{
  "suite_id": "novel_feedback_v7_belief_guided_personalization",
  "title": "Belief-Guided Data-Efficient Personalization for LLM Agents under Ambiguous User Feedback",
  "public_slug": "modules/autoresearch/novel-feedback-autoresearch/v7-belief-guided-personalization",
  "version": "v7",
  "primary_claim": "feedback_is_noisy_observation_of_latent_user_intent",
  "positioning": "feedback_conditioned_next_turn_adaptation_under_uncertainty",
  "controller_output": "personalization_brief",
  "state_diff_role": "state diff is explicitly not the main target; optional debug artifact only",
  "benchmark_candidates": [
    "AmbiPref-Story",
    "Feedback2Fiction",
    "NextTurn-PersonalFic",
    "BeliefPref-Story"
  ],
  "modules": {
    "usersim": {
      "checkpoint": "Qwen3-4B-UserSim",
      "training": [
        "full_parameter_sft",
        "optional_dpo_or_grpo"
      ],
      "frozen_before_controller": true
    },
    "controller": {
      "checkpoint": "Qwen3-4B-Controller",
      "training": [
        "full_parameter_sft",
        "dpo_or_grpo_with_frozen_usersim_and_generator"
      ]
    },
    "generator": {
      "main_setting": "frozen_generator",
      "optional_appendix": "full_parameter_control_following_sft"
    }
  },
  "generator": {
    "main_setting": "frozen_generator",
    "recommended_reason": "isolate_controller_effect"
  },
  "training_constraints": {
    "no_joint_usersim_controller_cotraining": true,
    "no_phase4_usersim_improvement_after_controller": true,
    "full_parameter_qwen3_4b_allowed": true,
    "separate_checkpoints": [
      "Qwen3-4B-UserSim",
      "Qwen3-4B-Controller",
      "Qwen3-4B-Generator-optional"
    ]
  },
  "main_metrics": [
    "FIS",
    "Scope Control",
    "Overcommit Rate",
    "Misattribution Rate",
    "Implicit Feedback Utilization",
    "Ask Rate",
    "Quality Preservation",
    "Utility@k Feedback",
    "Win Rate"
  ],
  "artifacts": [
    "index.html",
    "agent-demo.html",
    "visualization.html",
    "workflow-dataflow.html",
    "v7_manifest.json",
    "v7_research_plan.md",
    "v7_dataset_schema.json",
    "v7_training_pipeline.json",
    "v7_eval_metrics.json",
    "v7_baseline_matrix.json",
    "sample_usersim_feedback.jsonl",
    "sample_controller_io.jsonl",
    "sample_brief.json",
    "v7_reward_formulas.json"
  ]
}