PYTHOLOGY INTELLIGENCE · SUBTERRANEAN LAB · EARLY SIGNALS
Don't begin with the earthquake or eruption. Begin with the changes in the earth that come before it. We're studying which early patterns matter, which are ordinary noise, and whether any precursor signals can offer reliable awareness of what may happen next. We share the misses as well as the matches; these are not yet validated forecasts.
LIVE SCREENING RECORD · PUBLIC, READ-ONLY
The independent world-wide shadow screen reports catalogue patterns around tectonic and volcano-context locations. Screens, null results and source freshness are not proof of predictive ability.
Independent hourly screening of earthquake-catalogue patterns and known volcanic settings, with evidence limits kept visible.
Catalogue screening is not an earthquake or eruption prediction.
Review the latest screen ↓LATEST HOURLY SCREEN · READ-ONLY RESEARCH FEED
Connecting to the curated research record…
This is a display of the independent hourly watch's latest saved research scan, not a separate collector or a continuous live seismometer feed. No private logs, tokens or Slack message history are exposed. An unavailable or old scan is labelled explicitly. USGS global catalogue completeness varies; research observations are not public warnings or validated predictions.
10 October 2026 · first archived experiments
Selected historical replays, not prospective forecasts. Matching a known event's preceding catalogue does not establish prediction skill or false-alarm rates.
EARTHQUAKE DATES · Christchurch · Kaikōura · Fiordland
Multi-signal classifications occurred in 5 of 33 Christchurch cells, 3 of 31 Kaikōura cells, and 4 of 33 Fiordland cells. These are spatial counts, not independent predictions.
ERUPTION DATES · Nabro · Whakaari · La Palma
Seismic-catalogue patterns were identified in all three eruption-selected historic windows. That does not demonstrate volcano-specific sensing.
REPLAY WINDOW · – · no specific future event forecast
Panama showed some anomaly-classified checkpoints but no multi-signal cells under the tested rule. It is a useful negative result from the selected replay, not a formal quiet-region control or proof of safety.
Read the Panama study ↗The global prototype examines earthquake catalogues, with volcano location context, not a worldwide network measuring gas, tremor, deformation or thermal changes.
SUBTERRANEAN · GLOBAL VOLCANIC RESEARCH · READ-ONLY SHADOW
Smithsonian historical volcano and eruption records are joined to available USGS official notice metadata and nearby earthquakes. They are separate types of evidence. NZ Volcano Pulse remains its own detailed regional study.
Global volcanic research figures appear only after a successfully collected source-checked snapshot is published. Missing data is not interpreted as quiet volcanic activity.
— volcanoes with one or more catalogue earthquakes within 50 km. This counts geographic proximity only and does not establish volcanic seismicity.
— individually verified continuous volcanic sensor streams in this global collector; sensor availability remains unassessed.
— listed volcanoes with a recent elevated USGS notice. USGS HANS coverage is not worldwide.
— volcanic precursors independently validated by this new global collector.
Future work: per-volcano observing histories, deformation, thermal and gas feeds, eruption-free controls, monitoring gaps and prospective false-alarm tests.
Last successful snapshot: Not yet published. Sources: Smithsonian GVP · USGS HANS · USGS earthquake feed. Only aggregate research coverage is published, not private raw measurements, credentials or operator logs. For safety decisions follow official volcanic observatories and civil-protection authorities.
SUBTERRANEAN · AOTEAROA NEW ZEALAND · VOLCANIC + GEOTHERMAL PRECURSORS
Rotorua is part of the Taupō Volcanic Zone (TVZ) — a remarkable belt of volcanic and geothermal systems. We're studying the environmental conditions around these systems, asking which changes reflect everyday geothermal behaviour and which combinations deserve closer investigation. The same questions are compared with volcanoes worldwide.
Detailed research scope: Rotorua's geothermal fields, the Ōkataina volcanic centre and Tarawera, the Taupō volcanic system, and the wider Bay of Plenty volcanic setting, including Whakaari / White Island.
