Outputs
The two machine-readable surfaces AIVEX Feed produces — a daily context row and a per-article stream.
Two Outputs, One Entry Point
Everything the pipeline produces is read from these surfaces. There is no third interface to learn.
world_context — MySQL table
One row per day, carrying:
- Net sentiment — the day's aggregate directional score
- Per-theme breakdown — sentiment split across themes
- Regime indicators — arithmetic descriptors of the day's news regime
- Anomalies — deviations detected against the day's baseline
- Optional narrative — prose written by the local language model, when one is available
Read it with plain SQL. Later runs compare themselves against this row, which is what makes deviation detectable.
news:articles — Redis stream
Every enriched article is published as one message:
| Field | Description |
|---|---|
sentiment | Positive, neutral, or negative label |
confidence | Confidence score for the sentiment label |
theme | Assigned theme |
companies | Matched tracked-company IDs |
relevance | Relevance score used for ranking |
A consumer group is ready out of the box.
news:hot_signals — Redis stream
Anomaly alerts, published the moment a batch diverges from the day's baseline — sentiment deviation, category shift, or volume spike. These alerts are produced by pure arithmetic; no model call is involved.
The outlook Rule
The outlook field is always computed from net_sentiment by an arithmetic function. The
language model writes prose only — it never sets direction. If the model is offline, the
narrative stays empty and every number is identical.
Outputs are research artifacts for informational purposes only — not financial, investment, or trading advice, and not trading signals.