# Matrix Watcher

> An open, real-time experiment that monitors 12 completely independent real-world data streams and rigorously tests one honest question: do anomalies in unrelated domains line up more often than pure chance would produce? Live at https://matrixwatcher.space — open source, real public data, and honest about its results.

## What it is
- A measurement instrument, **not** a prediction service. It does not claim to predict crypto prices or earthquakes.
- Uses **no AI or neural networks** — only transparent, rule-based statistics (probability = occurrences / observations), validated with shuffle tests and out-of-sample backtesting.
- Founded 2025 by Aleksejs Moisejevs. Collecting data continuously since 12 December 2025.

## The 12 monitored sources (all real public feeds)
- **Crypto** — Bitcoin & Ethereum prices and volatility (Binance)
- **Blockchain** — network block times and on-chain anomalies
- **Quantum RNG** — hardware quantum randomness (ANU QRNG)
- **Space Weather** — geomagnetic Kp index (NOAA SWPC)
- **Solar Wind** — real-time solar-wind speed, density and IMF Bz; the physical precursor that drives geomagnetic storms (NOAA DSCOVR)
- **Solar Activity** — F10.7 radio flux, GOES X-ray flares, proton flux (NOAA)
- **Earthquakes** — global seismic activity, magnitude and location (USGS)
- **Volcanoes** — weekly volcanic activity report (Smithsonian / USGS)
- **Weather** — temperature and pressure (Open-Meteo)
- **News** — global headline volume (public RSS)
- **Wikipedia** — global human edit rate, an objective pulse of the information field (Wikimedia EventStreams, Wikipedia-only, bots filtered)
- **Earth Tides** — solid-earth gravitational tide from the Sun and Moon, computed locally from ephemerides; used as a phase reference (a covariate, not an anomaly stream), and immune to any network outage

## How it works
- Each source is polled in real time. Detection is **adaptive**: a reading is flagged only when it is unusual relative to that stream's own recent distribution (robust statistics, floating thresholds) — no hardcoded numbers. Genuine physical events (real earthquakes, geomagnetic storms, M-class flares) are flagged at their established physical levels.
- **Self-learning, forward-verifying loop:** for every condition the system learns P(event | condition), shows a prediction only when it beats the event's base rate by a meaningful margin and has enough observations, then verifies each prediction against what actually happened and keeps an honest running score. If a pattern stops working it stops being shown. Every domain can predict any other.
- When anomalies from several independent sources fall in the same 30-second window, that is a cluster. The level (1–5) is simply how many distinct domains coincided — a temporal coincidence, never a claim of causation.
- Apparent patterns are stress-tested to disprove them: a **shuffle test** vs time-randomized data, an **out-of-sample backtest with block bootstrap**, and **base-rate comparison** (a probability that merely matches chance is not a finding).

## Honest finding (as of June 2026)
- No cross-domain predictive edge has held up out of sample. The only relationships that survive testing are **within a single domain** (storm persistence, earthquake aftershocks) — known physics, not hidden links between unrelated worlds. The rebuilt adaptive system now accumulates forward evidence continuously; the honest verdict on cross-domain links is **“not proven — still gathering data,”** not a final “no.” If a genuine signal appears, the same strict tests will surface it credibly.

## Links
- Live dashboard: https://matrixwatcher.space
- Source code: https://github.com/amois3/matrixwatcher.space