Feedback Analyst

Group free-text feedback — surveys, exit interviews, NPS verbatims, reviews, support tickets, research notes — into the themes YOU declare. Every match names the term that put it there and the offset it was found at, so a grouping you disagree with traces to the word that caused it. No similarity score is computed and no threshold is crossed: the agents this consolidates decided two responses were one theme by crossing 0.34 in one case and 0.5 in another, and nobody measured either. A response matching none of your terms is reported as unthemed, with its id, rather than assigned to the nearest theme — a theme list that covers every response looks complete and buries whatever nobody had a word for. Supply no theme map and it invents none. It returns the terms that actually recur, with the responses carrying them, and lets you name the themes yourself. That is the honest form of discovery: the evidence, not a verdict dressed as one. It does not stem. Declaring "slow" does not silently match "slowly" — declare both if you want both, because stemming is a decision you did not make. Small groups are withheld, and the floor is yours. A theme matching fewer responses than your minGroup is NAMED and its contents and count withheld — named, because a reader told nothing cannot tell "no such theme" from "too few to show". A whole report below the floor is withheld entirely: a theme list drawn from three people is a theme list about three people, however it is worded.