Building an AI-Assisted Quant Strategy Development and Review Loop
Summary
The document presents a workflow for using an AI coding assistant alongside a local strategy project and FMZ. The local project stores source code, configuration exports, design notes, tests, and reports; remote synchronization updates platform source, while an extension API retrieves runtime details and logs. The article distinguishes platform source from the version currently running and recommends checking configuration and restart state after synchronization.
For review, it maps logs, status summaries, chart records, and profit records to questions about orders and strategy behavior. It emphasizes structured event records so that signals, submissions, acknowledgements, fills, cancellations, and settled positions are not confused. A case study uses runtime diagnostics and source inspection to explain a strategy’s lack of trades through conflicting recovery and cost thresholds. This is a process example, not evidence of profitability; the author cautions that synthetic tests, backtests, and limited live observation each leave important uncertainties. Credentials and API permissions also require careful handling.
Key ideas
- A persistent local project with design notes, configuration exports, tests, and reports helps an AI assistant resume strategy work with context.
- Remote source synchronization does not itself prove that a running instance has loaded the new version.
- Runtime details, logs, chart records, and profit records can provide evidence for diagnosing strategy behavior.
- Structured order events help distinguish a signal from an accepted order, a fill, and a reconciled position.
- A no-trade diagnosis can expose conflicting signal-recovery and transaction-cost thresholds, but does not demonstrate profitability.
- Testing should state its data, execution assumptions, and gaps because synthetic scenarios, backtests, and small live samples cannot guarantee real-world results.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.