{ "inputs": { "molmo2_npz": "reports/eval__fontaine_molmo2_ar_40k_ddp4__step_040000__panel_curated_v0_k4l2.npz", "anchor_npz": "reports/eval__fontaine_arb_rcond_40k_1xh100__step_040000__panel_curated_v0_k4l2.npz", "policy": "pred:bijou@40000", "anchor_policy": "pred:bijou@40000" }, "read1": { "molmo2_chunk": 6.0079, "molmo2_first": 2.1871, "anchor_chunk": 7.7966, "anchor_first": 3.9422, "beats_line": 7.3, "parity_band": 0.5, "classification": "BEATS", "paired": { "delta_pooled": -1.71667, "ci95": [ -1.79712, -1.63484 ], "frames": 17204 }, "pairing_failure": false, "alignment_columns": [ "index", "repo_id", "core", "truth", "valid" ], "decision": "Molmo2 becomes the phase-2 flow-trunk candidate (frozen AR-adapted prefix \u2014 kills the -2.7 confound)" }, "read2": { "state_copy_byte_match": true, "state_copy_norm_byte_match": true, "state_copy_pooled": [ 11.7847, 2.6202 ], "expected_pooled": [ 11.7847, 2.6202 ], "prereg_parenthetical_nonreproducing": [ 11.7639, 2.5851 ] }, "read3_context": { "e4b_probe_10k": 7.54, "statedrop_arm_c": 10.5 } }