The product thesis is narrow. Capture provider requests, convert them into structured risk analysis, and translate the result into short human-readable guidance before signature.
This request does not move funds now, but it opens an allowance path that can be reused later. The correct product behavior is not to warn abstractly, but to show exposure, rationale, and a safer alternative in one screen.
The architecture rejects the idea that a model should inspect raw calldata and decide everything on its own. Provider capture, transaction interpretation, and risk scoring are deterministic layers. The model receives structured analysis and emits the shortest explanation that can change user behavior. This separation is the product moat because models can be replaced, but captured transaction semantics and user safety logs accumulate locally.
The value event is a transaction decision, not generic session time. The right pricing shape is a free starter quota followed by usage-based charging per analysis event. That aligns cost, inference, and user value without forcing a subscription before trust exists.
The immediate validation surface is a browser extension because signature flows already converge there. The SDK path remains the natural second stage once the explanation and risk schema prove that they actually alter approval behavior.
The live page is intentionally simple. Its purpose is not marketing volume but operational proof: the concept is coherent, the UI exists, the extension shell renders from structured data, and the Cloudflare deployment path is real.
