Blog
Search does not change the score
Friday 09:22 the feed is four cutoff rows. The board above them is still NPS 30 on 180. Search text never reaches the histogram.
Friday, 6 November 2026, 09:14 UTC. Ren leads support at Lumenroute, a 22-person lane-quoting product. Survey product-nps, version 2, measured by UserVane, channel All, trait and value empty, no targeting rule. A cutoff clock shipped on the confirm step Monday, 2 November 2026, 15:10 UTC. Ren opens results the way the Surveys list opens it.
The board reads NPS 30, plus/minus 14, among measured by UserVane, N=180, response rate 100%. The lab says N=180, 95% confidence, true score almost certainly between +16 and +44. It does not say Still stabilizing, because 180 is past one hundred. The split is 90 promoters (9 or 10), 54 passives (7 or 8), and 36 detractors (0 through 6). NPS is 90 minus 36, over 180, which is 30. The passives sit in the denominator and do not get their own term. The measured card reads 180. Imported history reads 0. In-product 180, link 0, email 0.
Under Trend (7-day periods), This period is the 168 hours ending when the page opened: 30 October 09:14:00.000Z inclusive through 6 November 09:14:00.000Z exclusive. That card is NPS 29, plus/minus 34, N=28: 12 promoters, 12 passives, 4 detractors. Prior period, the 168 hours before that, is NPS 9, plus/minus 38, N=22: 8 promoters, 8 passives, 6 detractors. Neither card says Still stabilizing. The plus/minus is the whole warning on a card under one hundred. The headline post is those two clocks.
The 100% next to N=180 is 180 measured answers over 180 plus zero dismissal rows on this survey. It is not how many people saw the widget this week. The response-rate post is that denominator.
09:22. Ren types cutoff into Search text, leaves every other control where it was, and clicks Apply. The verbatim feed is four rows, newest first:
- 2026-11-06T08:40:00.000Z, rating 1, in-product, detractor. "Confirm hides the cutoff until after I accept the lane."
- 2026-11-05T19:12:00.000Z, rating 2, in-product, detractor. "The cutoff showed up on the invoice, not on the rate."
- 2026-11-04T11:05:00.000Z, rating 0, in-product, detractor. "The cutoff was not on the quote I sent the shipper."
- 2026-11-03T16:44:00.000Z, rating 6, in-product, detractor. "The cutoff is there. It is under the fold on mobile."
All four are inside This period, and all four are after Monday's 15:10 ship. They are also This period's only four detractors. The board is still NPS 30, plus/minus 14, N=180. This period is still NPS 29 on 28. The measured card is still 180. 09:31. Slack, drafted: "Searched cutoff. Four complaints, NPS still 30, this week still 29. Loud comments, not a score move. Leaving the confirm step."
The weak version, and what it costs by 09:41
The read that produced that draft looks like this:
Apply runs every field in the form. Search text is one of those fields, so the board above the feed is the cutoff board.
Four detractor sentences under NPS 30 means the people who mentioned the cutoff are a minority inside a healthy 30.
This period at 29, next to a lifetime 30, means the week of the ship did not move.
The read that keeps 09:31 honest looks like this:
Search text changed the verbatim feed. It did not change the histogram. NPS 30 is every measured rating on version 2, commented or not. The other 176 ratings are still in that 30: all 90 promoters, all 54 passives, and the other 32 detractors.
This period's 4 detractors are these four rows. The other 24 answers in that card are 12 promoters and 12 passives, and they are not in this feed. The card did not become a cutoff score when the box was filled.
Four rows are under ten. The product will not print an NPS for them. There is no cutoff NPS of 30, and there is no cutoff NPS of anything else.
The weak version costs 09:41. The confirm step keeps the cutoff under the fold through the weekend. The four people who named it are the entire detractor count for the week, and the draft treats them as color on a 180.
Version, source, channel, and trait move the board. Search and tag do not
The results form is seven fields and one Apply button. Version, source, channel, and a trait plus a value go into the histogram that builds the board, the two trend cards, and the measured, imported, and channel counts. Search text and tag do not. They are applied later, only when the page loads the verbatim rows.
The note under the heading Verbatim feed says the rows are shown as written, and that manual tags and search are the only extra cut, with no LLM summary. That sentence is about the feed. It does not rewrite the board above it. A board that still says N=180 after you search is not agreeing with the four rows. It never saw them as a filter.
Trait is the control that looks most like search and behaves the opposite way. Trait plan and value enterprise, both filled, rescores the board as that slice and labels it plan=enterprise. Search cutoff leaves the label on measured by UserVane. Empty the value and the trait does nothing. Search does nothing to the score even when it is filled.
The feed is a substring, the newest 50, and no count
Search keeps a row only when the comment is non-empty and the lowercased text contains the lowercased box. cutoff matches "the cutoff". It does not match a rating with no sentence. A promoter who wrote about onboarding and never used the word is not in this feed either. The match is a substring. It is not a tag, and it is not a theme.
The feed then keeps the newest 50 of those rows. This page does not print how many matched. With four, you can count them. With eighty, thirty are not on the form, and the form has no next page. The comments post is the same list with the box empty: newest rows that have text, not the 168-hour card. Search narrows that list. It still does not become the card.
Tag uses the same late cut and a stricter match. The tag on the row must equal the box after both are lowercased. cutoff does not match a tag named cutoff-mobile. Saving a tag is Pro+. The Tag field is on the form on every plan. A tag that was never saved returns "No verbatims match these filters." and leaves NPS 30, plus/minus 14, N=180 where it was.
Four detractors are not a score
Put the rows you can see on the sample size calculator before anyone quotes them: N=4, 0 promoters, 4 detractors. That is under ten, so the board would say not enough responses yet, the same floor as the thin-sample post. Then put This period on it too: N=28, 12 promoters, 4 detractors. Plus/minus 34 is why 29 is not yet the climate of the confirm step.
A row matches because the letters appear. It does not prove the cutoff is why the rating is a 1, and a sentence that says the cutoff was fine would sit in the same four. That is the limit.
Bottom line: at 09:22 the feed is four detractors and the board is still NPS 30 on 180. Search and tag never reach the histogram. Version, source, channel, and a filled trait do. Do not tell the founder the confirm step held because the number above a search did not move.