Video analysis was filling the conversation with bookkeeping. A local tool now keeps the raw artifacts on disk and returns compact, inspectable findings.

Game tests kept interrupting my desktop. Separating headless checks from guarded visual captures made the workflow safer and its evidence more reliable.

Walking exposed the gap between accurate foot placement and believable support. A precise solver helped, but cadence and contact tests still rejected the gait.

Aiming, sprinting, crouching, and reloading looked better once I tested the transitions between them and kept the weapon, view, and supporting hands connected.

A rising muzzle did not make our rifle feel heavy. Comparing firing footage helped separate shot onset, recovery, body support, and camera motion.

Starting a private Godot FPS prototype taught me to turn vague complaints about weight into useful observations, careful comparisons, and smaller experiments.

Locking a tip-point-first recipe — pose aim, farthest strong edge, short greens, cyan on midline, magenta clipped to the box — finally made barrel-axis Δ trustworthy.

AABB corners, long Hough rails, and clipped line segments can look like a barrel axis while measuring the sky — here is how the green lines lied.

Dual YOLO gun boxes plus pose skeletons let me associate wrists to weapons before I trust any barrel geometry for recoil angle Δ.

I needed deterministic barrel-axis numbers for game feel, so I turned public firearm YouTube clips into a measurement lab instead of guessing recoil curves.