Hydrothermal and geothermal changes are not automatically signs of magma movement or an impending eruption. Each field and volcano needs its own baseline.
Compare the TVZ against Ruapehu and Tongariro / Ngāuruhoe, Taranaki Maunga and the Auckland volcanic field. These systems have different histories, plumbing, and monitoring coverage.
The aim is to test whether an apparent signal is local, shared by similar volcanic systems, or merely common background variation.
Where official observations are available and authorised for reuse, compare earthquake depth and clustering, volcanic tremor, ground deformation, gas, surface temperature, and geothermal water chemistry or levels.
Account for rain, groundwater, seasonal effects, sensor changes, missing data and human activity before treating a change as volcanic.
Compare periods before unrest with quiet periods at the same site, then against similar and different volcanoes around the world. Freeze the screening rules before future tests, retain false alarms and non-detections, and check whether observations were actually available at the time. This NZ programme describes a research focus, not a verified continuous geothermal sensor network or a validated eruption prediction capability. GeoNet / GNS Science and New Zealand emergency authorities remain authoritative for volcanic hazard information.
01 / HISTORICAL EARTHQUAKE CATALOGUE EXPERIMENTS
The same seismic signal core and thresholds were replayed over known historic earthquake periods with revised GeoNet catalogue coverage. The output is a set of multi-signal classifications in spatial cells, not a list of predicted earthquakes. The revision changed results — useful evidence about the method's sensitivity to catalogue inputs.
Actual mainshock: (New Zealand local date). Replay first regional multi-signal: 12 February 2011, 00:00 UTC.
601 merged catalogue events; 199 unique GeoNet events after merging.
First regional multi-signal classification: , near -43.800, 172.258.
The near-Canterbury pattern persisted with GeoNet added, but one cell logged 11 qualifying six-hour checkpoints versus 23 previously. Both duration and geography changed.
Actual mainshock: (New Zealand local date). Replay first regional multi-signal: 5 November 2016, 06:00 UTC.
338 merged events; 109 unique GeoNet contributions.
First regional multi-signal classification: , near -40.851, 174.050.
Critical negative finding: the previously reported near-epicentral cell's short signal did not survive the V3 catalogue change.
Actual mainshock: (New Zealand local date). Replay first regional multi-signal: 1 July 2009, 06:00 UTC — distant from the epicentre.
358 merged events; 191 unique GeoNet contributions.
First regional multi-signal classification: , near -44.151, 168.925.
The first regional multi-signal cell lay away from the eventual mainshock. A closer cell qualified later, which is why regional occurrence and target proximity must be scored separately.
All three time windows were chosen because a significant earthquake was already known to have occurred. Historical catalogues may contain later revisions that would not have been available to an observer at the time. Multiple six-hour checkpoints in the same cell are not independent detections. These results do not establish a true prospective lead time, false-alarm rate or prediction accuracy.
02 / ERUPTION-SELECTED SEISMIC CATALOGUE REPLAYS
These experiments applied an earthquake-catalogue screen around places known to have erupted. They did not ingest continuous volcanic tremor, gas emissions, GNSS or InSAR deformation, or satellite thermal measurements. Nearby earthquakes may be tectonic.
Actual eruption: 12–13 June 2011; inferred onset around 20:27–20:42 UTC on 12 June. Replay multi-signal: 12 June, 18:00 UTC.
107 catalogue events merged. First reported regional multi-signal classification at .
The recorded classification preceded a literature-inferred eruption onset by approximately 2.5 hours. Each of the four qualifying cells had only one checkpoint, and the centre cell did not qualify.
Actual eruption: (New Zealand local date). The replay starts with multi-signal already present on 10 November, not a verified onset.
360 catalogue events merged. Multi-signal was already present at the first replay checkpoint on ; a true onset cannot be inferred.
