decisionlens/mdmp-staff-planning-pairs

mdmp-staff-planning-pairs

Leak-reviewed instruction-tuning pairs for MDMP staff-planning coaching. Public doctrine summaries and fictional scenarios only — no proprietary algorithms, customer data, or classified content.

Disclaimer: Unofficial educational dataset. Not affiliated with the U.S. Army.

Dataset description

324 human-reviewed {instruction, input, output} pairs for fine-tuning a Mistral-7B instruct model on Military Decision-Making Process vocabulary, step boundaries, and coaching responses.

• —GitHub source: dlens/mdmp-assistant
• —Model trained on this data: mistral7b-mdmp-lora

Fields

Buckets

• —mdmp_steps — step identification and ordering
• —step_boundaries — Step 4 vs Step 5 and related traps
• —glossary — term definitions
• —war_gaming — methods, synchronization matrix
• —coa_screening — FASDC, criteria development
• —scenario_coaching — fictional scenario trade-offs

Example

{"instruction": "What MDMP step is war gaming?", "input": "", "output": "War gaming is Step 4, COA Analysis. Staff visualize each COA through critical events and record strengths and weaknesses without head-to-head COA comparison; comparison belongs in Step 5.", "bucket": "mdmp_steps", "source": "mdmp_steps.md", "reviewed": true}

Leak-review policy

Before inclusion, each pair is checked for:

• —Customer names, OPNAV, or real unit designations
• —Proprietary algorithm or product terms
• —Verbatim golden-eval questions (held out in eval/golden_questions.json on GitHub)

See scripts/leak_review.py.

Usage

import json

pairs = []
with open("pairs.jsonl", encoding="utf-8") as f:
    for line in f:
        if line.strip():
            pairs.append(json.loads(line))

Split into train/eval with scripts/split_data.py (85% / 15%).

License

Apache 2.0