Not everything said in a meeting is a decision
“Maybe we should raise the price” and “we’re raising the price” read the same in a transcript. Jiva marks the difference.
Here is a subtle problem with transcripts. "Maybe we should raise the price," "I think we're raising the price," and "we're raising the price" all end up as plain text on the page, and to a machine skimming for what happened, they look nearly the same. But they are completely different things: a musing, a belief, and a decision. Treat them alike and your summary will confidently report a decision that was only ever a suggestion.
So Jiva tags each line with its epistemic status — roughly, is this something said, something decided, or an action someone committed to. Note that the primary driver is not linguistic tidiness, rather not overclaiming: the most damaging summary error is promoting a hedge into a decision, and the tag is what stops that from happening.
I built this deliberately, not with a model guessing, but with conservative deterministic rules — it would rather leave something as a plain "said" than wrongly stamp it "decided." That is the right bias. A missed decision you can find by reading; a fabricated one you will act on.
The general lesson is one about respecting uncertainty in language. Humans hedge constantly, and the hedge is information — it is the difference between "we might" and "we will." A tool that flattens all of it into confident text is not neutral; it is quietly upgrading everyone's maybes into commitments. Keeping the hedge is part of keeping the record honest.
More soon.