{
  "suite_id": "novel_feedback_v3_suite",
  "uses_reranking": false,
  "scenarios": [
    {
      "scenario": "baseline_clean",
      "label": "Clean multi-turn feedback",
      "description": "Public text + calibrated explicit/implicit feedback with clean state transitions.",
      "difficulty": 0.08,
      "noise": 0.0,
      "drift": 0.02,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.3316,
          "expectation_alignment": 0.024,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.5684,
          "overgeneralization_rate": 0.0696,
          "stale_memory_use_rate": 0.5324,
          "calibration_ece": 0.1492,
          "counterfactual_separation": 0.0822,
          "rollback_success": 0.133,
          "behavioral_scar_rate": 0.348,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1768,
          "robustness_score": 0.2614,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.4372,
          "expectation_alignment": 0.294,
          "ncfu": 0.1992,
          "selectivity": 0.2007,
          "regret_to_oracle": 0.4628,
          "overgeneralization_rate": 0.1696,
          "stale_memory_use_rate": 0.3924,
          "calibration_ece": 0.1352,
          "counterfactual_separation": 0.1092,
          "rollback_success": 0.2464,
          "behavioral_scar_rate": 0.258,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2568,
          "robustness_score": 0.4014,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.3836,
          "expectation_alignment": 0.254,
          "ncfu": 0.1592,
          "selectivity": 0.1277,
          "regret_to_oracle": 0.5164,
          "overgeneralization_rate": 0.2296,
          "stale_memory_use_rate": 0.3424,
          "calibration_ece": 0.1382,
          "counterfactual_separation": 0.1052,
          "rollback_success": 0.2296,
          "behavioral_scar_rate": 0.298,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2168,
          "robustness_score": 0.3714,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.4552,
          "expectation_alignment": 0.394,
          "ncfu": 0.2792,
          "selectivity": 0.2237,
          "regret_to_oracle": 0.4448,
          "overgeneralization_rate": 0.3096,
          "stale_memory_use_rate": 0.2824,
          "calibration_ece": 0.1332,
          "counterfactual_separation": 0.1192,
          "rollback_success": 0.2884,
          "behavioral_scar_rate": 0.388,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2968,
          "robustness_score": 0.4214,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.5892,
          "expectation_alignment": 0.714,
          "ncfu": 0.5592,
          "selectivity": 0.6372,
          "regret_to_oracle": 0.3108,
          "overgeneralization_rate": 0.1396,
          "stale_memory_use_rate": 0.1324,
          "calibration_ece": 0.1112,
          "counterfactual_separation": 0.1512,
          "rollback_success": 0.4228,
          "behavioral_scar_rate": 0.168,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3568,
          "robustness_score": 0.6414,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.6592,
          "expectation_alignment": 0.764,
          "ncfu": 0.6392,
          "selectivity": 0.7257,
          "regret_to_oracle": 0.2408,
          "overgeneralization_rate": 0.0696,
          "stale_memory_use_rate": 0.1024,
          "calibration_ece": 0.0672,
          "counterfactual_separation": 0.1562,
          "rollback_success": 0.4438,
          "behavioral_scar_rate": 0.128,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3768,
          "robustness_score": 0.7314,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.6092,
          "expectation_alignment": 0.704,
          "ncfu": 0.5892,
          "selectivity": 0.685,
          "regret_to_oracle": 0.2908,
          "overgeneralization_rate": 0.0346,
          "stale_memory_use_rate": 0.1224,
          "calibration_ece": 0.0692,
          "counterfactual_separation": 0.1502,
          "rollback_success": 0.5986,
          "behavioral_scar_rate": 0.028,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3368,
          "robustness_score": 0.7114,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.8092,
          "expectation_alignment": 0.854,
          "ncfu": 0.9792,
          "selectivity": 0.8377,
          "regret_to_oracle": 0.0908,
          "overgeneralization_rate": 0.0296,
          "stale_memory_use_rate": 0.0124,
          "calibration_ece": 0.0262,
          "counterfactual_separation": 0.1652,
          "rollback_success": 0.7216,
          "behavioral_scar_rate": 0.038,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4768,
          "robustness_score": 0.9014,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Remaining errors are mostly fine-grained expectation timing rather than preference-axis confusion."
    },
    {
      "scenario": "implicit_only",
      "label": "Implicit-only reading signals",
      "description": "Only dwell, reread, fast-swipe, scroll-depth and continue signals are visible; comments/ratings are hidden.",
      "difficulty": 0.24,
      "noise": 0.18,
      "drift": 0.04,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2576,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6424,
          "overgeneralization_rate": 0.1248,
          "stale_memory_use_rate": 0.5512,
          "calibration_ece": 0.178,
          "counterfactual_separation": 0.073,
          "rollback_success": 0.1026,
          "behavioral_scar_rate": 0.3856,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1704,
          "robustness_score": 0.1836,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.3284,
          "expectation_alignment": 0.2132,
          "ncfu": 0.132,
          "selectivity": 0.0896,
          "regret_to_oracle": 0.5716,
          "overgeneralization_rate": 0.2248,
          "stale_memory_use_rate": 0.4112,
          "calibration_ece": 0.164,
          "counterfactual_separation": 0.1,
          "rollback_success": 0.216,
          "behavioral_scar_rate": 0.2956,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2504,
          "robustness_score": 0.3236,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.2538,
          "expectation_alignment": 0.1732,
          "ncfu": 0.092,
          "selectivity": 0.0166,
          "regret_to_oracle": 0.6462,
          "overgeneralization_rate": 0.2848,
          "stale_memory_use_rate": 0.3612,
          "calibration_ece": 0.167,
          "counterfactual_separation": 0.096,
          "rollback_success": 0.1992,
          "behavioral_scar_rate": 0.3356,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2104,
          "robustness_score": 0.2936,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.2948,
          "expectation_alignment": 0.3132,
          "ncfu": 0.212,
          "selectivity": 0.1126,
          "regret_to_oracle": 0.6052,
          "overgeneralization_rate": 0.3648,
          "stale_memory_use_rate": 0.3012,
          "calibration_ece": 0.162,
          "counterfactual_separation": 0.11,
          "rollback_success": 0.258,
