Hypothesis Validator

A multi-agent hypothesis validation system using Claude Agent SDK that stress-tests business ideas across 5 dimensions: Market, Technical, Financial, Competitive, and Devil's Advocate.

Features

• Multi-dimensional analysis: Validates hypotheses across 5 critical dimensions
• Evidence-based: Every claim must have a source URL (enforced via hooks)
• Iterative deep-dives: Automatically investigates low-confidence areas
• Market-first weighting: Market dimension gets 35% weight in final scoring
• Structured reports: Generates comprehensive markdown reports with verdicts

Installation

# Clone the repository
git clone https://github.com/Sarangk90/hypothesis-validator.git
cd hypothesis-validator

# Install with pip
pip install -e .

# Or with poetry
poetry install

Configuration

1. Copy the example environment file:

   cp .env.example .env

2. Add your Anthropic API key to .env:

   ANTHROPIC_API_KEY=your_actual_api_key
   

Usage

Command Line

# Validate a hypothesis from a string
python -m hypothesis_validator "Your hypothesis here"

# Validate from a file
python -m hypothesis_validator hypothesis.txt

# With options
python -m hypothesis_validator hypothesis.txt \
    --max-iterations 5 \
    --confidence 0.8 \
    --output report.md

Python API

import asyncio
from hypothesis_validator import HypothesisValidator

async def main():
    validator = HypothesisValidator()
    report, state = await validator.run("""
        Product: AI-powered code review tool
        Target: Enterprise development teams
        Problem: Code reviews are slow and inconsistent
    """)
    print(report)

asyncio.run(main())

Validation Dimensions

Dimension	Weight	Focus Areas
Market	35%	TAM/SAM/SOM, pain points, timing, demand signals
Technical	20%	Feasibility, architecture, MVP timeline
Financial	15%	Pricing, unit economics, funding requirements
Competitive	15%	Direct/indirect competitors, defensibility
Devil's Advocate	15%	Assumption attacks, failure modes (inverted)

Verdict Types

• STRONG_GO: Weighted score ≥7.0 AND devil's advocate ≤4.0
• CONDITIONAL_GO: Weighted score ≥6.0 AND devil's advocate ≤6.0
• NEEDS_MORE_DATA: Weighted score ≥5.0 OR many unresolved gaps
• WEAK_NO: Weighted score ≥3.5
• STRONG_NO: Weighted score <3.5

Project Structure

hypothesis-validator/
├── src/hypothesis_validator/
│   ├── __init__.py
│   ├── main.py              # CLI entry point
│   ├── orchestrator.py      # Main orchestration logic
│   ├── state.py             # Pydantic models
│   ├── tools.py             # Custom MCP tools
│   ├── agents/              # Validator agents
│   ├── prompts/             # Agent prompts
│   ├── hooks/               # Quality validation hooks
│   └── report/              # Report generator
├── examples/
│   └── example_hypothesis.py  # Example usage
├── outputs/                   # Generated reports
└── tests/

License

MIT License