How dreamer-85/quant-lab works
QuantLab Research Engine is a bounded-memory market-data research pipeline built on Lean. Its principal workflow streams raw market events, deterministically reorders and replays them, reconstructs state and observations, computes features, optionally attaches delayed-return labels, then runs experiments and writes results. The README also documents checkpoint/resume, configurable feature and experiment plugins, and local/cloud orchestration. Sampled engine excerpts support the replay, observation, feature, and experiment responsibilities; storage, labeling, input merging, and orchestration details are represented only at the documented subsystem level.
Main entry points: EventReplayEngine.cs Deterministic replay · Runner Job runner · ObservationEngine.cs Observation engine · quantlab Unified CLI
Components 16
Data and replay
- Stream and merge
Research/Engine - Deterministic replay
Research/Engine/Replay/EventReplayEngine.cs - Market state
Research/Engine - Observation engine
Research/Engine/Observations/ObservationEngine.cs
Research analysis
- Future-return labels
Research/Engine - Feature engine
Research/Engine/Features/FeatureEngine.cs - Experiment engine
Research/Engine/Experiments/ExperimentEngine.cs
Job runtime
- Result streaming
Research/Engine - Job configuration
Research/Engine - Job runner
Research/Runner - Checkpoint resume
Research/Engine
Run operations
- Unified CLI
Research/Python/quantlab
Other components
- Researcher
- Market data
- CSV JSON Parquet
- Cloud VM
Connections 19
- Researcher → Unified CLI: submits job
- Researcher → Job configuration: defines job
- Job configuration → Job runner: configures
- Unified CLI → Job runner: runs locally
- Unified CLI → Cloud VM: dispatches job
- Cloud VM → Job runner: runs remotely
- Market data → Stream and merge: streams files
- Stream and merge → Deterministic replay: merged events
- Job runner → Deterministic replay: starts replay
- Deterministic replay → Market state: reconstructs state
- Deterministic replay → Checkpoint resume: snapshots state
- Checkpoint resume → Deterministic replay: restores state
- Deterministic replay → Observation engine: emits results
- Observation engine → Future-return labels: provides observations
- Observation engine → Feature engine: provides observations
- Future-return labels → Experiment engine: provides outcomes
- Feature engine → Experiment engine: provides features
- Experiment engine → Result streaming: produces results
- Result streaming → CSV JSON Parquet: streams output