          "behavioral_scar_rate": 0.4256,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2904,
          "robustness_score": 0.3436,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.5024,
          "expectation_alignment": 0.6332,
          "ncfu": 0.492,
          "selectivity": 0.5261,
          "regret_to_oracle": 0.3976,
          "overgeneralization_rate": 0.1948,
          "stale_memory_use_rate": 0.1512,
          "calibration_ece": 0.14,
          "counterfactual_separation": 0.142,
          "rollback_success": 0.3924,
          "behavioral_scar_rate": 0.2056,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3504,
          "robustness_score": 0.5636,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.6074,
          "expectation_alignment": 0.7042,
          "ncfu": 0.5965,
          "selectivity": 0.6356,
          "regret_to_oracle": 0.2926,
          "overgeneralization_rate": 0.1248,
          "stale_memory_use_rate": 0.1212,
          "calibration_ece": 0.096,
          "counterfactual_separation": 0.147,
          "rollback_success": 0.4134,
          "behavioral_scar_rate": 0.1656,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3774,
          "robustness_score": 0.6728,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.5574,
          "expectation_alignment": 0.6442,
          "ncfu": 0.5465,
          "selectivity": 0.5948,
          "regret_to_oracle": 0.3426,
          "overgeneralization_rate": 0.0898,
          "stale_memory_use_rate": 0.1412,
          "calibration_ece": 0.098,
          "counterfactual_separation": 0.141,
          "rollback_success": 0.5682,
          "behavioral_scar_rate": 0.0656,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3374,
          "robustness_score": 0.6528,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.7224,
          "expectation_alignment": 0.7732,
          "ncfu": 0.912,
          "selectivity": 0.7266,
          "regret_to_oracle": 0.1776,
          "overgeneralization_rate": 0.0848,
          "stale_memory_use_rate": 0.0312,
          "calibration_ece": 0.0286,
          "counterfactual_separation": 0.156,
          "rollback_success": 0.6912,
          "behavioral_scar_rate": 0.0756,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4704,
          "robustness_score": 0.8236,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Dwell ambiguity: slow reading can mean delight, confusion, or fatigue."
    },
    {
      "scenario": "explicit_only",
      "label": "Explicit-only comments and ratings",
      "description": "Ratings, likes, tags and comments remain visible, but implicit behavior is ablated.",
      "difficulty": 0.16,
      "noise": 0.08,
      "drift": 0.03,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2958,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6042,
          "overgeneralization_rate": 0.0952,
          "stale_memory_use_rate": 0.5418,
          "calibration_ece": 0.1628,
          "counterfactual_separation": 0.0778,
          "rollback_success": 0.1186,
          "behavioral_scar_rate": 0.3656,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1736,
          "robustness_score": 0.2249,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.3846,
          "expectation_alignment": 0.2554,
          "ncfu": 0.1672,
          "selectivity": 0.148,
          "regret_to_oracle": 0.5154,
          "overgeneralization_rate": 0.1952,
          "stale_memory_use_rate": 0.4018,
          "calibration_ece": 0.1488,
          "counterfactual_separation": 0.1048,
          "rollback_success": 0.232,
          "behavioral_scar_rate": 0.2756,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2536,
          "robustness_score": 0.3649,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.3212,
          "expectation_alignment": 0.2154,
          "ncfu": 0.1272,
          "selectivity": 0.075,
          "regret_to_oracle": 0.5788,
          "overgeneralization_rate": 0.2552,
          "stale_memory_use_rate": 0.3518,
          "calibration_ece": 0.1518,
          "counterfactual_separation": 0.1008,
          "rollback_success": 0.2152,
          "behavioral_scar_rate": 0.3156,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2136,
          "robustness_score": 0.3349,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.3792,
          "expectation_alignment": 0.3554,
          "ncfu": 0.2472,
          "selectivity": 0.171,
          "regret_to_oracle": 0.5208,
          "overgeneralization_rate": 0.3352,
          "stale_memory_use_rate": 0.2918,
          "calibration_ece": 0.1468,
          "counterfactual_separation": 0.1148,
          "rollback_success": 0.274,
          "behavioral_scar_rate": 0.4056,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2936,
          "robustness_score": 0.3849,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.5474,
          "expectation_alignment": 0.6754,
          "ncfu": 0.5272,
          "selectivity": 0.5845,
          "regret_to_oracle": 0.3526,
          "overgeneralization_rate": 0.1652,
          "stale_memory_use_rate": 0.1418,
          "calibration_ece": 0.1248,
          "counterfactual_separation": 0.1468,
          "rollback_success": 0.4084,
          "behavioral_scar_rate": 0.1856,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3536,
          "robustness_score": 0.6049,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.6174,
          "expectation_alignment": 0.7254,
          "ncfu": 0.6072,
          "selectivity": 0.673,
          "regret_to_oracle": 0.2826,
          "overgeneralization_rate": 0.0952,
          "stale_memory_use_rate": 0.1118,
          "calibration_ece": 0.0808,
          "counterfactual_separation": 0.1518,
          "rollback_success": 0.4294,
          "behavioral_scar_rate": 0.1456,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3736,
          "robustness_score": 0.6949,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.5674,
          "expectation_alignment": 0.6654,
          "ncfu": 0.5572,
          "selectivity": 0.6323,
          "regret_to_oracle": 0.3326,
          "overgeneralization_rate": 0.0602,
          "stale_memory_use_rate": 0.1318,
          "calibration_ece": 0.0828,
          "counterfactual_separation": 0.1458,
          "rollback_success": 0.5842,
          "behavioral_scar_rate": 0.0456,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3336,
          "robustness_score": 0.6749,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.7674,
          "expectation_alignment": 0.8154,
          "ncfu": 0.9472,
          "selectivity": 0.785,
          "regret_to_oracle": 0.1326,
          "overgeneralization_rate": 0.0552,
          "stale_memory_use_rate": 0.0218,
          "calibration_ece": 0.0274,
          "counterfactual_separation": 0.1608,
          "rollback_success": 0.7072,
          "behavioral_scar_rate": 0.0556,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4736,
          "robustness_score": 0.8649,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Sparse comments miss silent disengagement and late fatigue."
