PYTHOLOGY INTELLIGENCE · EXTREME WEATHER LAB · EARLY SIGNALS
Don't begin with the storm. Begin with the atmosphere that produces it. We're looking for the changes that happen first, which combinations matter, and whether those early signals can offer useful awareness before severe weather develops. This is open research, not an official warning or forecasting service.
01 / MOORE, OKLAHOMA · 20 MAY 2013 · HISTORICAL REPLAY
The research engine replayed 39 KTLX Level II radar volumes recorded between 18:07 and 20:59 UTC. Its inputs were gated by estimated operational availability, so the replay did not knowingly use later official tornado reports as early evidence. The engine was also developed using this same Moore case, making it a retrospective engineering test, not an independent validation.
EARLIEST PERSISTENT RADAR ROTATION
First persistent rotation at 19:13:35 UTC, about 42.4 minutes before touchdown and 26.4 minutes before the first NWS tornado warning at 19:40 UTC.
This early classification had 0.64 confidence, not a validated tornado probability; the same event was used while building and tuning the analysis engine.
HIGH-CONFIDENCE ROTATION
The first high-confidence rotation classification was emitted at 19:30:41 UTC, 9.3 minutes before the warning and 25.3 minutes before touchdown, with confidence 0.72.
Recognising rotating storm structure does not mean accurately forecasting tornado formation or issuing a public warning.
HIGH-CONFIDENCE FUSED STORM STRUCTURE
The fused tornado-supportive-structure hypothesis did not reach 0.7 confidence until 19:47:47 UTC — 7.8 minutes after the NWS warning and 8.2 minutes before touchdown.
That negative result prevents us claiming the full detector demonstrated a 26-minute warning advantage.
REPRODUCIBILITY · ENGINE RECORD
The pinned Moore baseline processed 68 engine inputs and recorded 168 finding revisions. Revisions preserve how evolving storm evidence changed classifications over time.
Missing scans, reduced radar resolution and absent soundings produced different timings; halved azimuthal resolution was particularly damaging.
All timestamps are UTC. This is a retrospective replay of a known tornado, not a live or prospective warning test. The model's early rotation signal is not equivalent to its later fused hypothesis.
02 / ONGOING OBSERVATIONS · READ-ONLY PUBLIC RESEARCH SUMMARY
The existing native Rust weather backend can poll NOAA NEXRAD Level II radar volumes at configured U.S. sites, official National Weather Service alerts and National Hurricane Center active cyclone advisories. The first operational radar trial was KTLX in Oklahoma. A separate Prometheus research watcher can review qualifying live radar finding revisions and preserve timestamped solar observations as context.
LATEST VERIFIED PUBLIC WEATHER RESEARCH SNAPSHOT
WHAT THE SOURCES ARE ACTUALLY SHOWING
Awaiting verified NHC advisory feed…
Awaiting verified NWS alert feed…
Awaiting fresh radar evidence…
These are current-source observations and official advisory fixes, not predicted storm paths, new tornado warnings or evidence of independent forecasting skill. If no active systems are listed, the corresponding source must first be verified fresh; otherwise the display reports the gap.
The source-level badges must be verified before counts are described as current. NWS warnings and NHC advisories originate from official agencies, not Weather Lab. This public projection does not expose investigation APIs, bearer tokens, Slack records or private radar archives. If unavailable it shows unverified, never fictitious zeroes.
Selected U.S. radar volumes are decoded, gridded and analysed for rotating storm structure. No worldwide tornado-radar coverage or independently evaluated forecast skill is implied.
Official U.S. warning polygons and tropical advisory fixes are time-stamped context. Current NHC positions and headings are not Weather Lab's predicted cyclone tracks.
Solar Lab observations can be attached using data actually available before the storm observation. Solar forcing of tornadoes or tropical cyclones has not been demonstrated by this work.
03 / PROMETHEUS · EXTREME-WEATHER PRECURSOR LAB · GLOBAL + AOTEAROA
Separate from the NEXRAD radar research above, this experimental screen looks for combinations and changes in global environmental conditions, with a dedicated New Zealand observational lens. Counts are transparent research snapshots, not forecasts; old and unavailable readings are labelled instead of silently presented as current.
Patterns in global environmental conditions, with separate New Zealand observational screening. These are research candidates, not weather warnings or validated predictions.
— NZ exploratory observations
Checking publication date…Research only; no validated weather forecast.
Review validation priorities →04 / HURRICANE MICHAEL · 9–11 OCTOBER 2018 REPLAY · 10 OCTOBER 2018 LANDFALL
The 2018 Hurricane Michael experiment combines observed radar, advisory fixes and GOES infrared context with a separately labelled, seeded model reconstruction for radar-unsampled inner-core regions. The interface distinguishes observed/derived layers from reconstructed estimates and their uncertainty.
The rapid-intensification precursor did not offer demonstrated lead over the truth onset in the evaluated replay; the first relevant revision appeared around 30 hours after a later intensification onset. The first land-interaction finding came 30 minutes after evaluated landfall. Live NHC advisory fixes are context only and do not yet drive a live tropical prediction engine.
05 / NEXT TESTS · SCIENTIFIC AND OPERATIONAL
Collect genuinely unseen storm episodes. Predefine target events and matching criteria, then measure false alarms, misses, robustness, source gaps and reproducible availability-time lead.
Add NEXRAD stations and independent regional weather agencies only where source provenance and operational access are verified. Western-Pacific typhoons require separate JMA/JTWC/RSMC integrations.
Investigate any solar-weather statistical relationship with quiet-weather controls, confounder checks and pre-registered analysis. Do not use solar observations to raise weather hazard scores in the current system.
