News-to-Signal Data Flow

Ingest, dedup, LLM extraction and scoring, point-in-time aggregation, and a blended trading signal

News-to-Signal Data Flow Ingest, dedup, LLM extraction and scoring, point-in-time aggregation, and a blended trading signal 01 / Sources 02 / Ingest 03 / Understand 04 / Signal 05 / Consume News Feeds · wires + web · 01 / Sources · streaming News Feeds wires + web streaming Filings + Wires · primary docs · 01 / Sources · timestamped Filings + Wires primary docs timestamped Asset Universe · tickers + aliases · 02 / Ingest · reference Asset Universe tickers + aliases reference Dedup + Cluster · hash + near-dup · 02 / Ingest · one story Dedup + Cluster hash + near-dup one story LLM Extract + Score · entity, event, impact · 03 / Understand · calibrated LLM Extract + Score entity, event, impact calibrated PIT Signal Builder · as-of, decay, net · 04 / Signal · no look-ahead PIT Signal Builder as-of, decay, net no look-ahead Signal Store · per-asset series · 05 / Consume · point-in-time Signal Store per-asset series point-in-time Strategy Engine · blend + size · 05 / Consume · positions Strategy Engine blend + size positions Quant Factors · price + vol · 05 / Consume · numeric Quant Factors price + vol numeric raw articles unstructured primary docs authoritative clean documents one per story ticker + alias map reference join scored events with confidence net signal time decay sentiment feature point-in-time price + factors numeric Legend primary data policy / PII async batch data store

Primary Signal Path

  • • Articles flow left to right through source, ingest, understand, signal, and consume
  • • Dedup collapses syndicated copies into one story before the LLM sees them
  • • Labels name data assets, not generic API verbs

No Look-Ahead

  • • The signal builder stamps every event with an as-of time
  • • Only information available at decision time reaches the signal
  • • Backtests replay the same point-in-time store used live

Blended Consumers

  • • The signal store exposes a per-asset time series
  • • The strategy engine joins the qualitative signal to quant factors
  • • Sizing and positions stay downstream of the blend