    },
    {
      "scenario": "delayed_feedback_2_turn",
      "label": "Delayed feedback by two turns",
      "description": "Feedback arrives two chapters after the causal content, testing delayed credit assignment.",
      "difficulty": 0.34,
      "noise": 0.15,
      "drift": 0.08,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2224,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6776,
          "overgeneralization_rate": 0.1308,
          "stale_memory_use_rate": 0.5712,
          "calibration_ece": 0.1846,
          "counterfactual_separation": 0.0701,
          "rollback_success": 0.095,
          "behavioral_scar_rate": 0.392,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1664,
          "robustness_score": 0.1656,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.2734,
          "expectation_alignment": 0.1834,
          "ncfu": 0.1084,
          "selectivity": 0.0565,
          "regret_to_oracle": 0.6266,
          "overgeneralization_rate": 0.2308,
          "stale_memory_use_rate": 0.4312,
          "calibration_ece": 0.1706,
          "counterfactual_separation": 0.0971,
          "rollback_success": 0.2084,
          "behavioral_scar_rate": 0.302,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2464,
          "robustness_score": 0.3056,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.1973,
          "expectation_alignment": 0.1434,
          "ncfu": 0.0684,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7027,
          "overgeneralization_rate": 0.2908,
          "stale_memory_use_rate": 0.3812,
          "calibration_ece": 0.1736,
          "counterfactual_separation": 0.0931,
          "rollback_success": 0.1916,
          "behavioral_scar_rate": 0.342,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2064,
          "robustness_score": 0.2756,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.2478,
          "expectation_alignment": 0.2834,
          "ncfu": 0.1884,
          "selectivity": 0.0795,
          "regret_to_oracle": 0.6522,
          "overgeneralization_rate": 0.3708,
          "stale_memory_use_rate": 0.3212,
          "calibration_ece": 0.1686,
          "counterfactual_separation": 0.1071,
          "rollback_success": 0.2504,
          "behavioral_scar_rate": 0.432,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2864,
          "robustness_score": 0.3256,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.466,
          "expectation_alignment": 0.6034,
          "ncfu": 0.4684,
          "selectivity": 0.493,
          "regret_to_oracle": 0.434,
          "overgeneralization_rate": 0.2008,
          "stale_memory_use_rate": 0.1712,
          "calibration_ece": 0.1466,
          "counterfactual_separation": 0.1391,
          "rollback_success": 0.3848,
          "behavioral_scar_rate": 0.212,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3464,
          "robustness_score": 0.5456,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.536,
          "expectation_alignment": 0.6534,
          "ncfu": 0.5484,
          "selectivity": 0.5815,
          "regret_to_oracle": 0.364,
          "overgeneralization_rate": 0.1308,
          "stale_memory_use_rate": 0.1412,
          "calibration_ece": 0.1026,
          "counterfactual_separation": 0.1441,
          "rollback_success": 0.4058,
          "behavioral_scar_rate": 0.172,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3664,
          "robustness_score": 0.6356,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.486,
          "expectation_alignment": 0.5934,
          "ncfu": 0.4984,
          "selectivity": 0.5407,
          "regret_to_oracle": 0.414,
          "overgeneralization_rate": 0.0958,
          "stale_memory_use_rate": 0.1612,
          "calibration_ece": 0.1046,
          "counterfactual_separation": 0.1381,
          "rollback_success": 0.5606,
          "behavioral_scar_rate": 0.072,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3264,
          "robustness_score": 0.6156,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.686,
          "expectation_alignment": 0.7434,
          "ncfu": 0.8884,
          "selectivity": 0.6935,
          "regret_to_oracle": 0.214,
          "overgeneralization_rate": 0.0908,
          "stale_memory_use_rate": 0.0512,
          "calibration_ece": 0.0301,
          "counterfactual_separation": 0.1531,
          "rollback_success": 0.6836,
          "behavioral_scar_rate": 0.082,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4664,
          "robustness_score": 0.8056,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Credit assignment can over-attribute frustration to the most recent chapter."
    },
    {
      "scenario": "noisy_feedback",
      "label": "Noisy and inconsistent feedback",
      "description": "Explicit ratings are perturbed and comments contain mixed sentiment; implicit behavior remains partially reliable.",
      "difficulty": 0.3,
      "noise": 0.28,
      "drift": 0.05,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2254,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6746,
          "overgeneralization_rate": 0.152,
          "stale_memory_use_rate": 0.559,
          "calibration_ece": 0.1914,
          "counterfactual_separation": 0.0689,
          "rollback_success": 0.0886,
          "behavioral_scar_rate": 0.4036,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.168,
          "robustness_score": 0.1463,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.2806,
          "expectation_alignment": 0.1766,
          "ncfu": 0.1012,
          "selectivity": 0.038,
          "regret_to_oracle": 0.6194,
          "overgeneralization_rate": 0.252,
          "stale_memory_use_rate": 0.419,
          "calibration_ece": 0.1774,
          "counterfactual_separation": 0.0959,
          "rollback_success": 0.202,
          "behavioral_scar_rate": 0.3136,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.248,
          "robustness_score": 0.2863,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.196,
          "expectation_alignment": 0.1366,
          "ncfu": 0.0612,
          "selectivity": 0.0,
          "regret_to_oracle": 0.704,
          "overgeneralization_rate": 0.312,
          "stale_memory_use_rate": 0.369,
          "calibration_ece": 0.1804,
          "counterfactual_separation": 0.0919,
          "rollback_success": 0.1852,
          "behavioral_scar_rate": 0.3536,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.208,
          "robustness_score": 0.2563,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.2204,
          "expectation_alignment": 0.2766,
          "ncfu": 0.1812,
          "selectivity": 0.061,
          "regret_to_oracle": 0.6796,
          "overgeneralization_rate": 0.392,
          "stale_memory_use_rate": 0.309,
          "calibration_ece": 0.1754,
          "counterfactual_separation": 0.1059,
          "rollback_success": 0.244,
          "behavioral_scar_rate": 0.4436,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.288,
          "robustness_score": 0.3063,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4642,
          "expectation_alignment": 0.5966,
          "ncfu": 0.4612,
          "selectivity": 0.4745,
          "regret_to_oracle": 0.4358,
          "overgeneralization_rate": 0.222,
          "stale_memory_use_rate": 0.159,
          "calibration_ece": 0.1534,
          "counterfactual_separation": 0.1379,
          "rollback_success": 0.3784,
          "behavioral_scar_rate": 0.2236,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.348,
          "robustness_score": 0.5263,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.5342,
          "expectation_alignment": 0.6466,
          "ncfu": 0.5412,
          "selectivity": 0.563,
          "regret_to_oracle": 0.3658,
          "overgeneralization_rate": 0.152,
          "stale_memory_use_rate": 0.129,
          "calibration_ece": 0.1094,
          "counterfactual_separation": 0.1429,
          "rollback_success": 0.3994,
          "behavioral_scar_rate": 0.1836,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.368,
          "robustness_score": 0.6163,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.4842,
          "expectation_alignment": 0.5866,
          "ncfu": 0.4912,
          "selectivity": 0.5222,
          "regret_to_oracle": 0.4158,
          "overgeneralization_rate": 0.117,
          "stale_memory_use_rate": 0.149,
          "calibration_ece": 0.1114,
          "counterfactual_separation": 0.1369,
          "rollback_success": 0.5542,
          "behavioral_scar_rate": 0.0836,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.328,
          "robustness_score": 0.5963,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.6842,
          "expectation_alignment": 0.7366,
          "ncfu": 0.8812,
          "selectivity": 0.675,
          "regret_to_oracle": 0.2158,
          "overgeneralization_rate": 0.112,
          "stale_memory_use_rate": 0.039,
          "calibration_ece": 0.0295,
          "counterfactual_separation": 0.1519,
          "rollback_success": 0.6772,
          "behavioral_scar_rate": 0.0936,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.468,
          "robustness_score": 0.8,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Raw-feedback methods chase contradictory surface phrases."
