Anyone can prompt a model to play a CISO and get back one confident, agreeable opinion. Príncipe pressure-tests your security product, pitch, or decision against a panel of 30–200 synthetic CISOs — engineered to disagree the way a real room does, and calibrated against real ones. You get a decision, the objections that block it, and an answer that's honest about its own confidence. That's the difference between asking a model and measuring an idea.
Ask a real question in plain language. A composition of synthetic CISOs answers it from their own seat. Then the answer is corrected for the known per-question-type bias and sized with an honest confidence band.
Variable-N, 30–200. Each CISO is built from independent axes — region, industry, size, background, and disposition (stance · posture · AI-posture · mandate) — so the panel splits where real CISOs split.
The question is classified, its framing corrected, and the result run through a calibration map learned from real CISO surveys. We measured mean error fall 47 → 18 points, step by step.
Not a vibe and a number — a stance, the ranked objections that block it, the most-opposed segment, and a statistical read on whether the panel was even the right shape for the question.
Where a question type isn't calibrated yet, the answer comes back marked directional — wide band, objections first, no false precision. Trustworthy where it's confident because it's honest where it isn't.
A signed knowledge feed updates the panel daily — fresh breaches, regulation, and vendor moves, tagged by region and industry so each persona reacts to the world that touches their seat.
Self-hosted in Docker, AGPL-3.0, bring your own Anthropic key. No telemetry, no accounts, no data leaving your machine. The whole engine is open and checkable.
Whether you're a founder about to spend a year of runway hunting for a hundred real CISOs, a VC stress-testing an investment thesis without ever sitting across from the buyer, or a security leader betting your next roadmap on an unverified hunch — Príncipe lets you measure the idea from a hundred different angles in an afternoon.
The name comes from a small volcanic island off the west coast of Africa. In May 1919 the astrophysicist Arthur Eddington sailed to Príncipe to watch the sky go dark — and for a few minutes of total eclipse he photographed the stars sitting just behind the sun, measuring how far they had shifted. He didn't prove Einstein right with conviction. He proved it by measuring a deflection against a known baseline, from the one place on earth you could actually see it. That's the whole idea here. We won't tell you whether your idea is right — we'll show you where the sky shifts, and where it doesn't. It's the only useful definition of validation we know.
No slides. A real security question goes in; a calibrated panel of 100 synthetic CISOs answers from their own seats; out comes a stance, the segments that split, the objections that block it, and an honest read on confidence — recorded live against a self-hosted instance.
Recorded live against a self-hosted Príncipe instance — real run, real output (compute wait trimmed for length).
Runs entirely on your machine in Docker. Bring your own Anthropic key. Your questions never leave your infrastructure.
Príncipe is open source under AGPL-3.0, so it behaves identically everywhere. One command installs Docker if you need it, clones the repo, and boots the stack — Postgres, the Bayesian statistician, and the web app. The only thing you bring is your Anthropic API key, pasted into the first-run wizard. Pick your OS:
Installs Docker Desktop via Homebrew (asks first), then boots.
macOS 13+. Already have Docker? It skips straight to booting.
Installs Docker via the official get.docker.com script (asks first).
Any modern distro with bash + systemd.
In PowerShell. Sets up WSL2 + Docker Desktop (one reboot, resumes itself).
Windows 10 (2004+) or 11 with winget.
install.sh (macOS / Linux) or install.ps1 (Windows) in the repo first.
No telemetry, no accounts, no phone-home. Questions and panels live only on your machine.
You pay Anthropic directly for inference. No middleman, no markup, full cost transparency.
The whole engine — personas, calibration, statistics — is in the repo. Verify every claim.