src/numeric.rs

This file is internal safe ratio math, marked pub(crate). Your crate denies arithmetic_side_effects and precision-loss casts under clippy pedantic.

So when you turn two counts into a float ratio, you do it through this module to stay lint-clean and panic-free. It is not public API.

Key items

trait RatioInt covers the integer families you use, u64 → f64 and usize → f32.

It also carries lossless narrowing types such as u32 and u16, plus the checked and saturating operations the lint set requires.

unit_ratio(numerator, denominator) -> Float is the one function everything calls.

Behavior notes

A zero denominator returns 0.0, never NaN/inf, which matches the empty-set reading at every call site you make.

Your results are not clamped to [0,1], because overflow and utilization reporting legitimately produces values above 1.0.

Large operands are scaled down through a lossless intermediate so precision survives, which you want. A small numerator over a huge denominator stays strictly positive, and boundary values do not collapse to 1.0.

Your callers include convergence similarity, health scores, context utilization, and memory relevance.

Deep dive: not user-facing, so this page is the whole story.