Genesis Protocol

Genesis Protocol is a deterministic synthetic-world simulation engine written in Rust. It does not predict markets. It does not ingest real data. It generates reproducible agent-based civilisations under configurable evolutionary and metabolic pressure, anchors every result to a SHA-256 hash chain, and publishes the full experiment record for independent replication.

That is not a weakness. It is the foundation. The engine has a fixed epoch ordering, five domain-separated ChaCha8 RNG streams, 10 sweep variables, 28 measurable metrics, and 6,820 fully reproducible worlds across 44 experiments. Zero collapses at the default floor. The collapse rate is a definition question, not an ambiguous result — change the extinction floor from 3 to 10 and you get 97.5% collapse from the same engine on the same seeds.

What the site currently shows accurately: the full crate architecture, real source constants, every experiment with its world count and collapse note, the four hard-coded engine elements, the 28 metrics, and the replication challenge leaderboard (currently 0 outside replicators — a gap, not a claim).

What it doesn’t yet show: calibration to real data, a measurement model that bridges simulation variables to market observables, regime detection, or any probabilistic forecast output. Those are the next layer. They require being built correctly, not claimed prematurely.

A simulation engine becomes a forecasting tool only when you can say precisely: this internal state variable corresponds to that observable real-world quantity. Without that bridge, you have a generative model with no feedback signal and no way to know if it is calibrated to reality.

Genesis internal state already tracks variables that have natural analogues in real markets. The mapping is not metaphorical — it is the basis for the observation equations that will connect to calibration:

The next build step is a genesis-observe crate that formalises these as configurable observation equations, with unit tests showing stable transforms and a measurement layer that can accept real time-series as input.

Real data sources required: market prices (BTC, ETH, SOL, SPX, DXY, gold, VIX), macro (CPI, payrolls, Treasury yields, credit spreads), on-chain (active addresses, stablecoin supply, exchange flows, realized vol, perp funding, open interest), and policy event flags.

The deterministic Rust core stays intact. No epoch ordering changes. No RNG changes. No hash chain changes. Every existing experiment remains reproducible. The predictive layer is additive — it wraps the simulation engine and adds calibration, data, and scoring on top.

Predictive ability has a precise definition: train on history, forecast on unseen data, score against benchmarks, track live. Any system that cannot beat naive baselines out of sample does not have real predictive ability, regardless of how coherent its internal mechanics are.

The backtesting design for Genesis Predict will enforce strict discipline:

Every live forecast published by genesis-live must include: model version, training window end date, data timestamp, and the most recent out-of-sample validation score. Forecasts are presented as probability distributions, never as certainties. If the system cannot beat benchmarks, that result is published openly.