2026-07-02_s3_m1_subtractive_ablation.json

{ "path": "real", "split": "dev", "n_seeds": 3, "timestamp": "2026-07-02T22:22:17Z", "stage_names": [ "Stage 0: Sense grounding", "Stage 1: Transfer within domain", "Stage 2: Scope control", "Stage 3: Dialog", "Stage 4: Consolidation stress", "Stage 5: Cross-domain disambiguation" ], "configs": { "FULL": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8095238095238095, "std": 0.08247860988423221, "min": 0.7142857142857143, "max": 0.8571428571428571, "ci95_half": 0.2049047619047618 }, "Stage 2: Scope control": { "n": 3, "mean": 0.717948717948718, "std": 0.044411559168432785, "min": 0.6923076923076923, "max": 0.7692307692307693, "ci95_half": 0.11033333333333342 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8518518518518519, "std": 0.06415002990995838, "min": 0.7777777777777778, "max": 0.8888888888888888, "ci95_half": 0.1593703703703703 } } }, "MINIMAL": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.4, "std": 0.0, "min": 0.4, "max": 0.4, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.9047619047619048, "std": 0.08247860988423228, "min": 0.8571428571428571, "max": 1.0, "ci95_half": 0.20490476190476198 }, "Stage 2: Scope control": { "n": 3, "mean": 0.6410256410256411, "std": 0.04441155916843272, "min": 0.6153846153846154, "max": 0.6923076923076923, "ci95_half": 0.11033333333333327 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.5555555555555556, "std": 0.0962250448649376, "min": 0.5, "max": 0.6666666666666666, "ci95_half": 0.23905555555555552 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.5925925925925926, "std": 0.1697250257391052, "min": 0.4444444444444444, "max": 0.7777777777777778, "ci95_half": 0.42165436635225684 } } }, "FULL-hippocampus": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.6666666666666666, "std": 0.04441155916843272, "min": 0.6153846153846154, "max": 0.6923076923076923, "ci95_half": 0.11033333333333327 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8148148148148148, "std": 0.06415002990995838, "min": 0.7777777777777778, "max": 0.8888888888888888, "ci95_half": 0.1593703703703703 } } }, "FULL-critic": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.6410256410256411, "std": 0.04441155916843272, "min": 0.6153846153846154, "max": 0.6923076923076923, "ci95_half": 0.11033333333333327 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8888888888888888, "std": 0.0, "min": 0.8888888888888888, "max": 0.8888888888888888, "ci95_half": 0.0 } } }, "FULL-identity": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.7435897435897436, "std": 0.0888231183368655, "min": 0.6923076923076923, "max": 0.8461538461538461, "ci95_half": 0.22066666666666668 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8518518518518519, "std": 0.06415002990995838, "min": 0.7777777777777778, "max": 0.8888888888888888, "ci95_half": 0.1593703703703703 } } }, "FULL-immune": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.6666666666666666, "std": 0.04441155916843272, "min": 0.6153846153846154, "max": 0.6923076923076923, "ci95_half": 0.11033333333333327 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8518518518518519, "std": 0.06415002990995838, "min": 0.7777777777777778, "max": 0.8888888888888888, "ci95_half": 0.1593703703703703 } } }, "FULL-autoencoder": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.717948717948718, "std": 0.044411559168432785, "min": 0.6923076923076923, "max": 0.7692307692307693, "ci95_half": 0.11033333333333342 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.8888888888888888, "std": 0.0, "min": 0.8888888888888888, "max": 0.8888888888888888, "ci95_half": 0.0 } } }, "FULL-dsi": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8571428571428571, "std": 0.0, "min": 0.8571428571428571, "max": 0.8571428571428571, "ci95_half": 0.0 }, "Stage 2: Scope control": { "n": 3, "mean": 0.8461538461538461, "std": 0.0, "min": 0.8461538461538461, "max": 0.8461538461538461, "ci95_half": 0.0 }, "Stage 3: Dialog": { "n": 3, "mean": 0.4, "std": 0.0, "min": 0.4, "max": 0.4, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 1.0, "std": 0.0, "min": 1.0, "max": 1.0, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.6666666666666666, "std": 0.0, "min": 0.6666666666666666, "max": 0.6666666666666666, "ci95_half": 0.0 } } }, "FULL-scope_slot_reranker": { "per_stage": { "Stage 0: Sense grounding": { "n": 3, "mean": 0.8, "std": 0.0, "min": 0.8, "max": 0.8, "ci95_half": 0.0 }, "Stage 1: Transfer within domain": { "n": 3, "mean": 0.8095238095238095, "std": 0.08247860988423221, "min": 0.7142857142857143, "max": 0.8571428571428571, "ci95_half": 0.2049047619047618 }, "Stage 2: Scope control": { "n": 3, "mean": 0.5128205128205129, "std": 0.0888231183368655, "min": 0.46153846153846156, "max": 0.6153846153846154, "ci95_half": 0.22066666666666668 }, "Stage 3: Dialog": { "n": 3, "mean": 0.2, "std": 0.0, "min": 0.2, "max": 0.2, "ci95_half": 0.0 }, "Stage 4: Consolidation stress": { "n": 3, "mean": 0.8333333333333334, "std": 0.0, "min": 0.8333333333333334, "max": 0.8333333333333334, "ci95_half": 0.0 }, "Stage 5: Cross-domain disambiguation": { "n": 3, "mean": 0.7777777777777778, "std": 0.0, "min": 0.7777777777777778, "max": 0.7777777777777778, "ci95_half": 0.0 } } } } }