ADR-0004: Use f32 for core process and tuning values

Status: Accepted

Context

OPC DA analog values commonly use single-precision representations, and replayable calculations need one consistent numeric width from sampled values through the tuning math. The safety guide also treats f32 precision as part of validating usable values before a live operation.

Decision

The core MRFT model, process ranges, and tuning calculations use f32 for process values and PID tuning numerics. Parsing and validation occur at the boundaries where external values enter the application.

Consequences

The engine's precision matches common analog-tag representations without widening and narrowing values at each calculation step. Range checks, actuation tolerances, and result-validity checks must account for single-precision limits.