Feedback and reuse¶
Rating replies¶
React to an RCA reply in the chat thread:
| Reaction | Verdict |
|---|---|
| 👍 | correct |
| 🤔 | partially right |
| 👎 | wrong |
Replies posted in the last NIMI_REACTION_DAYS (7) days are re-checked every 10 minutes, wherever they
were posted (posts in a webhook space count too, if the token's account can read it). Reactions are
stored in reactions and turned into rca_feedback rows; removing a reaction marks its feedback removed.
Skin-tone variants count the same.
Feedback is used in three places:
- Reuse: an RCA rated 👎 is never reused.
- The skill's history: previous RCAs come with their feedback counts, and the skill is told to treat ones rated wrong with distrust.
- Quality tracking: the
rca_feedback_summaryview.
When an earlier RCA is reused¶
Alerts often repeat while an incident is ongoing. Instead of a new RCA for each repeat, triage can reuse an earlier one per subject (a table, service, deployment or job the message is about):
- only RCAs for messages with the same signature,
- from the last
NIMI_REUSE_DAYS(7) days, - never one rated 👎,
- and only if it is the same ongoing incident: if history shows the subject recovered in between (an OK message, or a firing message that no longer listed it), it is a new incident and gets a fresh RCA.
A reused reply says which earlier RCA covers the alert, links to it and repeats its TL;DR. When only some subjects are covered, the fresh RCA analyzes just the new ones and the reply lists both ("partial"). When in doubt, triage runs a fresh RCA: a wrong reuse is worse than an extra RCA.