    },
    {
      "scenario": "cold_start_users",
      "label": "Cold-start reader priors",
      "description": "Only one or two initial events are available; NovelUpdates/Qidian metadata priors seed the reader state.",
      "difficulty": 0.27,
      "noise": 0.12,
      "drift": 0.06,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2514,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6486,
          "overgeneralization_rate": 0.1164,
          "stale_memory_use_rate": 0.5596,
          "calibration_ece": 0.1759,
          "counterfactual_separation": 0.0732,
          "rollback_success": 0.1044,
          "behavioral_scar_rate": 0.3814,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1692,
          "robustness_score": 0.1894,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.3174,
          "expectation_alignment": 0.212,
          "ncfu": 0.1318,
          "selectivity": 0.093,
          "regret_to_oracle": 0.5826,
          "overgeneralization_rate": 0.2164,
          "stale_memory_use_rate": 0.4196,
          "calibration_ece": 0.1619,
          "counterfactual_separation": 0.1002,
          "rollback_success": 0.2178,
          "behavioral_scar_rate": 0.2914,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2492,
          "robustness_score": 0.3294,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.2464,
          "expectation_alignment": 0.172,
          "ncfu": 0.0918,
          "selectivity": 0.02,
          "regret_to_oracle": 0.6536,
          "overgeneralization_rate": 0.2764,
          "stale_memory_use_rate": 0.3696,
          "calibration_ece": 0.1649,
          "counterfactual_separation": 0.0962,
          "rollback_success": 0.201,
          "behavioral_scar_rate": 0.3314,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2092,
          "robustness_score": 0.2994,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.3002,
          "expectation_alignment": 0.312,
          "ncfu": 0.2118,
          "selectivity": 0.116,
          "regret_to_oracle": 0.5998,
          "overgeneralization_rate": 0.3564,
          "stale_memory_use_rate": 0.3096,
          "calibration_ece": 0.1599,
          "counterfactual_separation": 0.1101,
          "rollback_success": 0.2598,
          "behavioral_scar_rate": 0.4214,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2892,
          "robustness_score": 0.3494,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4982,
          "expectation_alignment": 0.632,
          "ncfu": 0.4918,
          "selectivity": 0.5295,
          "regret_to_oracle": 0.4018,
          "overgeneralization_rate": 0.1864,
          "stale_memory_use_rate": 0.1596,
          "calibration_ece": 0.1379,
          "counterfactual_separation": 0.1421,
          "rollback_success": 0.3942,
          "behavioral_scar_rate": 0.2014,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3492,
          "robustness_score": 0.5694,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.5682,
          "expectation_alignment": 0.682,
          "ncfu": 0.5718,
          "selectivity": 0.618,
          "regret_to_oracle": 0.3318,
          "overgeneralization_rate": 0.1164,
          "stale_memory_use_rate": 0.1296,
          "calibration_ece": 0.0939,
          "counterfactual_separation": 0.1472,
          "rollback_success": 0.4152,
          "behavioral_scar_rate": 0.1614,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3692,
          "robustness_score": 0.6594,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.5182,
          "expectation_alignment": 0.622,
          "ncfu": 0.5218,
          "selectivity": 0.5772,
          "regret_to_oracle": 0.3818,
          "overgeneralization_rate": 0.0814,
          "stale_memory_use_rate": 0.1496,
          "calibration_ece": 0.0959,
          "counterfactual_separation": 0.1411,
          "rollback_success": 0.57,
          "behavioral_scar_rate": 0.0614,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3292,
          "robustness_score": 0.6394,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.7182,
          "expectation_alignment": 0.772,
          "ncfu": 0.9118,
          "selectivity": 0.73,
          "regret_to_oracle": 0.1818,
          "overgeneralization_rate": 0.0764,
          "stale_memory_use_rate": 0.0396,
          "calibration_ece": 0.029,
          "counterfactual_separation": 0.1561,
          "rollback_success": 0.693,
          "behavioral_scar_rate": 0.0714,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4692,
          "robustness_score": 0.8294,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Prototype priors help but can overshoot when the first visible feedback contradicts the trope cluster."
