Operational Doctrine

Epistemological Crisis

Historical logs are finite and regime-dependent. The engine addresses this by generating counterfactual market days that preserve microstructure physics while remaining reproducible via seeds and manifests (see Determinism).

Principles

• Determinism-first: Batch and stream outputs are designed to match for a given manifest + seed. Provenance metadata is returned so you can replay and verify.
• Schema-first: Events are normalized to a canonical schema and remain versioned + hashable. See Configuration and the schema endpoint in API Reference.
• Venue realism: Funding and feed behavior are venue-specific, not generic perp math. See Funding Models.
• Auditability: Authentication, rate limits, and structured telemetry make it easier to operate safely and investigate incidents.
• Non-goals: No price prediction and no “alpha generation”; the focus is reproducible simulation for engineering, testing, and research workflows.

What We Model

• L2/L3 microstructure: queueing, burst/staleness, toxicity-sensitive spreads.
• Venue settlement & funding paths: Binance, HyperLiquid, OKX, Bybit, CME, SGX.
• Deterministic delivery paths: Parquet artifacts, SSE streams, MCP cache exports.

What We Do Not Model

• Real order submission/clearing; the API focuses on simulation, validation, normalization, and cache control.
• Unknown venues or bespoke liquidity models without a documented preset.
• Probabilistic “black box” randomness—every stochastic element is seeded and declared in the manifest.

Counterfactual Safety Checks

• Re-run manifests with identical seeds to validate parity before using data for production decisions.
• Inspect /mcp/manifest and /mcp/config/schema to lock versioned contracts into downstream pipelines.
• Monitor request rates, error rates, and latency for long runs to catch backpressure or downstream bottlenecks early.