Pick (or create) the folder that holds your well image archive. Images are filed as Well Name/<date>/photo.jpg.
Each agent scans the photo pairs for its failure mode. Confirm or dismiss findings below after each report — every verdict is saved to _routeforge/labels.json as training data for the learned models that will replace these heuristics.
python thermal-service.py alongside this page to read true °C from the M30T's R-JPEGs
via DJI's Thermal SDK. Without it, the IR agent falls back to relative hotspot contrast.
Compares each well's latest flight against the previous one — photo by photo, camera by camera — and flags wells that look different.