Your claim goes eight rounds against a council of independent AI models.
They hit it from every angle. You call every decision.
What comes out the other side is yours to defend — with a sealed record proving every hit it took.
You paste your claim and the full source document it came from. FPAW assembles a council of AI models from different provider families, scans the text for fog language and bias before the council ever sees it, then walks you through eight gates — structural review, a technical stress test on every quantitative claim, an independent challenge from a separate model, a mandatory bias assessment, and a calibrated verdict with Monte Carlo exploration. At every gate, you read the output, write your inspector notes, and sign off. Nothing advances without you.
At the end, you have a Falsification Package — a sealed record containing every gate output, every inspector note, every evidence rating, your verdict tier, Monte Carlo sensitivity analysis, and pre-mortem findings. The cryptographic chain proves nothing was altered after signing.
Three or more AI models from different providers stress-test every quantitative assertion. A separate model runs an independent challenge. Disagreements between models are tracked, not averaged.
Every gate requires your inspector notes and sign-off before the next one opens. You adjudicate council splits, rate bias findings, and choose the verdict tier. The AI generates output — you own every decision.
Every sign-off creates a SHA-256 hash of the gate content at that moment. The finished audit is a tamper-evident record — your rationale, your evidence ratings, your verdict, sealed and verifiable.
Pick from three built-in example claims — or paste your own — then run it through five deterministic scans: fog language, bias signals, structural gaps, compound claim detection, and certainty inflation. Everything runs in your browser. Nothing is transmitted.
Gate order is enforced — you cannot skip or reorder them. Each gate must be signed before the next opens. Your sign-off creates a SHA-256 hash of the gate content, forming the tamper-evident audit trail.
Fair warning: A full audit takes 1–2 hours and requires real attention. That's by design. If you want a quick answer, there are faster tools. If you want a defensible one, this is yours.
FPAW is a desktop application. Your data lives in an encrypted vault on your hardware. Nothing is transmitted without your explicit action.
Every audit, gate result, and API key is stored locally in an AES-256-GCM encrypted vault. Your data never touches a server.
13+ AI providers configured as independent reviewers. No single model dominates the analysis. Disagreement is tracked, not hidden.
Every outbound API call is gated by jurisdiction rules. Sensitive material is hard-blocked from leaving your machine.
Sandboxed Electron with context isolation, no runtime eval, and locked-down content security policy.
FPAW beta is a working build, not a preview. Free access. Windows desktop app. Your vault stays yours.