In a perspective view, a world-space proximity test rejects the target the player is aiming at

Fact

When a 2D pointer aims into a 3D scene, the aim is a ray from the camera, and the cursor's world position is only one point on it — wherever you chose to put it. A hit test that measures world distance from that point therefore only works at that one depth.

A target further down the ray sits directly under the crosshair on screen while being metres away in world space, so a sphere around the cursor refuses the shot the player just lined up perfectly. Widening the sphere does not fix it: it buys depth tolerance by also accepting things the player is visibly not pointing at.

The test that matches what the player sees is angular — perpendicular distance to the ray, against a radius that grows with distance:

const t = dot(target - camera, rayDir);            // depth along the ray
const perp = length((target - camera) - rayDir*t); // off-axis distance
const hit = perp <= (radiusAtCursor / cursorDepth) * t;

Why it matters

The complaint is "it misses when I'm clearly on it", which sounds like sloppy collision or a lag problem, and both of those are worth ruling out first — so the depth-dependence goes unnoticed while near targets keep working fine.

The inverse costs more. A cone that runs the depth of the scene is a long tube, and everything in it is "aimed at", including whatever happens to be far behind the target and too small to notice. If some of those are hazards, the player is punished for a mistake they could not see.

How to apply

• Use the angular test for aiming, and keep the cone's half-angle equal to the on-screen size of the reticle — that is the promise the crosshair makes.
• Resolve a swing or shot to the single best-aimed candidate, ranked by off-axis distance relative to each candidate's own allowance, rather than taking everything the cone contains.
• Bound the cone's depth to the range the player can actually read.
• Test with a target at the far plane, not at the cursor's own depth; that is the only place the two models disagree.

Related: [[framing-a-box-is-a-per-corner-solve]].