Struct FFIMonteCarloResults 

#[repr(C)]
pub struct FFIMonteCarloResults {
    pub ranges: *mut c_double,
    pub impact_velocities: *mut c_double,
    pub impact_positions_x: *mut c_double,
    pub impact_positions_y: *mut c_double,
    pub impact_positions_z: *mut c_double,
    pub num_results: c_int,
    pub mean_range: c_double,
    pub std_dev_range: c_double,
    pub mean_impact_velocity: c_double,
    pub std_dev_impact_velocity: c_double,
    pub hit_probability: c_double,
}

Monte Carlo simulation results, one entry per simulated shot.

Every array holds num_results doubles and all of them are indexed by the same sample number.

The impact_positions_* arrays are DEVIATIONS, not positions. Each triple is the sample’s offset from the baseline point of aim, measured in the target plane (the plane at the requested target distance), in METERS — see crate::MonteCarloResults::impact_positions. A triple of zeros means “exactly on the point of aim”, not “at the muzzle”.

Axes follow FFITrajectoryPoint:

field	axis	use for dispersion?
impact_positions_z	HORIZONTAL / windage deviation, + right	YES — this is the horizontal dispersion axis
impact_positions_y	VERTICAL deviation, + up	YES — this is the vertical dispersion axis
impact_positions_x	downrange, and the target plane sits at a FIXED downrange	NO — never a dispersion axis

Consumers ported from engine 0.13.x (X lateral, Z downrange) commonly read impact_positions_x as the horizontal spread. Under this convention that reads the downrange component instead, which is not scatter, and it fails silently.

Filter the shortfall sentinel before computing any statistic

A sample that never reached the target plane has no deviation to report, so it is encoded as (0, TARGET_NOT_REACHED_SENTINEL_M, 0) — the Y component is crate::TARGET_NOT_REACHED_SENTINEL_M (-1.0e9 meters) — which keeps these arrays the same length as ranges and impact_velocities. The engine does NOT drop those entries for you. A sample is a finite arrival exactly when crate::MonteCarloResults::position_reached_target accepts it: every component finite AND the Y component not equal to the sentinel.

Feeding an unfiltered array into a mean or standard deviation drags the result toward -1e9 and produces a plainly absurd group size. Exclude sentinel samples from dispersion statistics, but keep them in the denominator for hit probability — they are definite misses, which is how hit_probability already counts them.

Fields

ranges: *mut c_double

impact_velocities: *mut c_double

impact_positions_x: *mut c_double

Downrange component of the target-plane deviation, meters. NOT a dispersion axis; see the struct docs.

impact_positions_y: *mut c_double

VERTICAL deviation from the point of aim, meters, positive up — and the field carrying the shortfall marker: crate::TARGET_NOT_REACHED_SENTINEL_M (-1.0e9) marks a sample that did not reach the target plane. Exclude those from dispersion statistics but retain them as misses for probability calculations. See the struct docs.

impact_positions_z: *mut c_double

HORIZONTAL / windage deviation from the point of aim, meters, positive right. This is the horizontal dispersion axis; see the struct docs.

num_results: c_int

mean_range: c_double

std_dev_range: c_double

mean_impact_velocity: c_double

std_dev_impact_velocity: c_double

hit_probability: c_double