Simulator Suite

A Claude Code plugin that stress-tests business ideas before you commit resources. Two complementary modes, one command.

/simulate

The Problem It Solves

Business ideas get greenlit on conviction rather than signal. Financial models feel rigorous but are really just optimistic assumptions dressed up in a spreadsheet. Customer interviews surface enthusiasm but not the friction that kills adoption after launch. Neither alone is enough — and teams rarely have the time to run both in a way that actually connects.

This plugin runs a probabilistic Monte Carlo simulation and a structured synthetic customer panel against the same concept, then synthesizes where the numbers and the people agree — and where they diverge. Both modes complete automatically. You make one decision at the end.

How It Works

Two modes, run independently or together:

| Mode | Question It Answers | Method | |------|-------------------|--------| | Monte Carlo | What are the probabilities? | 10K+ iterations across correlated variables — KPI distributions, sensitivity rankings, tail risk | | Customer Panel | Will people buy it, and why not? | 5-round structured discussion with synthetic personas — Mom Test discipline, forced rankings, no oracle-trap language |

Run them independently or use /simulate both to produce a synthesis that bridges the two signal types.

Installation

Requirements: Claude Code, Node.js, Python 3

npx simulator-suite-plugin
pip install numpy scipy pandas

Run simulations from the current folder. Open it in Claude Code and type /simulate.

Global install (available in all projects):

npx simulator-suite-plugin --global
pip install numpy scipy pandas

OpenCode

cp -r dist/opencode/.opencode your-project/

GitHub Copilot

cp -r dist/github/.github your-project/

Gemini CLI

cp -r dist/gemini/.gemini your-project/

Codex CLI

cp -r dist/codex/.agents your-project/
mkdir -p your-project/.codex
cp -r dist/codex/.codex/agents your-project/.codex/

Trae

cp -r dist/trae/.trae/skills/* ~/.trae/skills/

Rovo Dev

# Project-specific
cp -r dist/rovo-dev/.rovodev your-project/

# Or global
cp -r dist/rovo-dev/.rovodev/skills/* ~/.rovodev/skills/

Qoder

# Project-specific
cp -r dist/qoder/.qoder your-project/

# Or global
cp -r dist/qoder/.qoder/skills/* ~/.qoder/skills/

Pi

cp -r dist/pi/.pi your-project/

Usage

/simulate                    # Guided mode — asks which mode and gathers inputs
/simulate monte-carlo        # Run Monte Carlo directly
/simulate panel              # Run Customer Panel directly
/simulate both               # Run both + synthesis

All outputs save to simulations/YYYY-MM-DD-[name]/.

What Each Mode Produces

Monte Carlo

| Artifact | What It Contains | |----------|-----------------| | simulation-contract.json | Variables, KPIs, distributions, correlations — the full simulation spec | | simulation.py | Generated Python — shown to you before execution | | executive-report.md | KPI probability distributions, sensitivity rankings, tail risk, actionable takeaways |

Customer Panel

| Artifact | What It Contains | |----------|-----------------| | panel-brief.md | What's being evaluated and why | | personas/ | Individual activated persona files | | panel-transcript.md | Full verbatim 5-round discussion | | panel-results.md | Forced rankings, consensus map, adoption signals, confidence & limitations |

Both Modes Together

Adds synthesis.md — where the quantitative and qualitative signals agree, where they diverge, and what to trust more in each case.

Methodology

Monte Carlo — 7-Step Pipeline

1. Generate Contract JSON — Defines variables, KPIs, distributions, correlations
2. Validate Contract — Structural check before code generation
3. Estimate Distributions — Beta, Normal, Triangular, Lognormal per variable type
4. Generate Python Code — Standalone script shown to user before execution
5. Execute via Harness — harness.py runs the code, enforces caps, returns structured JSON
6. Analyze Results — KPI probabilities, Spearman rank sensitivity, tail risk at 5th/95th percentiles
7. Write Executive Report — Neutral, evidence-based, decision-ready

Correlation method: Gaussian copula. Max 50,000 iterations (harness-enforced). Only numpy, scipy, pandas permitted in generated code.

Customer Panel — 5 Rounds

| Round | Purpose | |-------|---------| | 1. Context Setting | Personas share current challenges — no concepts presented yet | | 2. Individual Reactions | Each persona evaluates each concept independently (Mom Test: no pitching) | | 3. Deep Dive | Pre-mortem, adoption barriers, concept combinations | | 4. Forced Ranking | Every persona ranks all concepts — no ties, trade-off rationale required | | 5. Cross-Panel Discussion | Personas engage each other on disagreements; final verdicts: Advance / Refine / Park / Kill |

Recommended panel: 6 personas (2 Champions, 2 Skeptics, 2 Pragmatists). Minimum: 2.

Guardrails

Monte Carlo:

• Generated code is shown before execution — required, not optional
• Max 50,000 iterations enforced by harness
• Max 2 retries on failure
• Only numpy, scipy, pandas permitted in generated code

Customer Panel:

• Oracle-trap language prohibited: "customers will..." → blocked, use "the panel surfaced..."
• No false consensus — persona agreement must trace to their card attributes
• Confidence & Limitations section required in all results files
• Forced rankings enforce trade-off discipline — no ties permitted

License

MIT