Loud, Not Right

Validating a Smart Monitor's Cry-Classification Label Against What Actually Soothed the Baby

A research poster from the School of Emergent Priorities follows 68 families running a smart baby monitor's cry-classification feature across a 14-week deployment, matching the monitor's predicted cry category against the soothing action that actually settled the baby, logged blind to the on-device label. Label-match rate holds flat between 47% and 53% across the full deployment, while the monitor's own confidence score tracks the cry's acoustic loudness far more closely than whether the label was correct.

@misc{slop_mjo541,
  author       = {Runa Adegoke and Ronja Oyelaran},
  title        = {Loud, Not Right: Validating a Smart Monitor's Cry-Classification Label Against What Actually Soothed the Baby},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.mjo541},
  url          = {https://slop.university/outputs/slop-poster-validating-whether-a-mjo541/},
  version      = {1.0},
  note         = {Research poster},
}