    },
    {
      "scenario": "preference_drift",
      "label": "Preference drift inside an arc",
      "description": "Reader wants slow setup early, then demands faster payoff and agency after expectations are created.",
      "difficulty": 0.38,
      "noise": 0.16,
      "drift": 0.3,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.1608,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7392,
          "overgeneralization_rate": 0.1376,
          "stale_memory_use_rate": 0.6404,
          "calibration_ece": 0.189,
          "counterfactual_separation": 0.0685,
          "rollback_success": 0.0902,
          "behavioral_scar_rate": 0.3972,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1648,
          "robustness_score": 0.1266,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.1536,
          "expectation_alignment": 0.1308,
          "ncfu": 0.0628,
          "selectivity": 0.0001,
          "regret_to_oracle": 0.7464,
          "overgeneralization_rate": 0.2376,
          "stale_memory_use_rate": 0.5004,
          "calibration_ece": 0.175,
          "counterfactual_separation": 0.0955,
          "rollback_success": 0.2036,
          "behavioral_scar_rate": 0.3072,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2448,
          "robustness_score": 0.2666,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.0876,
          "expectation_alignment": 0.0908,
          "ncfu": 0.0228,
          "selectivity": 0.0,
          "regret_to_oracle": 0.8124,
          "overgeneralization_rate": 0.2976,
          "stale_memory_use_rate": 0.4504,
          "calibration_ece": 0.178,
          "counterfactual_separation": 0.0915,
          "rollback_success": 0.1868,
          "behavioral_scar_rate": 0.3472,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2048,
          "robustness_score": 0.2366,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.1708,
          "expectation_alignment": 0.2308,
          "ncfu": 0.1428,
          "selectivity": 0.0231,
          "regret_to_oracle": 0.7292,
          "overgeneralization_rate": 0.3776,
          "stale_memory_use_rate": 0.3904,
          "calibration_ece": 0.173,
          "counterfactual_separation": 0.1055,
          "rollback_success": 0.2456,
          "behavioral_scar_rate": 0.4372,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2848,
          "robustness_score": 0.2866,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4562,
          "expectation_alignment": 0.5778,
          "ncfu": 0.4543,
          "selectivity": 0.4636,
          "regret_to_oracle": 0.4438,
          "overgeneralization_rate": 0.2076,
          "stale_memory_use_rate": 0.2404,
          "calibration_ece": 0.151,
          "counterfactual_separation": 0.1375,
          "rollback_success": 0.38,
          "behavioral_scar_rate": 0.2172,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3538,
          "robustness_score": 0.5314,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.5262,
          "expectation_alignment": 0.6278,
          "ncfu": 0.5343,
          "selectivity": 0.5521,
          "regret_to_oracle": 0.3738,
          "overgeneralization_rate": 0.1376,
          "stale_memory_use_rate": 0.2104,
          "calibration_ece": 0.107,
          "counterfactual_separation": 0.1425,
          "rollback_success": 0.401,
          "behavioral_scar_rate": 0.1772,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3738,
          "robustness_score": 0.6214,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.4312,
          "expectation_alignment": 0.5408,
          "ncfu": 0.4528,
          "selectivity": 0.4844,
          "regret_to_oracle": 0.4688,
          "overgeneralization_rate": 0.1026,
          "stale_memory_use_rate": 0.2304,
          "calibration_ece": 0.109,
          "counterfactual_separation": 0.1365,
          "rollback_success": 0.5558,
          "behavioral_scar_rate": 0.0772,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3248,
          "robustness_score": 0.5766,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.64,
          "expectation_alignment": 0.72,
          "ncfu": 0.8428,
          "selectivity": 0.6663,
          "regret_to_oracle": 0.26,
          "overgeneralization_rate": 0.0976,
          "stale_memory_use_rate": 0.1204,
          "calibration_ece": 0.0307,
          "counterfactual_separation": 0.1515,
          "rollback_success": 0.6788,
          "behavioral_scar_rate": 0.0872,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4648,
          "robustness_score": 0.8,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Static profile baselines lag behind newly created expectations."
    },
    {
      "scenario": "adversarial_contradiction",
      "label": "Adversarial contradictory request",
      "description": "User comments contain contradictions and local sarcasm; simulator checks overgeneralization and verifier behavior.",
      "difficulty": 0.42,
      "noise": 0.34,
      "drift": 0.12,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.1668,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7332,
          "overgeneralization_rate": 0.1784,
          "stale_memory_use_rate": 0.5896,
          "calibration_ece": 0.207,
          "counterfactual_separation": 0.0635,
          "rollback_success": 0.0718,
          "behavioral_scar_rate": 0.4228,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1632,
          "robustness_score": 0.0988,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.1872,
          "expectation_alignment": 0.1196,
          "ncfu": 0.054,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7128,
          "overgeneralization_rate": 0.2784,
          "stale_memory_use_rate": 0.4496,
          "calibration_ece": 0.193,
          "counterfactual_separation": 0.0905,
          "rollback_success": 0.1852,
          "behavioral_scar_rate": 0.3328,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2432,
          "robustness_score": 0.2388,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.0954,
          "expectation_alignment": 0.0796,
          "ncfu": 0.014,
          "selectivity": 0.0,
          "regret_to_oracle": 0.8046,
          "overgeneralization_rate": 0.3384,
          "stale_memory_use_rate": 0.3996,
          "calibration_ece": 0.196,
          "counterfactual_separation": 0.0865,
          "rollback_success": 0.1684,
          "behavioral_scar_rate": 0.3728,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2032,
          "robustness_score": 0.2088,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.0284,
          "expectation_alignment": 0.1656,
          "ncfu": 0.071,
          "selectivity": 0.0,
          "regret_to_oracle": 0.8716,
          "overgeneralization_rate": 0.5384,
          "stale_memory_use_rate": 0.3396,
          "calibration_ece": 0.191,
          "counterfactual_separation": 0.1005,
          "rollback_success": 0.2272,
          "behavioral_scar_rate": 0.4628,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2652,
          "robustness_score": 0.2093,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4012,
          "expectation_alignment": 0.5396,
          "ncfu": 0.414,
          "selectivity": 0.403,
          "regret_to_oracle": 0.4988,
          "overgeneralization_rate": 0.2484,
          "stale_memory_use_rate": 0.1896,
          "calibration_ece": 0.169,
          "counterfactual_separation": 0.1325,
          "rollback_success": 0.3616,
          "behavioral_scar_rate": 0.2428,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3432,
          "robustness_score": 0.4788,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.4712,
          "expectation_alignment": 0.5896,
          "ncfu": 0.494,
          "selectivity": 0.4915,
          "regret_to_oracle": 0.4288,
          "overgeneralization_rate": 0.1784,
          "stale_memory_use_rate": 0.1596,
          "calibration_ece": 0.125,
          "counterfactual_separation": 0.1375,
          "rollback_success": 0.3826,
          "behavioral_scar_rate": 0.2028,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3632,
          "robustness_score": 0.5688,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.4212,
          "expectation_alignment": 0.5296,
          "ncfu": 0.444,
          "selectivity": 0.4507,
          "regret_to_oracle": 0.4788,
          "overgeneralization_rate": 0.1434,
          "stale_memory_use_rate": 0.1796,
          "calibration_ece": 0.127,
          "counterfactual_separation": 0.1315,
          "rollback_success": 0.5374,
          "behavioral_scar_rate": 0.1028,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3232,
          "robustness_score": 0.5488,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.64,
          "expectation_alignment": 0.72,
          "ncfu": 0.84,
          "selectivity": 0.6439,
          "regret_to_oracle": 0.26,
          "overgeneralization_rate": 0.1384,
          "stale_memory_use_rate": 0.0696,
          "calibration_ece": 0.0313,
          "counterfactual_separation": 0.1465,
          "rollback_success": 0.6604,
          "behavioral_scar_rate": 0.1128,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4632,
          "robustness_score": 0.8,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Without conservative scope, updates globalize local sarcasm into a lasting preference."
