Research DAG

A Research DAG is a directed acyclic graph that represents a quant research workflow. Nodes perform steps such as loading data, building features, generating signals, modeling execution, running validation, and producing evidence artifacts. Edges describe dependencies between those steps.

Related search phrases include quant research DAG, strategy research pipeline, alpha workflow graph, factor pipeline DAG, and reproducible backtesting workflow.

Why DAG structure matters

Quant research is dependency-heavy. A signal depends on features. Features depend on data. Backtest returns depend on signal timing and execution assumptions. Validation metrics depend on the returns and on the registered search history.

When those dependencies live only in a notebook, it is easy to lose the exact path that produced a result. A Research DAG makes the path explicit. It also lets the system rerun, compare, audit, and freeze the workflow.

Typical nodes

A Corrai-style research DAG may include:

• Data source nodes for OHLCV, funding, on-chain, fundamentals, or alternative data.
• Cleaning nodes for schema normalization, missing data policy, and availability alignment.
• Feature nodes for returns, volatility, momentum, carry, spreads, flows, or custom factors.
• Signal nodes for ranking, thresholding, weighting, and portfolio construction.
• Execution nodes for next-bar fills, fees, slippage, capacity, and turnover.
• Validation nodes for walk-forward, purged CV, DSR, PBO, regime splits, and cost sensitivity.
• Evidence nodes for reports, verdicts, and review artifacts.

The graph does not need to be visually complex. It needs to be precise.

Research DAG and Judge DAG

Corrai distinguishes between a Research DAG and a Judge DAG.

The Research DAG is exploratory. It can branch, run variants, and allow agents to test ideas. The Judge DAG is controlled. It reads frozen artifacts, applies validation gates, and records the verdict. This separation prevents the validation procedure from being rewritten by the search process.

Benefits for AI agents

AI agents work better when they operate on structured artifacts. A Research DAG gives an agent a clear map of what exists and what can be changed. The agent can add a feature node, inspect a failed validation node, or explain why two workflows are equivalent.

The DAG also prevents hidden work. If an agent changes the execution assumption or the universe, that change becomes part of the workflow history and the trial ledger. The Judge can then account for it.

Benefits for SEO users

Users who search for reproducible quant workflow, AI backtesting pipeline, or factor research DAG are usually trying to solve the same operational problem: how to move from fast research to trustworthy evidence. The Research DAG is Corrai's answer at the workflow layer.