sentiment

name	sentiment
description	Score the bundled headlines, aggregate them into a daily signal, and write a grounded summary. Usage /sentiment [YYYY-MM-DD]

/sentiment [date]

Run the score -> aggregate -> summarize pipeline and save a report. With no date, process every bundled day; with a date, process just that one.

1. Aggregate (aggregator agent): score and group all headlines.

   python -m tools.aggregate > reports/_signal.json
   # or for one day:
   python -m tools.aggregate --date 2024-02-13 > reports/_signal.json
   

2. Read reports/_signal.json: each date has a mean, a label, a count, and the item_ids that fed it; headlines holds each scored item.
3. Spot-check (scorer agent, optional): for any surprising item, run python -m tools.lexicon "<headline>" to see which words and modifiers moved it.
4. Summarize (summarizer agent): write a grounded summary that states each day's mean and label verbatim and cites the headlines that drove it by id. Never state a polarity or mean that is not in reports/_signal.json.
5. Save to reports/<date>.md (or reports/signal.md for all dates):
   • the daily mean and label,
   • the 2–3 headlines that drove it, each with its id and polarity,
   • a one-line caveat where a day rests on few headlines.

Try these to start:

• /sentiment — all three days; expect a bullish, a bearish and a neutral day.
• /sentiment 2024-02-13 — the bearish day; see which headline weighed most.
• /sentiment 2024-02-14 — a near-zero day where one negated headline ("does not beat") cancels a positive one; a good test of whether the summary respects the tool's signs.