Per-Geyser Narrow Classifiers
Dedicated classifiers for individual named geysers (Steamboat, Daisy, Riverside, Beehive). Hit rates on small samples; too few to call skill.
What it does
Individual classifiers trained per named geyser, each consuming the geyser's local FDSN tremor signature + basin-wide multi-physics context. Regular geysers are easier to catch than Old Faithful's all-eruptions record, so read their counts as a record, not as skill.
Physics basis
Each geyser has a distinctive pre-eruptive tremor signature, fluid-pressure cycle, and local ground-motion pattern, which a classifier trained on that one geyser can learn.
When it fires
Continuous; auto-train daily evening on GeyserTimes.
What the customer receives
- Per-geyser eruption classification + lead-time prediction
- Confidence + waveform excerpt per fire
- Monthly hit-rate summary per geyser
Operational numbers (live)
Engagement paths
Per-query, subscription, territorial-exclusive, and royalty-on-find structures are all available. Specific commercial terms are scoped after a technical-fit conversation.
Honest caveats
- Small samples: a handful of eruptions per geyser per month, so a perfect count is not yet statistically meaningful.
- Each classifier is per-geyser; it does not transfer to a newly-monitored geyser without retraining.
- Ground-truth denominator is GeyserTimes community observations; a missed user report becomes a missed classification.
Michael Jessop: michael@deepmapai.com · partner portal