No entries match this view.
Keep source revisions and market timestampsNews and daily bars land separately in Snowflake; preserve when each fact became available.
Snowflake2h ago
Verified. Reviewed by a person on the team.
Schema card filed under stocks / ingestion, shared with the team.
Maya Chen reviewed it 2h ago.
The licensed market-data API supplies news revisions and daily OHLCV bars. Store the untouched responses in RAW.MARKET.NEWS and RAW.MARKET.DAILY_BARS. An article ID survives revisions; ticker plus session date identifies a daily bar.
Keep published_at, updated_at, available_at and source_url. Retry ingestion by source key without losing revision history. Downstream research uses only information available at its as-of time.
Attached lineage nodes
RAW.MARKET.NEWSSource identity, provenance and point-in-time availability.Use the approved research universeThe Google Docs research brief defines ticker coverage, corporate actions and reporting policy.
Google Docs3h ago
Verified. Reviewed by a person on the team.
Note filed under stocks / ingestion, shared with the team.
Maya Chen reviewed it 3h ago.
The team’s Equity Research Playbook defines the tracked companies and their stable security IDs. Tickers can be renamed or reused; keep the effective dates of each symbol mapping rather than joining only on today’s ticker.
Use split-adjusted prices for numerical features and disclose the adjustment convention. These outputs support analyst research; they are not automated trade recommendations.
Attached lineage nodes
ANALYTICS.STAGING.STG_DAILY_BARSEquity Research Playbook · Google Docs · reviewed by Maya Chen.Distinguish missing data from a quiet marketTrack API cursors and trading sessions so a failed fetch never becomes a zero-volume observation.Human3h ago
Verified. Reviewed by a person on the team.
Note filed under stocks / ingestion, shared with the team.
Maya Chen reviewed it 3h ago.
Record each API page and its ingestion status before advancing the source cursor. News can arrive late, while price bars follow the exchange calendar; neither should be inferred from an empty response alone.
Retry failed pages using the same source keys. Flag missing expected sessions for review, and keep holidays distinct from incomplete ingestion.
Attached lineage nodes
RAW.MARKET.DAILY_BARSIngestion completeness and retry policy.dbt joins must respect the research cutoffClean article revisions, validate daily bars and aggregate news without future information.Human4h ago
Verified. Reviewed by a person on the team.
Reference SQL filed under stocks / transformation, shared with the team.
Maya Chen reviewed it 4h ago.
stg_news normalizes timestamps and retains event identity. stg_daily_bars validates prices and volume. int_news_by_ticker groups eligible articles by security and research cutoff before joining numerical features.
where available_at <= as_of_at
Attached lineage nodes
ANALYTICS.INTERMEDIATE.INT_NEWS_BY_TICKERPoint-in-time joins prevent look-ahead leakage.Separate news sentiment from confidencePython extracts company signals with an LLM; confidence is not a probability of stock movement.Agent5h ago
Agent-set. Written by the agent; awaiting human review.
Learned fact filed under stocks / analysis, shared with the team.
Alkera recorded it 5h ago.
enrich_news.py extracts the company, event topic, sentiment and a short rationale. sentiment_score runs from −1 to 1. confidence runs from 0 to 1 and describes the model’s certainty in its extraction—not factual truth or expected return.
Scores below 0.80 require analyst review. Retain failures with null confidence and analysis_error. Carry source URL, prompt version and model version into the reporting mart.
Attached lineage nodes
ANALYTICS.RESEARCH.NEWS_SIGNALSSemantic extraction and human review policy.Evaluate trend forecasts out of timeNumerical price and volume features feed a separate ML model with walk-forward validation.AgentYesterday
Agent-set. Written by the agent; awaiting human review.
Learned fact filed under stocks / analysis, shared with the team.
Alkera recorded it yesterday.
predict_trends.py uses lagged returns, rolling volatility and volume changes to estimate the next five-session return. Fit on past sessions and validate on later sessions; never use a random train/test split for this time series.
Store training_cutoff, forecast_horizon and model_version with each prediction. Compare with a no-change baseline. An LLM confidence score is not a forecast probability or a performance claim.
Attached lineage nodes
ANALYTICS.RESEARCH.TREND_FORECASTSNumerical features and temporal evaluation.Publish two traceable research viewsBigQuery serves semantic signals and trend forecasts to BI, with separate review states.Google DocsYesterday
Verified. Reviewed by a person on the team.
Note filed under stocks / reporting, shared with the team.
Maya Chen reviewed it yesterday.
The Research Reporting Contract defines NEWS_SIGNALS and TREND_FORECASTS as separate BigQuery serving tables. BI dashboards can join them by security_id and as_of_at without confusing article-level signals with numerical predictions.
Publish only reviewed semantic labels to the default analyst view; retain all records for audit. Display source freshness, model version and forecast horizon beside every result.
Attached lineage nodes
research.news_signalsResearch Reporting Contract · Google Docs · reviewed by Maya Chen.