Interactive
Live Lab
Six working demonstrations of the projects behind this site, built to run for free on a static page.
Everything here runs in your browser. Panels marked LIVE call public APIs (GitHub, PyPI, USGS) directly from your browser; nothing is sent to the site owner and nothing is stored beyond a ten-minute cache in your own session. Panels marked SNAPSHOT show frozen registry data. Panels marked ILLUSTRATION use synthetic data or toy assumptions and say so.
Live repository and package data, plus a claim you can freeze and watch drift.
ClaimBound Evidence, right now
Freeze a claim
A claim card commits one narrow claim to a protocol and a hash. Edit anything after freezing and the status turns to drift. This is a concept demonstration in your browser, not the ClaimBound command line.
Live release data, plus a simulator that shows how the best of many strategies is just luck.
Monte-Neo, right now
Why the best of many strategies is usually luck
Every variant below is pure noise: daily returns with mean zero. Pick how many variants you try and for how long, and read off the best headline Sharpe. This is the selection effect that a verifier has to see through. Synthetic data; not a Monte-Neo run. The hypothesis factory behind this research generated 4,054 hypotheses and validated none, for the same reason.
The 26 laws as a clickable map; every verdict, every experiment.
The law map
Each dot is one named law, coloured by its verdict, grouped by family. Hover, focus or tap a dot for the measurement and result; click the legend to hide a verdict. Frozen from the author's registry; mechanics are not disclosed.
Threshold transfer
The one place where QWC beat a standard method in a replicated test.
Pre-registered experiments
Green passed, red failed or was withdrawn, amber was void. Each was frozen before it ran.
Cache, symbolic derivation, model, verifier: the routing principle, running.
Cache, then symbolic, then model, and always a verifier
NeoArc is built on the idea that a slow model should be the last resort and that no answer is accepted unless a real verifier agrees. This panel runs that routing for arithmetic: tier 1 is a cache, tier 2 a symbolic evaluator, tier 3 a model (not available in a static page), and an independent second algorithm verifies every answer. Illustration of the principle; no benchmark numbers are claimed.
This week's real earthquakes, normalised and pushed through a Monte Carlo chain.
The world, this week
From an event to a probabilistic path
Pick one of the real events above, then set your own assumptions about how it hits a hypothetical index. The platform estimates chains like this from history; here the numbers are yours and the model is a toy. Not a forecast, not advice. Paper trading only, by design.
Claims as content-addressed objects, sealed snapshots, agents that may only propose.
Claims as content-addressed objects
In a knowledge engine of this kind a claim is identified by the hash of its content, evidence is attached with an explicit relation, a snapshot is sealed with a root hash that covers everything before it, and an AI agent may propose but never commit. Two real registry entries are preloaded. The hashes below are computed live in your browser.
Add a claim or seal a snapshot.