Market analyst
Price action, indicators, and technical context for the as-of session—so the desk does not invent a chart story after the rating is chosen.
Cornerstone
Multi-agent stock analysis on WebserviceX assigns distinct jobs before a single desk rating appears. Specialists build evidence layers; synthesis and risk steps reconcile them; publish QC decides whether the note is fit to publish. The goal is structured disagreement and shared facts—not four chatbots glued into one paragraph.
Price action, indicators, and technical context for the as-of session—so the desk does not invent a chart story after the rating is chosen.
Headline and event flow that can invalidate or reinforce a thesis, including macro-linked news when tools or prefetch provide it.
Tone and positioning signals that may diverge from fundamentals—useful when the narrative is running hotter or colder than the numbers.
Business quality, financial structure, and valuation framing drawn from available fundamentals evidence rather than generic slogans.
After specialist packets exist, the research manager synthesizes and emits an INVESTMENT_RATING_12M footer. Optional debate or trader steps can add adversarial framing or a tactical view; production batch profiles may skip live debate LLM turns and rely on analyst packets plus programmatic risk. Either way, the report package records the pipeline profile used so readers are not guessing which path ran.
Programmatic risk constraints are deliberate: they clamp or challenge ratings when rules fire, instead of hoping the next prose paragraph remembers risk. The portfolio manager emits a STRUCTURED_RATING. Footer enforcement retries once via the model, then repairs programmatically if needed—so multi-agent analysis still ends in a machine-checkable rating contract.
Context engineering matters as much as role names. Evidence packs prioritize macro → fundamentals → news when configured; analyst tool-call limits and output-contract retries reduce runaway tool loops. Prefetch can warm OHLCV, indicators, FRED macro, peer comps, and street consensus into shared briefs so agents argue over the same facts. That is multi-agent analysis as a system, not as marketing language.
Browse the equity research directory for sector entry points, read methodology for the full flow, and use /eval to see how published directional calls aged. Educational demonstration only—not investment advice.