Why you can trust the answers
The app is built so it can’t confidently make things up — a few guardrails keep every answer honest.
Your question
↓
Guard 1
Search the meeting
Keep only chunks that are genuinely close — drop the rest as noise.
→
If nothing is relevant
“I couldn’t find that in this meeting.” — it refuses, instead of guessing. (It doesn’t even call the AI.)
If it found good chunks
Guard 2 — answer using only those excerpts
Guard 3 — every claim cites an excerpt [chunk:ID]
Guard 4 — our code deletes any citation that isn’t a real excerpt
It doesn’t guess, it answers only from evidence, and every claim is one click from the source — so you can always check it.