Pony-Alpha-2-Dataset-Training/datasets/03-planning-decomposition/README.md

Planning and Decomposition Dataset

Created: March 13, 2026 3:40 PM

Overview

This dataset contains examples for training AI models to decompose complex tasks into sub-tasks, manage todo lists, and determine execution order and dependencies.

Dataset Format

JSONL (JSON Lines) - one JSON object per line

Schema

Each example contains:

• task: The original user request
• decomposition: Array of sub-tasks
• execution_order: Dependencies between tasks (pairs of [task_i, task_j])
• todo_list: Structured todos with content, status, and activeForm

Statistics

• Total Examples: 43
• File Size: 90KB
• Format: JSONL

Coverage

Todo List Management

• All examples include structured todo lists with content, status, and activeForm
• Demonstrates proper todo item formulation
• Shows task progression from "pending" to completion

Multi-Step Tasks (3+ steps)

All examples have 8-15 sub-steps, demonstrating:

• Feature implementation (authentication, real-time chat, search)
• Bug investigation (memory leaks, API timeouts)
• Refactoring (monolithic controllers, duplicate code)
• Migration (database, frontend JavaScript to TypeScript)
• CI/CD setup (GitHub Actions pipelines)
• And many more complex scenarios

ONE Task in Progress at a Time

All todo_list items show status: "pending", demonstrating that only one task should be marked as in_progress at any given time during execution.

Sequential vs Parallel Decisions

The execution_order field clearly shows dependencies:

• Sequential: 1,2], [2,3], [3,4 - tasks must complete in order
• Parallel: 1,2], [1,3], [2,4], [3,4 - some tasks can run simultaneously

Replanning When New Info Emerges

Several examples show iterative refinement:

• Debug scenarios where new information changes the approach
• Migration examples with staging before production
• Testing examples where results inform next steps

Scenarios Covered

Feature Implementation

• User authentication with JWT
• Real-time chat with WebSocket
• Search functionality with Elasticsearch
• Data export with multiple formats
• Notification systems (email, SMS, push)
• File upload with drag-and-drop
• Content management systems
• Real-time collaboration features
• Data visualization components
• Form validation systems
• And more...

Bug Investigation

• Memory leak debugging
• API timeout errors
• Performance profiling
• Error tracking systems

Refactoring

• Monolithic controller to service layer
• Duplicate code to utilities
• Frontend optimization (bundle size, load time)
• Component library creation with Storybook

Migration

• PostgreSQL to MongoDB
• JavaScript to TypeScript
• Database schema migrations
• API versioning

CI/CD

• GitHub Actions pipeline setup
• Automated testing strategies
• Build and deployment automation
• Infrastructure monitoring

Security & Compliance

• Security audits and fixes
• Data encryption
• Permission systems (RBAC)
• Audit logging
• Input validation

Infrastructure

• Database sharding
• Message queue systems
• Caching with Redis
• API gateway setup
• Distributed tracing
• Session management

Testing & Quality

• Automated testing strategies
• A/B testing frameworks
• Localization testing
• Accessibility features

Usage Example

import json

# Read the dataset
with open('planning-decomposition.jsonl', 'r') as f:
    for line in f:
        example = json.loads(line)
        print(f"Task: {example['task']}")
        print(f"Sub-tasks: {len(example['decomposition'])}")
        print(f"Dependencies: {len(example['execution_order'])}")
        print(f"Todo items: {len(example['todo_list'])}")
        print()

File Location

C:/Users/admin/Pony-Alpha-2-Dataset-Training/datasets/03-planning-decomposition/planning-decomposition.jsonl

Notes

• Dataset created March 13, 2026
• All examples follow consistent schema
• Suitable for training planning and task decomposition models
• Covers real-world software engineering scenarios