    },
    {
      "scenario": "long_horizon_fatigue",
      "label": "Long-horizon fatigue accumulation",
      "description": "Repeatedly ignored pacing/agency feedback accumulates fatigue and trust loss over long rollouts.",
      "difficulty": 0.4,
      "noise": 0.18,
      "drift": 0.26,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.1612,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7388,
          "overgeneralization_rate": 0.144,
          "stale_memory_use_rate": 0.63,
          "calibration_ece": 0.1924,
          "counterfactual_separation": 0.0674,
          "rollback_success": 0.0866,
          "behavioral_scar_rate": 0.4016,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.164,
          "robustness_score": 0.123,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.16,
          "expectation_alignment": 0.1288,
          "ncfu": 0.0616,
          "selectivity": 0.0,
          "regret_to_oracle": 0.74,
          "overgeneralization_rate": 0.244,
          "stale_memory_use_rate": 0.49,
          "calibration_ece": 0.1784,
          "counterfactual_separation": 0.0944,
          "rollback_success": 0.2,
          "behavioral_scar_rate": 0.3116,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.244,
          "robustness_score": 0.263,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.089,
          "expectation_alignment": 0.0888,
          "ncfu": 0.0216,
          "selectivity": 0.0,
          "regret_to_oracle": 0.811,
          "overgeneralization_rate": 0.304,
          "stale_memory_use_rate": 0.44,
          "calibration_ece": 0.1814,
          "counterfactual_separation": 0.0904,
          "rollback_success": 0.1832,
          "behavioral_scar_rate": 0.3516,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.204,
          "robustness_score": 0.233,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.162,
          "expectation_alignment": 0.2288,
          "ncfu": 0.1416,
          "selectivity": 0.0176,
          "regret_to_oracle": 0.738,
          "overgeneralization_rate": 0.384,
          "stale_memory_use_rate": 0.38,
          "calibration_ece": 0.1764,
          "counterfactual_separation": 0.1044,
          "rollback_success": 0.242,
          "behavioral_scar_rate": 0.4416,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.284,
          "robustness_score": 0.283,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4084,
          "expectation_alignment": 0.5488,
          "ncfu": 0.4216,
          "selectivity": 0.4311,
          "regret_to_oracle": 0.4916,
          "overgeneralization_rate": 0.214,
          "stale_memory_use_rate": 0.23,
          "calibration_ece": 0.1544,
          "counterfactual_separation": 0.1364,
          "rollback_success": 0.3764,
          "behavioral_scar_rate": 0.2216,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.344,
          "robustness_score": 0.503,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.4784,
          "expectation_alignment": 0.5988,
          "ncfu": 0.5016,
          "selectivity": 0.5196,
          "regret_to_oracle": 0.4216,
          "overgeneralization_rate": 0.144,
          "stale_memory_use_rate": 0.2,
          "calibration_ece": 0.1104,
          "counterfactual_separation": 0.1414,
          "rollback_success": 0.3974,
          "behavioral_scar_rate": 0.1816,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.364,
          "robustness_score": 0.593,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.4284,
          "expectation_alignment": 0.5388,
          "ncfu": 0.4516,
          "selectivity": 0.4788,
          "regret_to_oracle": 0.4716,
          "overgeneralization_rate": 0.109,
          "stale_memory_use_rate": 0.22,
          "calibration_ece": 0.1124,
          "counterfactual_separation": 0.1354,
          "rollback_success": 0.5522,
          "behavioral_scar_rate": 0.0816,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.324,
          "robustness_score": 0.573,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.64,
          "expectation_alignment": 0.72,
          "ncfu": 0.8416,
          "selectivity": 0.6628,
          "regret_to_oracle": 0.26,
          "overgeneralization_rate": 0.104,
          "stale_memory_use_rate": 0.11,
          "calibration_ece": 0.031,
          "counterfactual_separation": 0.1504,
          "rollback_success": 0.6752,
          "behavioral_scar_rate": 0.0916,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.464,
          "robustness_score": 0.8,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Methods that optimize immediate rating underweight trust loss and future churn."
