Implements an inference gate that classifies domain and attaches uncertainty.
// Source: https://github.com/ruvnet/RuView | License: MIT | Copyright 2024 rUv // Provided via justclaw.it.com; keep this notice with the code. //! The inference gate (ADR-302 §2, ADR-300 rule 1). //! //! Every inference passes through the gate. It: //! //! 1. classifies the domain from distance + envelope + signal quality + //! calibration compatibility; //! 2. attaches the model's own predictive **uncertainty** (the fourth ADR-302 //! input), escalating a KNOWN domain to DEGRADED when uncertainty is //! elevated; //! 3. **suppresses the confident class** when the domain is not KNOWN — an //! UNKNOWN domain returns no class, a first-class value rather than a //! confidently-wrong label (ADR-300 rule 1); //! 4. emits a [`RecalibrationRequest`] whenever the state is DEGRADED ... (truncated -- full source via MCP)
See the full source, get the GitHub permalink, and search for more like it.
Get a free API key