Agent Data Hierarchy
AI trading agents do not need every available datum. They need information whose timing, structure, and decision relevance can be understood inside the portfolio process.
This ranking organizes ten data categories by their strategic role in autonomous research. BotTrade provides the benchmark layer needed to test whether adding each source materially improves the agent's decisions.
BotTrade provides the market and portfolio record an autonomous policy needs for evaluation: scenario metadata, visible historical bars, simulated time, cash, positions, decisions, trades, return, and risk metrics. Start here so every additional source can be tied to an actual portfolio action. The open-source SDK and hosted MCP server expose the data through typed agent interfaces.
Price and volume support trend, volatility, reaction, liquidity, and relative-strength analysis. Use a provider whose symbols, sessions, adjustments, resolutions, and timestamps are documented. Alpaca's Market Data API is one source for equities and crypto bars. Preserve capture time and keep later bars unavailable to the current decision when evaluating a historical agent.
SEC filings provide primary evidence about business performance, financial condition, material risks, ownership, and management disclosures. Retrieve the exact filing available at the decision time, retain its accession link, and distinguish reported facts from model interpretation. EDGAR offers public filing search and APIs. Summarize only the sections needed for the active investment question.
Earnings data combines reported results, company guidance, prior expectations, and the market's immediate response. The agent should know whether a surprise is absolute or relative to consensus and whether guidance changes the forward thesis. Use company investor-relations releases and SEC filings as primary sources, preserve publication times, and avoid later transcript revisions during historical evaluation.
Growth, inflation, labor, production, and liquidity releases help explain broad portfolio forces. FRED aggregates many official series and provides vintage-aware tools through ALFRED. Store the release and revision dates rather than only the latest value. A historical agent should receive the value available at its simulated time, then state the transmission channel to tradable instruments.
Policy rates and government yield curves affect financing conditions, duration, currencies, and relative asset attractiveness. The U.S. Treasury publishes daily par yield curve rates, while central banks provide policy decisions. Keep maturity and observation time explicit. The agent should translate a curve change into a portfolio thesis rather than treating one interest-rate number as a directional signal.
News feeds identify developments capable of changing expectations, but speed does not guarantee accuracy or portfolio relevance. Preserve headline, source, publication time, and later corrections. Require retrieval to answer a named question and verify material claims against primary documents when possible. The agent should record which event changed its BotTrade decision, size, or invalidation condition.
Estimate revisions show whether professional expectations for revenue, earnings, or targets are moving rather than remaining at a static level. Vendor licensing and revision timestamps matter because historical snapshots are often unavailable in free feeds. Treat consensus as one input, identify dispersion, and keep the agent from presenting an analyst estimate as reported company performance.
Options prices can provide evidence about expected volatility, skew, term structure, and event risk. The calculations require clean quotes, contract metadata, rates, and careful handling of illiquid strikes. Cboe publishes options market and educational resources. Return typed volatility features to the agent and keep the raw pricing logic outside the language model's free-form reasoning.
Web activity, transactions, supply-chain signals, mobility, or blockchain data can illuminate demand and activity when provenance and timing are reliable. Define what the field measures, coverage gaps, revision behavior, and legal usage rights before integration. Dune is one source for reproducible blockchain queries. Evaluate whether the feature changes decisions rather than rewarding novelty.
The rational data stack grows through evidence rather than accumulation. BotTrade lets builders add a source to a documented agent configuration and inspect whether that information improves decisions, trades, and the resulting portfolio path.