Papajams/body-debt-finetune-dataset

Body Debt Fine-Tuning Dataset

Structured recovery prescription dataset for fine-tuning small language models on the Body Debt 4-agent health coaching pipeline.

Dataset Description

• —Domain: Healthcare / physiological recovery
• —Base model: Qwen3-1.7B-Instruct
• —Total examples: 12,800 (3,000 train + 200 test per agent × 4 agents)
• —Generation method: Deterministic scoring engine + template-based labels
• —License: Apache 2.0

Structure

The dataset contains 4 sub-tasks, one per QVAC pipeline agent:

Each JSONL line is a chat-formatted example:

{
  "messages": [
    {"role": "system", "content": "<agent system prompt>"},
    {"role": "user", "content": "<formatted input>"},
    {"role": "assistant", "content": "<deterministic label>"}
  ]
}

Generation Pipeline

1. 1.Profile sampling: Stressor profiles sampled from a constrained parameter space (alcohol type/count, training area/intensity, sleep hours, stress, illness, care actions).

1. 1.Deterministic scoring: Each profile is scored by the Body Debt 5-system engine (scoring.py), producing cardiovascular, brain, liver, muscular, and gut scores (0-100).

1. 1.Label generation: Deterministic template functions produce the ground-truth output for each agent:
2. 2.Triage: _fallback_plan() logic (system ranking + avoid mapping)
3. 3.Coach: _fallback_advice() logic (severity-tiered prescriptions)
4. 4.Schedule: generate_schedule() (top-4 systems → time blocks)
5. 5.Reflection: apply_voice() (deterministic voice transformation)

1. 1.Chat formatting: Each example is wrapped in the exact system prompt used by the QVAC pipeline at inference time, ensuring train/inference prompt alignment.

Reproducibility

python generate_finetune_dataset.py --n-train 3000 --n-test 200 --seed 42

Intended Use

Fine-tuning small language models (1-2B parameters) for structured health recovery coaching. NOT for medical diagnosis or treatment recommendations. The deterministic labels are physiologically grounded but simplified — they are training targets, not clinical guidelines.

Citation

If you use this dataset, cite the Body Debt project and the AutoScientist Challenge submission.