{ "all_passed": true, "checkpoints": { "grounded_cut_s811": "3ccd298e2f653aca93a1ea99dd6ff6c28867b3c027cefefa99945f871dcb4d56", "grounded_cut_s812": "8e53788f69360d855248ff162f7d04ff58a48b670a5f5178506946e70ad2235c", "grounded_cut_s813": "ad33c81366d0f263633a1507f751af99af4f1e32fb9a270407a51050947ef7df", "grounded_future_s811": "84809db146ceb43bb2b58bc84ae23b04dc507bf1c8bf05edd464d95f04d70fc4", "grounded_future_s812": "2fc34355d170abf02105af7f8a0d9478dc56bca57137aa2d227a3f50d925d720", "grounded_future_s813": "1499a605b5a622515da6e7e35c715902507b0ac0dd8495316282c767e0b6d6ce", "numeric_future_s811": "88e9b5f28581717bd9c61bc6f94650e3ae516495623b12d52d1fdbd7b6dfccdb", "numeric_future_s812": "57dd646fd8e971e5949848e2faecfd17bbc4501c42c3ed5a6cf2e446bee353f8", "numeric_future_s813": "a6dbd3fac5e821d7d259791f2f579b16f40f15cede3e81ec9fe56997176b4e61" }, "code": { "PREREGISTRATION.md": "a8b9f2421e2b2a34401a0189c11930e2003d395e348adf5328a909d4fd8aca6d", "campaign.py": "c27af6e25d71e88c48ce1d1b70986b33f301ce15016434efcfc804c2c1fcee04", "core.py": "33905ffd455385a657c066648349e0595aa165daecd61380b7e8c52b47968425" }, "per_seed": { "811": { "all_passed": true, "effects": { "advantage_grounded_cut": { "mean": 0.49609375, "paired_bootstrap_95": [ 0.4657958984375, 0.5263916015624999 ] }, "advantage_numeric_future": { "mean": 0.9267578125, "paired_bootstrap_95": [ 0.91015625, 0.9423828125 ] }, "drop_zero_relation": { "mean": 0.908203125, "paired_bootstrap_95": [ 0.8896484375, 0.92578125 ] }, "drop_zero_second_value": { "mean": 0.9609375, "paired_bootstrap_95": [ 0.94921875, 0.97265625 ] } }, "gates": { "advantage_grounded_cut": true, "advantage_numeric_future": true, "candidate_permutation_agrees": true, "current_only": true, "drop_zero_relation": true, "drop_zero_second_value": true, "identity_0": true, "identity_128": true, "identity_512": true, "identity_retention": true, "onehop_0": true, "onehop_128": true, "onehop_512": true, "oracle": true, "relation_cf_answer": true, "relation_cf_identity": true, "replay": true, "replay_agrees": true, "replay_margin": true, "restore_agrees": true, "twohop_0": true, "twohop_128": true, "twohop_512": true, "twohop_ci_0": true, "twohop_ci_128": true, "twohop_ci_512": true, "twohop_retention": true, "unrelated_changed": true, "value_cf_answer": true } }, "812": { "all_passed": true, "effects": { "advantage_grounded_cut": { "mean": 0.619140625, "paired_bootstrap_95": [ 0.5888671875, 0.650390625 ] }, "advantage_numeric_future": { "mean": 0.9248046875, "paired_bootstrap_95": [ 0.9072265625, 0.9404296875 ] }, "drop_zero_relation": { "mean": 0.908203125, "paired_bootstrap_95": [ 0.8896484375, 0.92578125 ] }, "drop_zero_second_value": { "mean": 0.9697265625, "paired_bootstrap_95": [ 0.958984375, 0.9794921875 ] } }, "gates": { "advantage_grounded_cut": true, "advantage_numeric_future": true, "candidate_permutation_agrees": true, "current_only": true, "drop_zero_relation": true, "drop_zero_second_value": true, "identity_0": true, "identity_128": true, "identity_512": true, "identity_retention": true, "onehop_0": true, "onehop_128": true, "onehop_512": true, "oracle": true, "relation_cf_answer": true, "relation_cf_identity": true, "replay": true, "replay_agrees": true, "replay_margin": true, "restore_agrees": true, "twohop_0": true, "twohop_128": true, "twohop_512": true, "twohop_ci_0": true, "twohop_ci_128": true, "twohop_ci_512": true, "twohop_retention": true, "unrelated_changed": true, "value_cf_answer": true } }, "813": { "all_passed": true, "effects": { "advantage_grounded_cut": { "mean": 0.5732421875, "paired_bootstrap_95": [ 0.54296875, 0.6044921875 ] }, "advantage_numeric_future": { "mean": 0.91796875, "paired_bootstrap_95": [ 0.9013671875, 0.9345703125 ] }, "drop_zero_relation": { "mean": 0.908203125, "paired_bootstrap_95": [ 0.8896484375, 0.92578125 ] }, "drop_zero_second_value": { "mean": 0.9697265625, "paired_bootstrap_95": [ 0.958984375, 0.9794921875 ] } }, "gates": { "advantage_grounded_cut": true, "advantage_numeric_future": true, "candidate_permutation_agrees": true, "current_only": true, "drop_zero_relation": true, "drop_zero_second_value": true, "identity_0": true, "identity_128": true, "identity_512": true, "identity_retention": true, "onehop_0": true, "onehop_128": true, "onehop_512": true, "oracle": true, "relation_cf_answer": true, "relation_cf_identity": true, "replay": true, "replay_agrees": true, "replay_margin": true, "restore_agrees": true, "twohop_0": true, "twohop_128": true, "twohop_512": true, "twohop_ci_0": true, "twohop_ci_128": true, "twohop_ci_512": true, "twohop_retention": true, "unrelated_changed": true, "value_cf_answer": true } } }, "split": "confirmation" }