Reachability, collapse rate, adversarial survival. Verified by a deterministic engine, confirmed by you before any verdict.
●Sandbox mode. Full engine. Try it with sample data or your own CSV. The complete decision engine runs; nothing is stored and your data auto-deletes at session end.
Your casesFull mode
1situation
2read + validate
3confirm
4verdict
5audit
!
1. The situation
Describe the question in plain words, then bring the data. The agent proposes a structural reading; you review it before anything is solved.
— or bring your own —
Drop CSV files here, or click to browse
Sandbox: held only for this session, auto-deleted. Never tracked.
2. How we read your data
The agent proposes a structural reading of your data and a plain-English interpretation. The engine's deterministic validator (/v1/validate) gates the structure. This is what you sign off on.
!
No model is configured, so this is a placeholder, not a real reading of your data.
The reading below is a structurally valid stub that does not map your CSV. Do not sign off on it.
Configure a model (AZURE_OPENAI_KEY or ANTHROPIC_API_KEY) and re-map, or hand-author the schema.
What changed after your correction
Describe what to adjust, then re-map. The reading is deterministic, so the same input gives the same result -- a correction here steers it precisely. Every correction is saved to the audit trail.
Human semantic gate
3. Confirm the semantics
The validator checks structure, not meaning. You own the meaning. Nothing solves until you sign off that the reading above matches your real-world situation.
4. Structural verdict
Routed deterministically over the engine surface. The verdict is computed by the engine, never by the agent.
5. The audit
The engine's own verdict, laid into a branded structural audit. Presentation only, it never re-derives or softens the verdict.