[P2] trimpToStrain diverges for denominator at the logarithm boundary #36

Summary

Swift and Kotlin return materially different results when trimpToStrain receives a positive TRIMP with denominator == 1.

This was found by the Strain differential rollout tracked in #17.

Minimal reproduction

trimp = 1.0
denominator = 1.0

Observed behavior

The Swift parity runner rejects the non-finite result. Kotlin succeeds with a very large finite value.

Likely cause

Neither implementation validates the denominator before evaluating the logarithmic formula. The intermediate result becomes infinite at this boundary:

• Swift preserves +Inf.
• Kotlin applies roundToLong(), which saturates before division by 100.

The normal default denominator (7201) is unaffected.

Impact

A public non-default calibration parameter can produce either a rejected non-finite result or an extreme finite strain value depending on platform.

Severity: P2.

Acceptance criteria

• Define the valid denominator domain consistently on both platforms.
• Invalid boundary inputs fail or normalize identically.
• Add mirrored deterministic Swift and Kotlin regression tests for 1, values below 1, and a valid denominator.
• Re-run the Strain differential corpus.

No hardware testing is required; this is pure analytics behavior.