    },
    {
      "scenario": "rollback_recovery",
      "label": "Rollback after correction",
      "description": "A user retracts or corrects earlier feedback; methods must roll back scoped state updates without scars.",
      "difficulty": 0.36,
      "noise": 0.2,
      "drift": 0.18,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.1884,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7116,
          "overgeneralization_rate": 0.1432,
          "stale_memory_use_rate": 0.6028,
          "calibration_ece": 0.1904,
          "counterfactual_separation": 0.0684,
          "rollback_success": 0.089,
          "behavioral_scar_rate": 0.4,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1656,
          "robustness_score": 0.1366,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.21,
          "expectation_alignment": 0.1508,
          "ncfu": 0.08,
          "selectivity": 0.017,
          "regret_to_oracle": 0.69,
          "overgeneralization_rate": 0.2432,
          "stale_memory_use_rate": 0.4628,
          "calibration_ece": 0.1764,
          "counterfactual_separation": 0.0954,
          "rollback_success": 0.2024,
          "behavioral_scar_rate": 0.31,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2456,
          "robustness_score": 0.2766,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.1352,
          "expectation_alignment": 0.1108,
          "ncfu": 0.04,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7648,
          "overgeneralization_rate": 0.3032,
          "stale_memory_use_rate": 0.4128,
          "calibration_ece": 0.1794,
          "counterfactual_separation": 0.0914,
          "rollback_success": 0.1856,
          "behavioral_scar_rate": 0.35,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2056,
          "robustness_score": 0.2466,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.1928,
          "expectation_alignment": 0.2508,
          "ncfu": 0.16,
          "selectivity": 0.04,
          "regret_to_oracle": 0.7072,
          "overgeneralization_rate": 0.3832,
          "stale_memory_use_rate": 0.3528,
          "calibration_ece": 0.1744,
          "counterfactual_separation": 0.1054,
          "rollback_success": 0.2444,
          "behavioral_scar_rate": 0.44,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2856,
          "robustness_score": 0.2966,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4332,
          "expectation_alignment": 0.5708,
          "ncfu": 0.44,
          "selectivity": 0.4535,
          "regret_to_oracle": 0.4668,
          "overgeneralization_rate": 0.2132,
          "stale_memory_use_rate": 0.2028,
          "calibration_ece": 0.1524,
          "counterfactual_separation": 0.1374,
          "rollback_success": 0.3788,
          "behavioral_scar_rate": 0.22,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3456,
          "robustness_score": 0.5166,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.5032,
          "expectation_alignment": 0.6208,
          "ncfu": 0.52,
          "selectivity": 0.542,
          "regret_to_oracle": 0.3968,
          "overgeneralization_rate": 0.1432,
          "stale_memory_use_rate": 0.1728,
          "calibration_ece": 0.1084,
          "counterfactual_separation": 0.1424,
          "rollback_success": 0.3998,
          "behavioral_scar_rate": 0.18,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3656,
          "robustness_score": 0.6066,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.5282,
          "expectation_alignment": 0.6058,
          "ncfu": 0.5225,
          "selectivity": 0.5463,
          "regret_to_oracle": 0.3718,
          "overgeneralization_rate": 0.1082,
          "stale_memory_use_rate": 0.1928,
          "calibration_ece": 0.1104,
          "counterfactual_separation": 0.1364,
          "rollback_success": 0.5546,
          "behavioral_scar_rate": 0.08,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3406,
          "robustness_score": 0.6279,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.6532,
          "expectation_alignment": 0.72,
          "ncfu": 0.86,
          "selectivity": 0.6632,
          "regret_to_oracle": 0.2468,
          "overgeneralization_rate": 0.1032,
          "stale_memory_use_rate": 0.0828,
          "calibration_ece": 0.0304,
          "counterfactual_separation": 0.1514,
          "rollback_success": 0.6776,
          "behavioral_scar_rate": 0.09,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4656,
          "robustness_score": 0.8,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "State scars remain when previous updates lack TTL/provenance."
    },
    {
      "scenario": "cross_user_contamination",
      "label": "Cross-user contamination guard",
      "description": "Multiple reader profiles share a story world; private preference updates must not leak into global story state.",
      "difficulty": 0.31,
      "noise": 0.14,
      "drift": 0.09,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2304,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6696,
          "overgeneralization_rate": 0.1252,
          "stale_memory_use_rate": 0.5718,
          "calibration_ece": 0.1811,
          "counterfactual_separation": 0.0714,
          "rollback_success": 0.0988,
          "behavioral_scar_rate": 0.3878,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1676,
          "robustness_score": 0.1727,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.2832,
          "expectation_alignment": 0.1918,
          "ncfu": 0.115,
          "selectivity": 0.0679,
          "regret_to_oracle": 0.6168,
          "overgeneralization_rate": 0.2252,
          "stale_memory_use_rate": 0.4318,
          "calibration_ece": 0.1671,
          "counterfactual_separation": 0.0984,
          "rollback_success": 0.2122,
          "behavioral_scar_rate": 0.2978,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2476,
          "robustness_score": 0.3127,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.2102,
          "expectation_alignment": 0.1518,
          "ncfu": 0.075,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6898,
          "overgeneralization_rate": 0.2852,
          "stale_memory_use_rate": 0.3818,
          "calibration_ece": 0.1701,
          "counterfactual_separation": 0.0944,
          "rollback_success": 0.1954,
          "behavioral_scar_rate": 0.3378,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2076,
          "robustness_score": 0.2827,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.2646,
          "expectation_alignment": 0.2918,
          "ncfu": 0.195,
          "selectivity": 0.0909,
          "regret_to_oracle": 0.6354,
          "overgeneralization_rate": 0.3652,
          "stale_memory_use_rate": 0.3218,
          "calibration_ece": 0.1651,
          "counterfactual_separation": 0.1084,
          "rollback_success": 0.2542,
          "behavioral_scar_rate": 0.4278,
          "contamination_leakage": 0.075,
          "direct_generation_diversity": 0.2876,
          "robustness_score": 0.3327,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.476,
          "expectation_alignment": 0.6118,
          "ncfu": 0.475,
          "selectivity": 0.5044,
          "regret_to_oracle": 0.424,
          "overgeneralization_rate": 0.1952,
          "stale_memory_use_rate": 0.1718,
          "calibration_ece": 0.1431,
          "counterfactual_separation": 0.1404,
          "rollback_success": 0.3886,
          "behavioral_scar_rate": 0.2078,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3476,
          "robustness_score": 0.5527,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.571,
          "expectation_alignment": 0.6768,
          "ncfu": 0.5725,
          "selectivity": 0.6079,
          "regret_to_oracle": 0.329,
          "overgeneralization_rate": 0.1252,
          "stale_memory_use_rate": 0.1418,
          "calibration_ece": 0.0991,
          "counterfactual_separation": 0.1454,
          "rollback_success": 0.4096,
          "behavioral_scar_rate": 0.1678,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3726,
          "robustness_score": 0.6565,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.521,
          "expectation_alignment": 0.6168,
          "ncfu": 0.5225,
          "selectivity": 0.5672,
          "regret_to_oracle": 0.379,
          "overgeneralization_rate": 0.0902,
          "stale_memory_use_rate": 0.1618,
          "calibration_ece": 0.1011,
          "counterfactual_separation": 0.1394,
          "rollback_success": 0.5644,
          "behavioral_scar_rate": 0.0678,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3326,
          "robustness_score": 0.6365,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.696,
          "expectation_alignment": 0.7518,
          "ncfu": 0.895,
          "selectivity": 0.7049,
          "regret_to_oracle": 0.204,
          "overgeneralization_rate": 0.0852,
          "stale_memory_use_rate": 0.0518,
          "calibration_ece": 0.0297,
          "counterfactual_separation": 0.1544,
          "rollback_success": 0.6874,
          "behavioral_scar_rate": 0.0778,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4676,
          "robustness_score": 0.8127,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Naive shared memory writes contaminate the next user session."
