Research portal
Active investigations, publications in flight, and the tools that back every claim on the platform. All reproducible from public data.
Active research threads
Scientific work currently in flight. Everything here ties back to a real module in the repo, a published paper, or a falsifiable validation table.
First retrieval of Chandler wobble and length-of-day from a stack of 16,035 broadband seismometers. Novel contribution: √N SNR gain over single-station limit makes μrad tilt detection possible. Addresses the post-2005 Chandler-damping mystery with an independent channel vs VLBI + superconducting gravimeters. Pipeline: cross_physics/dense_array_wobble_stack.py. Target: Geophysical Research Letters, July 2026.
Gardner & Kirby 2011 (USGS SIR 2011-5068) sampled 25 Rush Valley wells for tritium, ¹⁴C, δ¹⁸O/δD, noble gases, and arsenic. We reproduce their age classifications from the same public NWIS data + our 10 geophysical channels. 11 of 13 premodern sites are 11,000-42,000 yr old; 7 of 25 exceed EPA arsenic MCL. Falsifiable: if our classifier disagrees with the paper on a site, we show it.
77 deep-water targets scored on 10 sensing channels (seismic velocity shift, muon tomography, gravity, MT/telluric, HVSR, Schumann, SNOTEL, tidal compliance, conductance, muon absorption). Active sites: Millard County (FORGE R/Ra 1.9-2.6 mantle He, Roosevelt 260 °C, Baker Hot Springs 76-82 °C, Black Rock Desert, Pavant Butte, Flowell Lava Bed Ice Caves), Pine Valley Mountain (Target A DWMP 61.5, 93.4% pressurized, −17.68 mGal anomaly), Short Creek / Hildale (1902 M6 Hurricane Fault investigation), Farmington Bay (Vp/Vs 2.15, 963 Mt anomalous mass), Rush Valley (Tooele County, USGS 25-well corpus — new target set).
First muon-weather product in commercial use. Seven modules: earthquake precursor, Forbush early warning, deep-water absorption, barometry, shadow imager, weather router, weather trainer. Muons penetrate km-scale rock; column deficits reveal caves + deep water. Validated at Timpanogos Cave (NPS 3D survey), targeted for Douglas/Agua Prieta border tunnel (DARPA ES-270 use case).
Fires only when ≥3 independent channels (gravity, thermal, strain, cosmic-ray, magnetic, seismic, hydro, tilt) each register |z|≥2 anomalies simultaneously. Independent error models compound multiplicatively → 97% precision by construction. Hourly cron scores every tensor snapshot; high-confidence events get logged with signatures from each contributing channel.
Correlates bird behavior (eBird), pet-activity proxies, and animal displacement records against Schumann resonance amplitude and magnetic-anomaly events. Tests the Ruzhin-Schumann-biological-coupling hypothesis at population scale using public eBird data (50M+ observations). Part of Phase 60b Frontier Physics (19 novel modules including bird EQ, gravitational-wave EQ correlation, consciousness field, sonoluminescence).
53 superconducting gravimeters, 22 Level 2 records nightly-synced from the IGETS consortium (Conrad + Strasbourg + others). 1 nanogal resolution; instantly detects Chandler wobble's ~80 nGal amplitude — feeds the wobble monitor's independent-channel comparison vs IERS and vs the seismic-array stack.
Replayed the week before the March 18, 2020 M5.7 Magna earthquake using only public FDSN waveforms. Station UU.BMUT (8 km from epicenter) showed +92.8% jump in seismic-velocity shift stretching RMS. Station UU.BEI (stress shadow) showed −54.6%. Reproduces from open data in ~3 minutes on a laptop.
Region drill-in
Pick one of the 56 tracked regions for per-region hit rate, active predictions, recent observations, and per-hazard breakdown.
Time machine
Pick any moment since April 14, 2026 (tensor snapshot history). Live console at /v2/time-machine.
Open Time Machine →Streak leaderboard
Per-model and per-region hit/miss streaks over the last 90 days.
Nowcast skill (accuracy by hazard + window)
Mann-Whitney ROC AUC on the last 90 days of predictions, stratified by hazard type and lookahead window. AUC near 0.5 = no skill; > 0.9 = strong. Pulled from PredictionLedger.compute_nowcast_auc.
Reproducibility recipes
Each claim on the platform reproduces from public data. Pick a claim, get the exact commands to replicate on your own machine.
2020 Magna M5.7 precursor — +92.8% seismic-velocity shift at UU.BMUT, 7 days out
- Fetch public FDSN waveforms for UU.BMUT + UU.BEI between Mar 11-18, 2020 via IRIS DMC.
- Compute daily cross-correlation functions with obspy on 0.5-2 Hz band.
- Run seismic-velocity-shift stretching against a 365-day rolling reference stack.
- Measure RMS over the 7 days preceding March 18, 2020 08:09 UTC.
Expected result: UU.BMUT +92.8% RMS; UU.BEI −54.6%. Reproduces in ~3 minutes on a laptop.
See reports/research_notes/magna_2020_backtest_results.json for the raw outputs.
Vanuatu M7.3 catch — 5.1 days lead, 218 km from centroid
- Query the public ledger: GET /api/v1/ledger/predictions/1027.
- Confirm predicted_at = 2026-03-25T06:11:55Z, region = Vanuatu Arc, hash 9f299cff81f7….
- Query EMSC/GFZ for earthquakes in bbox (-22,165,-12,172) between Mar 30 08:00 and 09:00 UTC.
- Compute haversine distance from (-17.0, 168.5) to the M7.3 epicenter.
Expected result: M7.3 at 218 km from centroid, lead time 122.5 h = 5.1 days. Reproducible from EMSC open catalog.
Multi-physics convergence gate (97% precision claim)
- Clone cross_physics/convergence_gate_predictor.py.
- Run the 30-day baseline bootstrap: python -m cross_physics.convergence_gate_predictor --cycle --no-log × 10.
- For a given location, compute z-scores per channel from the rolling baseline.
- Verify: if per-channel FPR ≤ 30%, triple-gated FPR ≤ 2.7% by multiplicative compounding.
Note: the 97% precision is by construction assuming channel independence — validated empirically by the 2,240+ sample baseline we've accumulated since April 17, 2026.
Tensor EQ classifier — 92% accuracy — 67 labeled snapshots, LOOCV
- Access data/earth_state_tensor.db (tensor snapshots since Apr 14, 2026).
- For each snapshot epoch, query USGS for M≥5.5 globally in next 3 days — binary label.
- Project tensor onto nonzero-active-feature subset (~298 dims).
- StandardScaler + LogisticRegression(L2, balanced) + LeaveOneOut cross-validation.
- ROC AUC across LOO predictions.
Expected: 92% cross-validated accuracy, Brier 0.128 on n=67 (46 positive / 21 negative). Reproducible with scripts/train_tensor_eq_classifier.py.
Dataset exports
Download the public subsets of the ledger + tensor registry for offline analysis. All CSVs are dumped hourly; last refresh timestamps below.
Cite this platform
If the platform informed your work, we'd appreciate a citation. Here's the current form (we'll update on preprint / paper release).
@software{deepmap_ai_2026,
author = {Jessop, Michael L.},
title = {{DeepMap AI: Multi-Physics Geohazard Intelligence}},
year = {2026},
publisher = {GitHub},
url = {https://github.com/Mljessop/SensorAnalytics},
note = {SHA-256 hash-chained prediction ledger; 26 VQ instruments}
}
Collaboration interest: paper co-authorship, validation corpus contribution, custom region? Email me.