Data extractor
Fetches OHLCV for a symbol and date range, writes it as parquet into the strategy workspace, and returns price stats, data-quality checks, a volatility regime classification, and starting-point parameters in a data brief.
Scout Research by Polarline
AI quant research assistant
Scout Research investigates a market premise, gathers data and news through focused research specialists, turns the idea into strategy code, and keeps the backtest evidence in a session your team can review.
From premise to reviewed evidence
Frame the question
Start with a premise: a market, a timeframe, and what you expect to happen. Research mode lets you investigate before any code is written.
Gather evidence
Specialists collect price data, news, filings, and sentiment in parallel. When you give a date range, data extraction and news research run before the strategy is generated, so the builder works from what came back.
Turn it into code
The premise becomes a Python strategy in a versioned workspace. The code is validated against the strategy contract and lookahead rules before it is allowed to run.
Test it until it breaks
Backtests run with walk-forward, robustness, and Monte Carlo checks. The harness assigns a deterministic verdict, and most honest ideas end as research only.
Hand it to a reviewer
Prompts, tool calls, outputs, and follow-ups stay in a shared session, so a trader or portfolio manager can question the result without rebuilding the context.
Specialists with narrow jobs
A single agent that designs strategies, scrapes headlines, and cleans data at once loses track of what it knows. Scout delegates research to specialists, each with its own context and budget. None of them can create strategies, run backtests, or decide that partial data is good enough.
News researcher
Runs on a budget of roughly 30 tool calls across web search, full-article fetches, and curated market-news channels, then writes dated headline tables and cited article summaries the builder reads on demand.
SEC filings
Pulls EDGAR filings, ownership, insider, and institutional information when a company-level thesis depends on them.
Sentiment analysis
Summarizes aggregate crowd positioning and attention as context for the premise, not as a trading signal on its own.
Built to doubt its own results
A language model will happily report a strategy with a spectacular Sharpe ratio. Usually the cause is lookahead bias, an overfit parameter sweep, or bad data. Scout's harness checks data coverage against the market calendar, runs walk-forward and robustness tests, and returns one of three verdicts: research only, candidate, or recommended for paper. Recommended for paper also requires positive total return, positive stitched out-of-sample return, and positive alpha over buy-and-hold, and moving to paper trading is always an explicit human call.
Validation before runs
Generated code is checked for forbidden imports, exec and eval, and lookahead before it can be backtested.
A record of the work
Session history replaces screenshots and pasted logs when someone asks how a result was produced.
Captured examples
A failed qualification stays research only. A research result is not approval to trade.
Measured, not claimed
ScoutBench scores the harness on hallucination, numeric discipline, lookahead repair, and research memos.
Where it runs
Individual traders can open Scout Research in the browser with the Plus plan ($20 per month) or Pro plan ($50 per month), and run the free Scout CLI locally with their own model key. Funds that need research inside their own environment can choose Scout Private, which runs in your environment, or Scout Managed, where we host and operate Scout within an agreed service scope, or discuss an enterprise deployment. Integrations, permissions, and model connections for private deployments are scoped with each team.
Quant research guides
- Scout Cloud, the AI quant research workspaceShared research sessions in the browser or your own cloud.
- Research subagents referenceHow the data extractor and news researcher work, and what they return.
- Plain English to trading strategyTurn an idea into rules a backtest can test.
- AI trading benchmark resultsHow Scout's harness scores on hallucination, lookahead, and memo tasks.
- AI backtesting toolThe engine behind every research verdict.
- Shared research sessionsWhy research needs a record a team can follow.
Questions
- What is an AI quant research assistant?
- An AI quant research assistant helps a trader or quant investigate a market idea: it gathers data and context, turns the idea into testable strategy code, runs backtests, and summarizes the evidence. Scout Research does this in a shared session so the work can be reviewed.
- How is Scout Research different from a general chatbot?
- Scout Research runs specialist research tasks with their own budgets, writes strategy code into a versioned workspace, validates it for lookahead, and backtests it on real historical data. A general chatbot answers in text and cannot show where its numbers came from.
- Will Scout Research tell me what to trade?
- No. Scout Research produces research and evidence, not investment advice. A strategy can be marked research only, a candidate, or recommended for paper trading, and promotion to paper is always a human decision.
- Can a team share research sessions?
- Yes. Scout Cloud keeps prompts, tool calls, outputs, and follow-ups in one session that permitted teammates can inspect and continue in the browser. Private cloud or VPC deployments are scoped with each team.
- Which AI models does it use?
- In the CLI you bring your own key for Anthropic Claude, OpenAI, or Google Gemini, or run a local model through Ollama or LM Studio. Model connections for private deployments are agreed with each team.
Scout is research software. Research output, backtests, and verdicts are not investment advice.