Documentation
AI quant research guides, validation methodology, long-tail glossary entries, and the evidence behind the Judge.
Product
Getting StartedWhat Corrai is, how the Judge Engine works, and the three ways to produce strategy candidates — agents, Alpha Canvas, and your own data.AI Quant Research WorkstationWhat an AI quant research workstation should do: agent-assisted research, reproducible backtests, local-first data, evidence gates, and audit trails.Agent Alpha DiscoveryHow agent alpha discovery works in Corrai: AI research agents propose hypotheses, run experiments, record trial history, and submit evidence to the Judge.Alpha Canvas Research WorkflowsAlpha Canvas explained: a visual quant research workflow for data, factors, signals, validation gates, backtests, and Judge-ready evidence packages.Local-First Data EngineHow a local-first quant data engine supports point-in-time market data, schema fingerprints, staleness checks, reproducible backtests, and audit trails.
Methodology
Why Backtests LieFour mechanisms that make backtests systematically overstate performance: selection bias, look-ahead leakage, ignored costs, and regime dependence.The Deflated Sharpe RatioHow the Deflated Sharpe Ratio corrects observed Sharpe for selection bias, non-normal returns, and the number of trials — and why N must be honest.Purged Cross-Validation and EmbargoWhy random K-fold leaks in financial time series, and how purging and embargo remove label overlap and serial correlation from cross-validation.Evidence-Based Alpha ValidationA practical framework for evidence-based alpha validation: registered trials, point-in-time data, cost-aware backtests, walk-forward robustness, DSR, PBO, and human review.Backtest Overfitting ChecklistA checklist for detecting backtest overfitting: multiple testing, look-ahead leakage, same-bar fills, ignored costs, regime dependence, and missing trial history.Cost-Aware BacktestingCost-aware backtesting explained: next-bar execution, fees, spread, slippage, turnover, funding, capacity, and why zero-cost Sharpe is not evidence.Point-in-Time Data LineagePoint-in-time data lineage for quant research: availability timestamps, publication lag, revisions, survivorship bias, schema fingerprints, and leakage prevention.
Glossary
Probability of Backtest Overfitting (PBO)Probability of Backtest Overfitting (PBO): the probability that an in-sample winner falls below the out-of-sample median, estimated via CSCV.Walk-Forward ValidationWalk-forward validation defined: rolling train windows followed by out-of-sample test windows, why it beats random K-fold for time series, and its limits.Alpha DiscoveryAlpha discovery defined for quantitative research: hypothesis generation, factor search, validation, overfitting control, and why discovery is not strategy approval.AI Research FeedAI Research Feed defined: a structured stream of agent observations, hypotheses, trials, evidence events, failed runs, and Judge verdicts for quant research.Judge EngineJudge Engine defined: the conservative evidence gate that reviews quant strategy candidates using DSR, PBO, walk-forward validation, costs, data lineage, and human approval.Research DAGResearch DAG defined: a directed acyclic graph for quant research workflows that connects data, features, signals, execution assumptions, validation gates, and evidence artifacts.