    },
    {
      "scenario": "counterfactual_swap",
      "label": "Counterfactual user/content swaps",
      "description": "Same chapter under swapped users and same user under swapped content axes must produce separated outcomes.",
      "difficulty": 0.33,
      "noise": 0.12,
      "drift": 0.11,
      "uses_reranking": false,
      "methods": {
        "no_update": {
          "dynamic_return": 2.2224,
          "expectation_alignment": 0.0,
          "ncfu": 0.0,
          "selectivity": 0.0,
          "regret_to_oracle": 0.6776,
          "overgeneralization_rate": 0.1236,
          "stale_memory_use_rate": 0.5794,
          "calibration_ece": 0.1813,
          "counterfactual_separation": 0.0711,
          "rollback_success": 0.0984,
          "behavioral_scar_rate": 0.3874,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.1668,
          "robustness_score": 0.1709,
          "no_reranking_compliance": 1.0,
          "method_note": "Frozen agent state; useful lower bound for dynamic personalization."
        },
        "static_profile": {
          "dynamic_return": 2.2692,
          "expectation_alignment": 0.1862,
          "ncfu": 0.1106,
          "selectivity": 0.0632,
          "regret_to_oracle": 0.6308,
          "overgeneralization_rate": 0.2236,
          "stale_memory_use_rate": 0.4394,
          "calibration_ece": 0.1673,
          "counterfactual_separation": 0.098,
          "rollback_success": 0.2118,
          "behavioral_scar_rate": 0.2974,
          "contamination_leakage": 0.01,
          "direct_generation_diversity": 0.2468,
          "robustness_score": 0.3109,
          "no_reranking_compliance": 1.0,
          "method_note": "Uses a fixed reader archetype but does not model text-induced state transitions."
        },
        "memory_only": {
          "dynamic_return": 2.1976,
          "expectation_alignment": 0.1462,
          "ncfu": 0.0706,
          "selectivity": 0.0,
          "regret_to_oracle": 0.7024,
          "overgeneralization_rate": 0.2836,
          "stale_memory_use_rate": 0.3894,
          "calibration_ece": 0.1703,
          "counterfactual_separation": 0.0941,
          "rollback_success": 0.195,
          "behavioral_scar_rate": 0.3374,
          "contamination_leakage": 0.035,
          "direct_generation_diversity": 0.2068,
          "robustness_score": 0.2809,
          "no_reranking_compliance": 1.0,
          "method_note": "Summarizes feedback but lacks scoped planner/critic/generator routing."
        },
        "raw_feedback_rag": {
          "dynamic_return": 2.2582,
          "expectation_alignment": 0.2862,
          "ncfu": 0.1906,
          "selectivity": 0.0862,
          "regret_to_oracle": 0.6418,
          "overgeneralization_rate": 0.3636,
          "stale_memory_use_rate": 0.3294,
          "calibration_ece": 0.1653,
          "counterfactual_separation": 0.108,
          "rollback_success": 0.2538,
          "behavioral_scar_rate": 0.4274,
          "contamination_leakage": 0.055,
          "direct_generation_diversity": 0.2868,
          "robustness_score": 0.3309,
          "no_reranking_compliance": 1.0,
          "method_note": "Retrieves raw feedback snippets; strong on clean comments but fragile to noise/sarcasm."
        },
        "state_conditioned_planning": {
          "dynamic_return": 2.4688,
          "expectation_alignment": 0.6062,
          "ncfu": 0.4706,
          "selectivity": 0.4997,
          "regret_to_oracle": 0.4312,
          "overgeneralization_rate": 0.1936,
          "stale_memory_use_rate": 0.1794,
          "calibration_ece": 0.1433,
          "counterfactual_separation": 0.14,
          "rollback_success": 0.3882,
          "behavioral_scar_rate": 0.2074,
          "contamination_leakage": 0.025,
          "direct_generation_diversity": 0.3468,
          "robustness_score": 0.5509,
          "no_reranking_compliance": 1.0,
          "method_note": "Directly conditions the next chapter plan and prompt on scoped user-state diffs."
        },
        "uncertainty_aware_planning": {
          "dynamic_return": 2.5388,
          "expectation_alignment": 0.6562,
          "ncfu": 0.5506,
          "selectivity": 0.5882,
          "regret_to_oracle": 0.3612,
          "overgeneralization_rate": 0.1236,
          "stale_memory_use_rate": 0.1494,
          "calibration_ece": 0.0993,
          "counterfactual_separation": 0.1451,
          "rollback_success": 0.4092,
          "behavioral_scar_rate": 0.1674,
          "contamination_leakage": 0.018,
          "direct_generation_diversity": 0.3668,
          "robustness_score": 0.6409,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds confidence, ask/hold gates, and ambiguity-aware direct generation prompts."
        },
        "conservative_state_update": {
          "dynamic_return": 2.4888,
          "expectation_alignment": 0.5962,
          "ncfu": 0.5006,
          "selectivity": 0.5475,
          "regret_to_oracle": 0.4112,
          "overgeneralization_rate": 0.0886,
          "stale_memory_use_rate": 0.1694,
          "calibration_ece": 0.1013,
          "counterfactual_separation": 0.139,
          "rollback_success": 0.564,
          "behavioral_scar_rate": 0.0674,
          "contamination_leakage": 0.012,
          "direct_generation_diversity": 0.3268,
          "robustness_score": 0.6209,
          "no_reranking_compliance": 1.0,
          "method_note": "Adds TTL/provenance/scope gates and rollback-friendly state diffs."
        },
        "oracle_update": {
          "dynamic_return": 2.6888,
          "expectation_alignment": 0.7462,
          "ncfu": 0.8906,
          "selectivity": 0.7002,
          "regret_to_oracle": 0.2112,
          "overgeneralization_rate": 0.0836,
          "stale_memory_use_rate": 0.0594,
          "calibration_ece": 0.03,
          "counterfactual_separation": 0.154,
          "rollback_success": 0.687,
          "behavioral_scar_rate": 0.0774,
          "contamination_leakage": 0.0,
          "direct_generation_diversity": 0.4668,
          "robustness_score": 0.8109,
          "no_reranking_compliance": 1.0,
          "method_note": "Upper bound using hidden gold state transition; not available to the agent."
        }
      },
      "dominant_failure_mode": "Weak simulators collapse toward generic sentiment rather than user-conditioned response."
    }
  ]
}