Function validate_fill_fractions 

pub(crate) fn validate_fill_fractions(
    fill: &PixFillConfig,
) -> Result<(), TransportSessionError>

Range-checks the two PixFillConfig fractions, ahead of anything that computes with them.

Extracted from validate_pix_supervision because it is not the first validator to see them. validate_incoming_session_pix_config derives the fill floor a cycle of the widest accepted quota needs, and SessionManager::start calls it before validate_pix_supervision — as does HoprProtocolConfig, where one is a field validator and the other a schema-level one. That derivation runs Duration::mul_f64(finish_fraction), which panics on a NaN, a negative, or a product the monotonic clock cannot represent. A single finish_fraction: .nan in a YAML file would therefore abort the node at load instead of being reported as the configuration error it is; values above one, or a negative loss_margin, would quietly compute a floor lower than the one the cycle needs. Both callers run this first so neither outcome is reachable.

Finiteness is tested before every range, and the ranges are two-sided, for the reason spelled out on max_off_front_share_fraction: every IEEE comparison against NaN is false, so a one-sided test admits it — and a NaN here is not inert. It propagates through the aim point into the required rate, and a NaN rate compares false against every hysteresis threshold, so the planner would hold whatever it last emitted for the life of the Session.