ask

name	ask
description	Answer a question about the bundled NovaCorp filings, grounded and graded. Usage /ask "<question>"

/ask ""

Run the full Perceive → Reason → Act → Check loop and save a report.

1. Retrieve (retriever agent): python -m tools.retrieve "<question>" -k 4 > reports/_context.json
2. Answer (analyst agent): write a grounded answer from reports/_context.json, citing chunk ids for every figure.
3. Grade (grader agent): python -m tools.grade --question "<question>" --answer "<answer>" --context reports/_context.json
4. Gate: if faithfulness < 0.7 revise the answer (or declare it unanswerable); if relevance < 0.5 re-retrieve with a better query. Loop at most 3 times.
5. Save to reports/<slug>.md:
   • the question,
   • the final answer with citations,
   • the two scores,
   • the chunk ids used.

Try these to start:

• /ask "What is NovaCorp's customer concentration risk?"
• /ask "How did gross margin change year over year, and why?"
• /ask "What was NovaCorp's net income?" ← not in the corpus; the agent must say so.