LeanOS Core

Autonomous business strategy for founders. One agent builds and stress-tests your strategy before you commit to building.

How It Works

You describe a problem space. The system researches, constructs hypotheses, and attacks its own conclusions. What survives is your strategy.

Governor -> provides problem space, constraints, values
Strategist -> researches (WebSearch/WebFetch)
           -> constructs hypotheses (compression model)
           -> self-challenges (adversarial protocol)
           -> writes strategy/hypotheses.md
           -> escalates only when it cannot decide
Governor -> responds to escalations, provides ground-truth evidence

The output is a hypothesis register — four validated hypotheses plus a solution design — ready to feed the Sell & Grow execution chain.

See It Running

Watch the demo — one command, regulation in, business model out.

The Hypothesis Register

One file. Four hypotheses. Each earns its confidence.

Hypothesis	What It Answers
Problem	What problem exists, for whom, and how acute?
Segment	Who has it badly enough to pay? Observable filters, not demographics.
Value Proposition	Why will your solution win? Functional, emotional, and social jobs.
Unit Economics	Does the business model sustain itself? LTV:CAC, payback, margins.

Plus a Solution Design section (growth architecture, feature map, MVP scope, growth loops) derived from the four hypotheses.

Confidence States

Every hypothesis must earn its confidence through evidence:

State	Meaning
`UNVALIDATED`	Stated, no qualifying evidence
`RESEARCHED`	System did autonomous research (T1/T2 evidence), awaiting ground truth
`SUPPORTED`	Governor provided customer conversations or test results that validate
`BROKEN`	Evidence contradicts the claim

Epistemic Tiers

Every claim carries its tier — no false confidence:

Tier	What It Means	Example
T1	Derivable from public data	Market size from census + industry reports
T2	Synthesized hypothesis	"This segment likely has this problem because..."
T3	Requires ground truth	"Customers will pay $X for this"

The Strategist

One agent. Three modes. No orchestrator — you invoke it directly.

Mode	When	What Happens
BUILD	First use	Autonomous research, hypothesis construction via compression (enumerate possibilities, eliminate, what survives is the strategy), adversarial destruction pass, complete register
CHALLENGE	Default	Re-evaluate existing register with fresh research, test evidence quality, run destruction on changed hypotheses, update confidence
REVIEW	Readiness check	Full challenge pass plus cross-hypothesis consistency, Sell & Grow readiness evaluation

What It Does Autonomously

Market research (TAM/SAM/SOM from public data)
Competitive mapping (features, pricing, positioning gaps)
Problem scoring (frequency, severity, breadth, alternatives' inadequacy)
Hypothesis construction via compression model
Adversarial self-challenge (assumption extraction, pre-mortem, red-team, constraint inversion)
Evidence quality assessment and overconfidence detection

What It Escalates to You

Mode selection (VENTURE / BOOTSTRAP / HYBRID)
UVP and unfair advantage approval
Tradeoff preferences when options have quantified consequences
Ground-truth gaps that block strategy (load-bearing T3 assumptions)

Strategy Dashboard

A visual command center for the hypothesis register. Parses the markdown and presents six views:

View	Question It Answers
Readiness	Am I Sell & Grow ready? What's blocking?
Evidence Quality	Where is my evidence weak? What's T3 that should be T1?
Risk Map	Which assumptions will kill the strategy if wrong?
Destruction	What did the adversarial self-challenge find?
Solution	What does the solution look like? Features, MVP, growth loops.
Hypothesis Detail	Full deep-dive on any hypothesis (click through).

Run the Dashboard

cd tools/dashboard
npm install
npx vite

The dashboard reads strategy/hypotheses.md directly from the project root. No file copying needed.

Sync Script (Optional)

Pre-generates a JSON file from the register for faster dashboard loading:

./tools/scripts/sync-register.sh

Runs the TypeScript parser against strategy/hypotheses.md, writes tools/dashboard/public/register.json. The dashboard tries JSON first, falls back to live markdown parsing.

File Structure

strategy/
  hypotheses.md              The hypothesis register (source of truth)

execution/
  queue/                     Escalations awaiting governor response

.claude/
  agents/
    strategist.md            The autonomous strategist agent

  skills/
    stg-sizing-markets/      Market opportunity (TAM/SAM/SOM)
    stg-segmenting-customers/ Customer segments with observable filters
    stg-scoring-problems/    Problem validation (frequency/severity/breadth)
    stg-analyzing-competition/ Competitive landscape and red-team input
    stg-designing-pricing/   Value-based pricing methodology
    stg-calculating-economics/ Unit economics (LTV/CAC/payback/margins)
    stg-designing-solutions/ Growth architecture, MVP scope, feature map
    stg-extracting-insights/ Process governor-provided expert sources

tools/
  dashboard/                 Strategy visualization dashboard (React)
  scripts/
    sync-register.sh         Parse markdown register to JSON

Quick Start

# Clone
git clone https://github.com/BellaBe/leanos my-project
cd my-project

# Run the strategist in BUILD mode
claude --agent strategist

# View the dashboard
cd tools/dashboard && npm install && npx vite

Requires Claude Code CLI. The system uses the filesystem for all communication — no external services, no database, no API keys beyond Claude.

Sell & Grow Interface

The register feeds directly into the Sell & Grow execution chain. The chain proceeds when all four hypotheses reach SUPPORTED:

Chain Step	Consumes From Register
Strategic Positioning	VP hypothesis (claim + jobs), Segment hypothesis (target customer)
Revenue Architecture	Solution Design (growth architecture, MVP, growth loops), Unit Economics (pricing)
Negotiation	VP (value anchor), Unit Economics (pricing structure)
Content Distribution	VP (positioning), Solution Design (demo formats)
Channel Intelligence	Segment (where they gather), Unit Economics (CAC benchmarks)

License

MIT. Use it, fork it, build on it.