An independent research platform built from scratch: five private machines, a 563 GB geophysical archive with records reaching back to 1700, and a full development record of how it was made.
Privately owned · locally hosted · no cloud dependencyTeslaQuake was built by one person, self-funded, in evenings and early mornings — working alongside AI as a development partner. This is the honest sequence, including the result that forced the program to change direction.
The starting point was the primary document, not the folklore: Tesla's Colorado Springs Notes 1899–1900 — roughly 716,000 characters — read end to end and reduced to a dated ledger of every passage where Tesla described stationary waves, nodal points and Earth currents, alongside the 1978 editorial commentary.
Five machines were assembled and networked into a private research studio: an always-on collector, a data-processing node, two GPU workstations and a development box. Every model runs locally. Nothing about the research depends on an outside API remaining available, affordable or unmonitored.
Every collector, parser, dashboard and statistical routine in this platform was designed and written in working sessions with Claude (Anthropic) — used as an engineering partner rather than an autocomplete: architecture reviewed before code was written, code read back and challenged, statistics checked against a second method, and every build signed off against a written nine-step verification procedure requiring a clean console and reproducible output.
Local open-weight models on the private machines carry the offline work — transcription, document reading and routine parsing — so the archive can be processed without sending research data anywhere.
Rather than start a fresh dataset, the project went after everything the world's institutions had already published and made it queryable in one place: 563 GB across 107,919 files in 13 categories, with continuous series reaching back through the twentieth century and, in the case of sunspot number, to 1700.
Each category was pulled from its institutional source, normalised to common units and timestamps, and checked for coverage gaps — which are published on this page rather than hidden.
Collectors run continuously on the always-on nodes, with a live feed-health view that reports the last successful pull for every source, so a silently dead feed shows up as a status line instead of a hole discovered months later.
Welford, CUSUM and EWMA run side by side on each data stream, so sudden spikes, sustained regime shifts and slow drift are each caught by a method suited to them. Detected anomalies are stored with their scores and remain queryable.
The platform was originally built to test whether geomagnetic and solar data carried a detectable seismic signal. A formal statistical audit of that question — run against the archive rather than against selected cases — found no such signal.
That result is reported here as plainly as a positive one would have been. An instrument that only ever confirms its builder is not an instrument.
Following the audit, TeslaQuake was redirected away from forecasting entirely. The program now does three things: keep a public observation log, maintain and open up the historical archive, and pursue Tesla's nodal and stationary-wave work as a documented question in the history and physics of science — with instrumentation and education, not prediction, as the output.
The infrastructure, the archive and the tooling are already built and paid for. What a sponsor funds is not construction — it is independent review, instrumentation, and publication of an archive that currently exists on five machines in one building.
Five machines split into two systems: always-on collection and analysis, and design/development workstations. Network addresses, service endpoints and access configuration are deliberately not published.
| Node | Class | GPU / VRAM | Role |
|---|---|---|---|
| Collector | Linux server | 8 GB | Always-on archive host, retrieval and orchestration |
| Processor | Linux workstation | 16 GB | Continuous feed collection, space-weather and magnetometer processing |
| GPU node | Windows workstation | 24 GB | Vision models, rendering, heavy local inference |
| Command | Windows workstation | 24 GB | Development, dashboards, day-to-day operations |
| Dev box | Linux workstation | 12 GB | CAD, 3D and code development |
| Total | 5 nodes | 84 GB | 17 local models · no cloud inference |
563 GB across 107,919 files, drawn from institutional sources and normalised into one queryable archive. Coverage gaps are listed rather than smoothed over.
| Category | Code | Size | Files | Coverage | Gaps |
|---|---|---|---|---|---|
| Ionospheric | I | 215.5 GB | 22,916 | 1950–2026 | No gaps in window |
| Atmospheric | A | 140.8 GB | 5,178 | 1934–2026 | No gaps in window |
| Bioacoustic | B | 75.8 GB | 10,289 | 2015–2025 | 37 gap years |
| Solar / Geomagnetic | W | 58.1 GB | 20,085 | 1963–2026 | No gaps in window |
| Schumann Resonance | R | 35.3 GB | 36,091 | 2000–2026 | 20 gap years (1980–1999) |
| Telluric | T | 28.2 GB | 5,941 | 2005–2026 | 31 gap years |
| Seismic | S | 1.2 GB | 3,547 | 1990–2026 | No gaps in window |
| Events / GDACS | E | 1.1 GB | 457 | 2010–2026 | 30 gap years |
| Ocean / Tidal | O | 400.5 MB | 1,175 | 1909–2025 | 35 gap years |
| Celestial | Z | 197.5 MB | 461 | 1900–2026 | 40 gap years |
| Gravity | G | 118.1 MB | 21 | In collection | Emerging category |
| Animal Behaviour | AN | 7.6 GB | 1,716 | In collection | Emerging category |
| Volcanic | V | 6.0 MB | 42 | 2026 | 46 gap years |
| Total | 13 | 563 GB | 107,919 | 1700–2026 | Published as-is |
The archive's reach is its most unusual asset. Sunspot number runs from 1700, the official planetary K index from 1932, atmospheric series from 1934, tide-gauge records from 1909, ionospheric data from 1950 and the disturbance-storm index from 1957 — alongside 358,000 historical earthquakes at M4.0+ over 25 years. Very little of it had previously been assembled in one place, in one time base, by one project.
Three complementary statistical methods run at once on each stream. Each catches a different shape of departure from baseline; none of them is used to forecast anything.
Online streaming computation of mean and variance per metric, weekday and month. Flags statistical spikes at z-score thresholds of 1.5, 2.0, 3.0 and 4.0, with a minimum of 10 samples before it will report anything.
Two-sided cumulative sum for regime-shift detection. Identifies sustained directional drift that never produces a single dramatic spike and would therefore slip past a threshold detector entirely.
Exponentially weighted moving average at α = 0.2, balancing reactivity against stability. Flags when a smoothed trend breaches control limits, and supplies trend context for the other two.
A managed relational archive handles ingestion, historical storage and detector state, currently being migrated onto the private machines so the full record is locally held.
Every figure in this archive traces to an established governmental, academic or research institution. Nothing is generated, estimated or synthesised.
The infrastructure is built and the archive is collected. These are the steps that require institutional support, independent oversight and peer collaboration.
Submit the archive, the detection methodology and the null-result audit to independent seismological and space-weather researchers for formal review and publication in a refereed journal.
Publish the 563 GB archive with documentation, provenance and gap reporting so other researchers can query three centuries of normalised geophysical data without rebuilding the collection layer.
Deploy dedicated Schumann resonance and telluric monitoring to address the 20-year Schumann gap (1980–1999) and the 31-year telluric gap, including a self-operated ELF installation.
Complete a history-and-physics paper documenting Tesla's stationary-wave and nodal work from the primary Colorado Springs record, and assess it against modern measurement — as a scholarly question, not a forecasting claim.
TeslaQuake does not predict earthquakes, does not publish forecasts, and makes no claim that electromagnetic activity causes or precedes seismic activity. The program maintains a public observation log, an open historical archive, and a documented research question in the history of science.
Review the observation log, the Tesla primary-source record, or discuss partnership.
IN HONOR OF NIKOLA TESLA · 1856–1943
