GhostPlay — Engine Proof & Methodology

Our Engine, Scored Against Reality

GhostPlay's economy-health signal is backtested against two documented real-world outcomes: a documented collapse (Axie Infinity) and a documented clean-ship control (Path of Exile). The reconstructions below show how the engine performs on reconstructed persona mixes that match public post-mortems.

Honesty guardrail: These reconstructions are built from publicly available post-mortems and economy reports. They demonstrate the engine's ability to identify the directional health signal (collapse vs. healthy) from persona-mix parameters that approximate documented real-world structures. This is not a claim that GhostPlay predicted these outcomes in advance — it is a backward-facing validation that the signal works on reconstructed historical data. Real per-title numerics depend on your actual player behavior, not these synthetic personas.

Methodology

Mint-to-Burn Ratio (MBR) is the economy-health signal computed from a simulated 360-day trajectory. It measures: MBR = (cumulative currency inflow) / (cumulative currency sinks)

Collapse Threshold: MBR ≥ 1.5 flags collapse risk. This reflects the structural signature of faucet/sink hyperinflation — currency entering circulation faster than sinks remove it. Real-world Axie SLP reached approximately 4:1 before its economy collapsed; Path of Exile runs below 1.0 by design.

What this backtest validates: That the engine's collapse signal correctly identifies the direction of health (collapse risk vs. healthy) on reconstructed persona mixes that approximate documented real economies. The stress vectors (whale flight, exploit shock) are deliberately not injected here — these are organic reconstructions under baseline dynamics.

What this backtest does not claim: That GhostPlay predicted these outcomes before the fact, or that the numerics apply to your title without customization. Every real economy requires ingestion of your own model and personas to produce a valid forecast.

Run it yourself

The backtest harness is open for local verification:

python -m forecast_engine.backtest