What problem does it solve? Quantitative research often suffers from lookahead bias when retrieval systems return documents published after the simulated historical date. This Skill enforces strict point-in-time filtering so every retrieved research report, announcement, or earnings call transcript satisfies published_at <= as_of_date. ## Core Features & Use Cases - Point-in-Time Filtering: Discards any document whose published_at is later than the requested as_of_date, guaranteeing temporal correctness. - Structured Retrieval Pipeline: Parses the query into entities, performs vector search over a Chroma corpus, applies time filtering, then reranks by relevance plus time decay. - Schema-Based I/O: Accepts a PITQuery (query, as_of_date, optional corpus scope) and returns a PITResult that also fills the retrieval field of the research-spec schema. - Use Case: A fundamental analyst backtesting a strategy asks what was known about a company on 2023-06-30; the Skill returns only reports and announcements published on or before that date. ## Quick Start Retrieve all research reports and announcements about a given company as of 2023-06-30, ensuring no document published after that date appears in the results.
