Nothing here asks you to trust the numbers on the front page. This page re-runs them. It loads the stored inputs, re-executes the same engine the site ships, and compares the result against the stored output field by field. Your browser does the work, so you can watch it in the network tab.
Every epoch derives a seed from the cohort root, turns each token's holder snapshot into a signal, plans a tick budget from that signal, and runs the worm. All of it is deterministic. If our stored feed disagrees with your browser by a single digit, the rows below go red.
| check | recomputed | stored | n |
|---|
What this does not prove. Level 1 verifies the simulation, not the chain data. The holder counts, entries, exits and concentration come from our snapshot of Robinhood Chain. If that snapshot is wrong, every check above still passes. To close that gap you need Level 2.
The window is Robinhood Chain blocks … to …, chain …. Re-derive holder sets by replaying Transfer logs for each token across that range against an archive node you control, then diff against data/snapshots_v3.json. We cannot do this for you in a browser tab, and you should not take our word for it either. The ingest script that produced the snapshot ships in the repo as scripts/ingest_colony_rh.py.
Per epoch, per token: the raw holder counts, the derived signal, the tick plan, and the resulting path efficiency. This is the file the checks above read.
The chain snapshot itself: holder counts, entries, exits, concentration per token per block. The input Level 2 exists to challenge.
The 30 epoch back-test: both arms, all three forward horizons, bootstrapped confidence intervals.
Partial correlations, controlling each arm for the raw inputs it was computed from.
The wiring. A clean room derivation of White et al. 1986, with its provenance block inside the file.
The engine the checks above execute is the same module the front page loads: js/engine.js. There is no second implementation kept for verification, because a second implementation is a place for the two to quietly disagree.