Ant3D is a fisheye multi‑camera system for 3D mapping in confined environments, designed and refined over a seven‑year period. This paper has two aims: to present the latest tests reflecting recent system developments and to summarize the design journey and milestones. We compare three successive surveys tests conducted in the same challenging test site: meandering masonry interiors and rock‑cut passages within San Vigilio Castle (Bergamo, Italy). The 2018 test used a low‑cost GoPro multi‑camera rig; the 2020 test used Ant3D v1, a custom prototype improved based on the previous experiences; the 2025 test employed the first industrialized Atom‑Ant3D, integrating SLAM‑based pose estimation, pose‑graph optimization, and SLAM‑aided SfM to accelerate processing. The comparison considers reconstruction drift against a TLS ground‑truth point cloud and processing times. Results show Ant3D v1 achieves high accuracy (max deviation ~3cm) but required extensive processing times (~20h), Atom‑Ant3D offers major time savings (~4.5h) with moderate accuracy trade‑offs, and the GoPro rig performs worst. The discussion outlines design and processing guidelines for reliable, fast mapping of narrow spaces.