The volcano's centre cell never reached multi-signal. Many qualifying cells were in the wider Bay of Plenty — this was not a successful volcano-localised eruption signal.
Actual eruption: . Replay first regional multi-signal: 12 September 2021, 03:00 UTC.
1,451 catalogue events merged. The first reported regional multi-signal state occurred on , roughly seven days before the eruption.
The nine qualifying cells overlap geographically, so they are not nine independent successes. Later-reviewed data availability and quiet-period controls remain unresolved.
Finding an unusual earthquake-catalogue pattern in a region selected after a known eruption is not equivalent to predicting an eruption. To evaluate genuine skill we need eruption-free volcano controls, tectonic comparisons, data-as-available-at-the-time, false-positive rates and an independently frozen prospective protocol.
03 / A NEGATIVE RESULT WE KEEP
A retrospective regional replay near 7.80° N, 81.00° W covered late September to early October 2026. Several cells were classified as anomalous at some checkpoints, but none met the multi-signal threshold. Negative findings belong in the public record too.
This experiment is a non-detection for this chosen region, rule and catalogue sample. It is not a properly designed quiet-period control and does not establish safety, future earthquake risk or a meaningful forecasting skill score.
04 / METHOD · PROVENANCE · DEFINITIONS
These public notes are condensed from labelled operator console transcripts and research summaries archived on 10 October 2026. Some cell-summary tables were reconstructed from printed console output; the original full-fidelity time series and local audit records have not all been independently rechecked or published here.
| Term | Plain-language meaning | What it does not mean |
|---|---|---|
| Catalogue event | An earthquake recorded in a public source catalogue, subject to coverage and revision. | A complete or uniform account of all earthquakes, everywhere. |
| Spatial cell | A geographical area over which the screening rules examine preceding catalogue patterns. | An independent sensor, confirmed hazard zone or prediction target. |
| Multi-signal | Several configured earthquake-catalogue pattern rules qualified at a sampled checkpoint. | Proof that an earthquake or eruption is imminent. |
| “Foreshocks” label | Historical name for the screening core's recent local event-count condition. | A scientifically identified foreshock before a mainshock is known. |
| Replay checkpoint | A sampled point in a known historical period, typically six hours apart. | A separate independent prediction or proof of historic publication-time availability. |
| Shadow research watch | A scheduled catalogue-screening prototype that records timestamped findings and negative scans. | A public safety alert system or validated forecast product. |
Historical tests used public earthquake catalogues from GeoNet, USGS, EMSC and ISC in differing combinations. The newer worldwide shadow scanner uses the USGS all-month catalogue and location context from the Smithsonian Global Volcanism Program. USGS HANS is independent, official U.S.-only volcano-status context.
Detailed machine-readable experiment outputs, run journals and private watcher state remain in the research workspace. This page deliberately publishes reviewed summaries and scientific boundaries only. It does not expose credentials, private alert text, raw operational files or a public link into the working code repository.
05 / NEXT PHASE · PROSPECTIVE RESEARCH
The current worldwide programme is an earthquake-catalogue shadow scanner with volcano-location context. A scientifically useful next phase will require more than finding interesting examples in the past.
Preserve exact code and thresholds before measuring future outcomes. Retain all scanned cells, negative results, source outages and original observation times.
Use eruption-free volcanoes, tectonic regions and event-free time windows. Measure the background rate of the same patterns, including false alarms and missed events.
Separate earthquake origin times from when a catalogue actually published each record. Track revisions, regional detection completeness and cross-provider de-duplication.
Integrate station tremor/RSAM, deformation, gas and thermal evidence source by source. Keep volcanic and tectonic hypotheses separate until independent evidence supports a mechanism.
These experiments do not establish earthquake or volcanic eruption forecasting skill. No finding on this page is an official warning or a declaration that a location is safe or unsafe. Follow GeoNet, GNS Science, NEMA and the relevant national geological or volcanological authorities for public safety information.
