The Mirror and the Factory: Mastering Shadow vs. Generation Mode

In the world of synthetic data, there is a constant tension between Reality and Requirement. Do you want your test data to perfectly mirror the chaos of production, or do you want it to strictly adhere to the ideal logic of your schema? In Causal Foundry, we solve this by providing two distinct operacional modes: Shadow Mode and Generation Mode.

1. Shadow Mode: The Mirror

In Shadow Mode, Causal Foundry acts as an intelligent observer. It connects to your production database via PostgreSQL CDC (Change Data Capture) and "watches" live events pass by. It does not store this data; instead, it extract the *causal distribution* of those events.

This allows the engine to mutate live streams in real-time. If production data evolves—if customers start buying a new product category or if a specific edge case becomes common—your synthetic stream updates automatically. This is the ultimate defense against Synthetic Data Rot.

2. Generation Mode: The Factory

Generation Mode is a pure manufacturing engine. It doesn't care about what *is* happening in production; it only cares about what *can* happen according to the rules of your StateMap and Schema.

This mode is essential for "Cold Starts" or testing speculative features that don't exist in production yet. If you are building a new loyalty program, you don't have production data to shadow. You must manufacture the entire lifecycle—from enrollment to redemption—from scratch. Generation Mode ensures that every manufactured row obeys 100% of your foreign keys and constraints.

Summary: Which Mode to Deploy?

Regardless of the mode you choose, the Safety Gate remains the same: 100% deterministic redaction and verifiable causal integrity before any data hits a downstream stream.