The state of the world as a time series
The system combines finance, geomagnetism, weather, earthquakes, collective attention and the lunar cycle. Results are research correlations, not proof of causality.
What is most interesting in the data today?
A short summary of the current World State Vector, largest deviations and historically similar states.
Guided analysis
Choose what you want to know. The application selects a suitable analysis and explains the result in ordinary language.
Main chart
Choose any signal and optionally compare it with a second signal on another axis.
Current anomalies
Z-scores relative to the historical baseline.
Most similar historical states
Cosine similarity of normalized vectors.
What followed similar states?
Future outcomes after the historically most similar World State Vectors.
Moon ↔ EUR/USD
Quick view of average daily returns by lunar-cycle quarter.
World fingerprint, time machine and live forecast challenge
A view of the latest sufficiently complete daily world state. By default it avoids the still-forming current day; compare it with history and track frozen forecasts.
World State Fingerprint
A visual signature of the latest sufficiently complete daily state built from several independent signal domains. By default it uses no later than yesterday and falls back to a more complete recent date when needed.
What stands out most
Signals with the largest standardized deviation in the selected world vector.
Time Machine
Choose a historical date. The system reconstructs that World State Fingerprint, searches only older similar states and shows what followed them.
Fingerprint at that time
Most similar earlier days
What followed similar states
What actually happened
Live Forecast Challenge
A forecast is written and frozen before the future outcome is known. The system scores it later, separating interesting historical patterns from real predictive usefulness.
Most interesting discoveries
Public shortlist of independent predictive relationships that survived at least one separate validation. A discovery is still not proof of causation.
Discovery of the Week
The most interesting currently validated result is frozen once per week, so it is not replaced later merely because a prettier result appears.
Autonomous Research Agent
The system generates bounded research questions, searches for candidates, validates them on separate data, rejects failed replications and tracks live forecasts.
What it is researching now
Research rules
Knowledge Graph
A visual network of validated time-lagged relationships across domains. Domain areas are only visual guides; nodes can move freely across the whole map. Arrows show temporal ordering, not proven causality.
Research Timeline
A log of important events: validations, robust discoveries, combinations, and live forecasts that were issued or scored.
What did the system actually find?
The most interesting results are summarized in plain language: what was found, why it matters, how many validations it survived and what it does not prove.
What did the system actually find?
The most interesting results are summarized in plain language: what was found, why it matters, how many validations it survived and what it does not prove.
Signal explorer
Choose any signal and time range.
Signal library
Find a signal by a familiar name or topic. Signals are grouped by meaning; the technical code is secondary information.
Lagged relationships
A positive lag means: compare the source today with the target N days later. For EUR/USD, returns are recommended instead of the raw exchange-rate level.
Strongest lags
Advanced analysis
Test temporal stability, lag structure, extreme events, shared influences and what followed similar historical states.
Lead/Lag heatmap
Compares many sources with one target across negative and positive lags. Green indicates positive and red negative correlation.
Rolling correlation and stability
Shows whether a relationship keeps the same direction through time or exists only in a short period.
Partial correlation
Tests how much of the X ↔ Y relationship remains after linearly removing a third signal.
Event study
Collects extreme events in one signal and shows the target’s average behavior before and after each event.
Regime analysis
Tests whether the same relationship is stronger only when a third context signal is low, medium or high.
Two-signal combination
Tests whether X1 + X2 predicts the target better than the best of the two signals alone.
Autonomous advanced findings
After ordinary discovery, cron tests a bounded number of the best combinations and looks for regimes where existing relationships become stronger.
Combinations
Regimes
Research hypotheses
A discovery is only a candidate. The system freezes the hypothesis and checks it on later data.