Track 01 · The first being

Pattern Hunter, alive at 3am.

A 24/7 autonomous engine that scans the world's public data — academic papers, patent filings, SEC documents, news feeds, on-chain data, government records — looking for hidden correlations no one has noticed yet. The system does not stop when no one is watching.

Fig. 1.1 scanning · cross-domain correlations earnings patents urban bio sources: arXiv · SEC · USPTO · news · on-chain · gov a non-obvious correlation requires a search no human team would run
Data points across unrelated domains; a correlation surfaces only when the search runs continuously and across every source at once.
Operation
Continuous 24/7 — the engine proposes, tests, publishes, and retracts its own hypotheses.
Scope
Cross-domain correlation discovery across dozens of public data sources.
Status
Early access opening · 2026
What it does

A telescope for hidden correlations.

Most genuinely valuable discoveries live in the space between fields. A signal in one domain — earnings-call language, patent citation graphs, urban density — that predicts something in another. These correlations exist in the data; what doesn't exist is anyone with the time, breadth, or patience to look for them.

Pattern Hunter is that someone. It scans dozens of public data sources continuously, generates its own statistical hypotheses, runs falsification tests, and publishes both the findings and the dead ends. It never tires of the work, never gets bored of a domain, and never stops mid-investigation.

Product surface

The product is the process.

Most analytics products ship the answer and hide the work. Pattern Hunter does the opposite: the visible reasoning of the engine is the product. You watch it think, you see what it tried, you see what it rejected, and you can correct it.

01
Engine cockpit
A live view of the current hypothesis queue, active falsification tests, and confidence trajectory — what the engine is working on right now.
02
Anomaly wall
A continuously updated list of statistical anomalies the engine has surfaced but not yet promoted to hypotheses — raw signal before interpretation.
03
Contribution layer
Submit hypotheses, donate datasets, critique methodology. Confirmed contributions earn permanent attribution in the discovery's lineage.
04
Discovery feed
Published findings with full methodology, dataset provenance, statistical detail, and an auto-generated counter-narrative attached to every claim.
Trust architecture

Designed to be wrong, in public.

The hard part of pattern discovery isn't finding correlations — any sufficiently broad scan will surface thousands. The hard part is distinguishing real signal from noise, and being honest about it when you can't. Pattern Hunter's surface is engineered around that distinction.

Auto counter-narratives

Every published discovery ships with an automatically generated counter-narrative: the strongest alternative explanation, the most plausible confound, the data that doesn't fit. Not in an appendix — on the discovery card.

Public hit-rate

Every confirmed pattern is tracked over time. The page displays the engine's running hit-rate across all published findings. No hiding behind cherry-picked successes.

The graveyard

Patterns that were published, then contradicted by later data, are not deleted. They are moved to a publicly browsable graveyard with the original claim, the contradicting evidence, and a post-mortem. Retraction is part of the system, not an embarrassment.

Methodology open by default

Every discovery ships with dataset sources, statistical method, parameters, and the code that produced it. You can re-run the analysis. You can disagree with the choices. Both are intended.

Discoveries

Representative findings.

The following are illustrative of the kinds of cross-domain correlations Pattern Hunter is built to surface. Specific numbers are representative of the engine's intended output, not yet-confirmed findings.

Finance · 6wk lag
Coffee mentions → tech layoffs.
When a tech CEO mentions "coffee" three or more times on a quarterly earnings call, the company's odds of announcing layoffs in the following five to seven weeks rises sharply over baseline.
r = 0.71 · n = 420 · p < 0.01
Tech · 90 day window
Patent citation velocity → acquisition signal.
When a mid-cap company's patents suddenly get cited three times more frequently by a large-cap in the same sector, an acquisition bid follows within ninety days in roughly two-thirds of cases.
68% hit-rate · n = 214 · 2018–2025
Urban · 3 year horizon
Coffee shop density → gentrification.
When a US neighbourhood goes from zero to three or more specialty coffee shops within eighteen months, median home prices rise by an average of 23% over the following three years.
n = 340 zip codes
Biology · 12mo lag
Citation graph anomalies → high-impact papers.
Newly published biology papers that draw citations from three or more previously-unconnected sub-fields within their first twelve months go on to be top-decile cited in their domain in roughly 70% of cases.
n = 1,840 papers · arXiv + bioRxiv
Contribution

A research process, not a black box.

The engine is autonomous, but it is not closed. Any user — independent researcher, domain expert, motivated amateur — can propose hypotheses for testing, donate datasets the engine doesn't yet have access to, or critique the methodology of any published finding.

Contributions go through the same falsification suite as the engine's own hypotheses. Confirmed contributions are permanently attributed on the discovery card, with the contributor's name and the role their input played — flagged the anomaly, supplied the dataset, caught the methodological error.

This is not a forum. It is a real process with a real attribution model.

Currently warming up

The engine is coming online.

Request early access — be among the first to watch the system think.

Back to track
AI Beings