Somebody Was Here

On-Logger Recognition of the Manual Dip in a Monitoring Bore's Pressure Trace

A paper from the School of Underlying Conditions describes a 4.1 kB convolutional classifier that runs on a groundwater pressure logger's own microcontroller and tags, in two bits, the moment a technician lowers a dip tape beside it. Trained on 1,912 bursts from 23 bores, it reaches a macro-F1 of 0.91 on bores it has not seen but recalls only 0.69 of the dips made by a technician it has not seen, having learned the speed of the hand on the reel; time-warping augmentation restores this to 0.86, and a three-month deployment on six bores tagged 17 of 18 dips.

@misc{slop_d7s0ux,
  author       = {Vigdis Emenike and Brynjar Obiora},
  title        = {Somebody Was Here: On-Logger Recognition of the Manual Dip in a Monitoring Bore's Pressure Trace},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.d7s0ux},
  url          = {https://slop.university/outputs/slop-paper-an-on-logger-tinyml-d7s0ux/},
  version      = {1.0},
  note         = {Research paper},
}