# DeltaBook

> DeltaBook (deltabook.ai) is an analysis workbench: describe a hypothesis in plain English, a
> deterministic engine tests it against point-in-time historical data, and the result comes back
> with its sample size, uncertainty interval, and full provenance. AI assists with building and
> interpreting; it never computes the numbers. DeltaBook reports what the data says - it never
> recommends an action.

Key rules:
- Every probability is computed by a deterministic engine (same inputs, same number). No LLM
  output is ever stored as a prediction, backtest number, or record entry.
- Point-in-time discipline: an analysis only sees data knowable before the moment it predicts,
  enforced in the query layer on dual timestamps.
- Nothing runs until the user confirms; the AI proposes, the user approves, the engine computes.
- The forward record is append-only and database-enforced: locked before the event, graded by
  the platform, never edited.
- Calibration is the headline metric; win rate is shown and labeled the least meaningful number.
- Pricing: $49/month or $499/year, card at signup, no free trial. Unlimited deterministic
  backtesting included. AI work meters as jobs on the user's confirm; scoping chat is free;
  everything already computed is free to read forever.

Docs:
- [How the engine works](https://deltabook.ai/docs.md): deterministic compute, the copilot
  boundary, the confirm gate, the fact firewall.
- [The data rules](https://deltabook.ai/docs.md): point-in-time enforcement, sources,
  provenance, verified vs unverified substrate.
- [The record](https://deltabook.ai/docs.md): locking, grading, DECLARE, what survives
  cancellation.
- [The meter](https://deltabook.ai/docs.md): what costs a credit and what never does.
- [Pricing](https://deltabook.ai/pricing): the full feature table.
- [Attribution](https://deltabook.ai/attribution): data source licences and credits.

Optional:
- [Full AI knowledge base](https://deltabook.ai/llms-full.txt): the complete rules in one file.