", "task_id": "triviaqa-qz_2379", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" }, { "run": "study2-2x2", "cell": "HOMO_RET", "task_id": "triviaqa-qz_1720", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" }, { "run": "study2-2x2", "cell": "HOMO_RET", "task_id": "triviaqa-qz_2379", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" }, { "run": "study2-2x2", "cell": "HET_NT", "task_id": "triviaqa-qz_1720", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" }, { "run": "study2-2x2", "cell": "HET_NT", "task_id": "triviaqa-qz_2379", "reason": "content-match alignment: unmatched or ambiguous segment (retried/orphaned item within a multi-pass run)" } ], "n_total_records": 1226, "n_total_skipped": 10 }