Research question
How can a market-intelligence platform preserve the timing and provenance of price observations, filings, news, corporate actions and portfolio exposures while generating real-time signals?
Proposed model
Every observation is represented as an event with source time, ingestion time, effective time and provenance. Instruments resolve to stable identities; issuers, listings, venues and portfolios remain separate but connected concepts. Corporate actions and disclosures modify the graph through time rather than overwriting history.
Signal context
A prediction node links to the exact feature set, market regime, input events, model version and portfolio exposure used at inference time. This creates an evidence path that can be queried after the market moves.
Expected value
The temporal graph supports point-in-time reconstruction, feature reproducibility, model-risk review and explanations that distinguish correlation from known business relationships. It also reduces leakage caused by mixing event time with processing time.
Next step
The SDS research program is evaluating temporal subgraph retrieval, graph-based feature propagation and confidence-aware signal ontologies for intraday stock and ETF analysis.