Keel

Trident converts the planner's waypoints into a local frame. t.geo.carto.geodesicInverse solves the leg on the WGS 84 ellipsoid and returns its bearing and length. Leg 4 runs 2,402 m east-northeast at 120 m altitude. Its end lies 2,265 m east and 799 m north of its start.

const f64 wgs84SemiMajor = 6378137.0

f64 radians(f64 degrees) {
    return degrees * t.math.pi() / 180.0
}

# East, up and north of the leg's end, in meters from its start.
dvec3 localEnd(f64 lat1, f64 lon1, f64 lat2, f64 lon2, f64 altitude) {
    f64 flattening = 1.0 / 298.257223563
    f64[4] solved = t.geo.carto.geodesicInverse(wgs84SemiMajor, flattening,
        radians(lat1), radians(lon1), radians(lat2), radians(lon2), 100)
    f64 azimuth = solved[2]
    f64 meters = solved[3]
    return dvec3(meters * t.math.sin(azimuth), altitude, meters * t.math.cos(azimuth))
}

A Keel kernel measures the closest approach between the leg and mast T-12, modeled as a 135 m vertical segment. t.collide.segmentClosest returns the distance and the closest point on each segment. It knows nothing about drones, masts or the 150 m rule.

f64 closestApproach(dvec3 legStart, dvec3 legEnd, dvec3 mastBase, dvec3 mastTop) {
    chest closest = t.collide.segmentClosest(legStart, legEnd, mastBase, mastTop)
    f64 distance = closest.distance
    return distance
}

f64 legFourAtT12() {
    dvec3 legStart = (0.0, 120.0, 0.0)
    dvec3 legFinish = (2264.98, 120.0, 799.23)
    return closestApproach(legStart, legFinish,
        (1250.0, 0.0, 573.0), (1250.0, 135.0, 573.0))    # 124.40
}

The decision stays in Trident. The clearance is a named constant, and the verdict names the leg, the mast and the distance. Leg 4 passes T-12 at 124 m, 57 percent of the way along, so the program holds the route.

const f64 clearanceMeters = 150.0

# The rule lives here, in source, and nowhere in the kernel.
text legVerdict(text leg, text mast, f64 closest) {
    if (closest < clearanceMeters) {
        f64 shown = t.math.round(closest)
        return "hold " + leg + ": " + mast + " at " + text(shown) + " m"
    }
    return "clear " + leg
}

Bad input refuses by name. When a GPS dropout writes NaN into a waypoint, the kernel refuses with collide_nonfinite and publishes nothing. Without that refusal, a NaN distance fails every comparison, so closest < clearanceMeters reads false and the leg would pass as clear.

text checkLeg(dvec3 legStart, dvec3 legEnd, dvec3 mastBase, dvec3 mastTop) {
    text outcome = "unchecked"
    task {
        chest closest = t.collide.segmentClosest(legStart, legEnd, mastBase, mastTop)
        f64 distance = closest.distance
        outcome = "closest " + text(distance) + " m"
    } else (e) {
        text why = e.message
        outcome = "refused: " + why
    }
    return outcome
}