suggestion events and learning loop
suggestion events record what users do with feedback: accept, reject, ignore, or edit. those choices help the app understand which advice is useful.
quick answer
the product learns from repeated user behavior, so the best signal is honest action on each suggestion.
use this for
- explaining accepted and rejected suggestions
- understanding drift proposals
- teaching agents to ask before applying bulk fixes
steps
- review each suggestion.
- accept only when it improves the line.
- reject or ignore advice that is wrong.
- edit manually when the suggestion is close but not right.
- watch for recurring lessons after enough signal exists.
details to know
- learning events can be tied to a document, profile, rule, and action.
- edited suggestions are valuable because they show how the user corrected the correction.
- recurring patterns can turn into drift proposals or memory signals.
limits and edge cases
- one event is not enough to redefine a voice.
- agents should not silently accept all suggestions just to speed up cleanup.
- private learning signals should not be exposed in public shared docs.
ready to use the product instead of reading about it?
open app