Test functions
1,706 426 Python test filesPrivate runtime infrastructure
One strategy contract.
Four operating environments.
CrackTrader is a production-grade quantitative runtime—from backtest through live—with deterministic replay, session-owned state, central risk and route-aware execution. Used internally. Not a SaaS.
Internal production system · selective access by agreement · no analytics or solicitation
Contract tests
47 modules Runtime and API boundariesFeed throughput
642k/s Synthetic aggregation benchmarkOperating modes
4 One strategy contractExecution modes
Mode differences kept at the edges.
Strategy logic targets a stable contract. The runtime owns shared state, attribution, risk and inventory; execution adapters absorb what must differ.
Historical replay
Strong deterministic expectation for research, regression testing and reproducibility—not a black-box result.
Simulated execution
Live or replayed inputs with simulated fills. Strategy shakeout without venue-side orders.
Real adapters, test venues
Broker adapters against testnets or venue sandboxes for integration and pre-live validation.
Full execution
Real market input and venue execution with operational safeguards and environment-dependent timing.
Architecture
One attributable path from event to inventory.
Decision provenance stays intact through the full path. Strategy, execution and route identity travel with every intent and report.
- 01Market eventfeeds · channels
- 02Snapshotimmutable · shared
- 03Strategy intentidentity preserved
- 04Risk decisioninventory · limits
- 05Executionroute-aware adapter
- 06Inventorypost-trade state
Session-owned construction
One session coordinates reference data, market state, strategies, routes, inventory, risk and control-plane services.
Immutable per-step snapshots
Market inputs produce one shared snapshot per step. Strategy-specific views filter it without duplicating state.
Full execution attribution
Strategy, execution and route identities remain attached to intents, reports and diagnostics.
Explicit runtime edges
Risk, execution adapters, post-trade sinks and kill switches are first-class boundaries—not strategy side effects.
Deterministic replay
Reproduce the decision path, not just the final number.
Runtime history, snapshots, attribution, order transitions and versioned checkpoints support replay, recovery and diagnosis end to end.
// replay guarantees
- Event order
- Verified sequence
- Snapshot state
- Coherent per step
- Attribution
- Strategy · route · execution
- Recovery
- Checkpoint + replay
Divergence ledger
Differences are intentional and inspectable.
One architecture does not mean pretending a simulator and a live venue behave identically. The differences are isolated, named and documented.
| Surface | Backtest / paper | Sandbox / live |
|---|---|---|
| Clock source | Replay clock or local scheduler | Venue and network timing |
| Fill formation | Explicit simulation and realism policy | Venue acknowledgements and fills |
| Determinism | Strong where inputs are hermetic | Environment-dependent execution |
| Safety defaults | Reproducibility and test controls | Operational guards and kill switches |
Venue support
Different markets, consistent contracts.
Venue details stay behind feed, broker and execution-adapter boundaries. Reference data normalises instruments and constraints for the rest of the runtime.
Centralised exchanges
CCXT-backed market data and execution with REST, streaming and historical ingestion paths.
- REST
- streaming
- historical
Event venues
Dedicated Polymarket and Kalshi sessions, feeds, stores, brokers and simulation surfaces.
- Polymarket
- Kalshi
- simulation
Decentralised venues
Uniswap and PancakeSwap connectors with quoting, slippage, token and transaction boundaries.
- Uniswap
- PancakeSwap
- derived OHLCV
Research workflow
Built to test ideas without quietly fooling itself.
The research stack is opinionated about leakage, provenance and comparable artefacts—not merely invoking a backtest function.
- 01Request hashstrategy · config · seed
- 02Dataset identityversioned input
- 03Validationwalk-forward · purge · embargo
- 04Evaluationruntime + stress reruns
- 05Artefactsreports · trades · registry
Research discipline & evidence
Inspectable evidence. Private implementation.
High-level architecture, runtime guarantees, scoped benchmark methods and selected interfaces are public. Source code, strategy logic, live results, configuration and proprietary datasets remain private.
Named test functions
Unit, contract, integration, end-to-end, performance and mode-parity concerns across 426 Python test files.
Contract-test modules
Runtime and public-API boundaries tested explicitly rather than left as implicit conventions.
Modes on one contract
Backtest, paper, sandbox and live retain one strategy-facing model while divergences stay explicit.
Synthetic ticks/sec
50,000 ticks processed in 77.9 ms in the documented July 2026 feed workload. CI reruns the benchmark gate; this measures parsing and OHLCV aggregation, not exchange latency.
Clear boundaries
What this is not.
- Not a public SaaS product
- Not a signals or copy-trading service
- Not an open-source trading bot
- Not a managed-investment offering
Serious enquiries welcome.
CrackTrader is a private internal system, not a public SaaS product. Serious enquiries about licensing, acquisition, technical collaboration or due diligence are considered case by case.
Strategy logic, live results and proprietary datasets remain private unless separately agreed.