Blog
The headline has no date. The 7-day cards do.
Monday 09:12 the headline is NPS 20 on 160 answers. This period, the 168 hours just ended, is NPS 75 on 20.
Monday, 19 October 2026, 09:12 UTC. Alex owns customer experience at Harbor, a 22-person crew-scheduling product. Survey product-nps, version 2. The question has not changed since August. Filters on the results page: source measured by UserVane, channel In-product only, trait blank, no targeting rule. The app calls survey("product-nps") when a dispatcher saves the week on /calendar.
The board at the top is NPS 20 on N=160. It does not say still stabilizing. Under it, the heading is Trend (7-day periods). This period is NPS 75 on N=20. Prior period is NPS -17 on N=12. Both cards print a plus/minus.
The calendar rewrite shipped Monday, 12 October, at 10:30 UTC. Every one of the 20 answers in This period is after that. Sam, a dispatcher, scored 9 at 11:41 UTC the same morning: "Drag onto Wednesday stayed." Riley scored 9 on Friday, 16 October, at 16:05 UTC: "Week view held. I stopped retyping the crew list." Riley's other row is in Prior period. Thursday, 8 October, 14:22 UTC, score 4: "I dragged Tuesday onto Wednesday and it snapped back to Sunday's jobs." The widget stays dark for seven days after a measured answer, so Friday was the first time Riley could be asked again. Two rows, one dispatcher.
09:20. Weekly. Slide: "Harbor NPS is 20 (n=160). The calendar rewrite did not move the company number. Park the drag-and-drop follow-up."
The weak version, and what it costs at 09:20
The read that produced that slide looks like this:
The big number is the week, because the meeting is weekly. NPS 20, n=160.
This period is a sparkline. 75 on 20 people sits next to 160, so it can wait.
Prior period can stay on the page. The slide does not need a negative week.
Park the follow-up. The score did not move.
The read that keeps Monday honest looks like this:
The board is every in-product answer on v2, with no date cut. NPS 20, N=160. That is the climate of this wording.
This period is the 168 hours that ended when the page opened at 09:12 UTC. NPS 75, N=20. Prior period is the 168 hours before that. NPS -17, N=12.
Riley's Friday line stays with the 75. Riley's Thursday line stays with the -17. The follow-up has an owner because of those two sentences.
75 does not replace 20 on the board deck.
The weak version costs the follow-up. The week the drag snapped back is -17 on the prior card, and the week after the ship is 75 on twenty answers, and the slide says nothing moved. The other cost shows up if someone pastes 75 into the deck as Harbor's NPS. That number drops the 128 answers older than 12 October 09:12 UTC. Those 128 are 53 promoters and 34 detractors. With the two windows, the full count is 72 promoters and 40 detractors, which is the 20.
The headline is the slice with no date
The score on the board uses the filters and then every stored timestamp. Version, source, channel, and trait cut the rows. A date does not. The search box and the tag box cut the verbatim feed under the cards. They do not cut the board, and they do not cut either card.
On this page the slice is measured by UserVane, channel in-product, version 2. NPS is promoters minus detractors over that whole slice: 72 minus 40, over 160, which is 20. The channel post is why In-product only is set before anyone reads the 20. Leave the filter on All channels and both clocks become the pool.
N=160 is past one hundred, so the board does not add still stabilizing. The plus/minus on that board is the margin for the 160. It is not a margin for the week.
This period ends when the page opens
The two cards are that same slice inside two windows. This period is timestamps from 09:12 UTC on 12 October, inclusive, up to 09:12 UTC on 19 October, exclusive. Prior period is the 168 hours before that boundary. The instant on the boundary belongs to This period. The cut is seven times 24 hours of UTC, counted back from the instant the results page rendered. It is not Monday 00:00 to Sunday night, and it is not the calendar week the meeting has in its head.
Open the same page at 15:40 UTC and both windows move. None of the 12 in Prior period is earlier than Thursday 8 October, and nobody else answers before 15:40, so the only row that changes cards is Sam. At 09:12 that 9 sits in This period: 16 promoters, 3 passives, 1 detractor, NPS 75 on N=20. At 15:40 the window starts at 12 October 15:40, so Sam's row is in Prior period. This period is then 15 promoters, 3 passives, 1 detractor, N=19, NPS 74. Prior period picks up that 9 and becomes 4 promoters, 4 passives, 5 detractors, N=13, NPS -8. The heading still says Trend (7-day periods). The labels still say This period and Prior period. Nothing on the card names the instant you opened.
Each window is scored on its own. Under ten answers the card says not enough responses yet and prints N. These two windows clear ten, so they print a score. They also print a plus/minus, because 20 and 12 are under one hundred. The words still stabilizing are not on the cards. They are on the big board, and this morning the big board is the 160, so the thin week is the number without that label.
Check N=20 and N=12 on the sample size calculator before either card becomes the climate. The calculator wants the people inside that window, not the 160. The older post on why a naked score lies is the same floor, applied to each card.
Twenty is a week you can read
The 75 is 16 promoters, 3 passives, and 1 detractor. The -17 is 3 promoters, 4 passives, and 5 detractors. Riley is one of the 16 and one of the 5, on different days. Read those two lines in the same sitting. The follow-up is the drag that used to snap back. The 75 is not a verdict that the ship worked, and the 20 is not a verdict that it did nothing.
A version filter is part of both clocks. August's wording, if it is still on v1, stays on v1. The rewording post is that break. This morning the question did not change, so v2 is the whole slice on purpose.
Two clocks show that the week and the lifetime are different counts of the same slice. They do not prove the Monday ship caused the 75. Twenty answers are enough to print a score and small enough that the next dispatcher moves it. That is the limit.
Bottom line: at 09:12 the headline is NPS 20 on 160 in-product answers, with no date. This period is NPS 75 on the 20 who answered in the 168 hours just ended. Prior period is NPS -17 on 12. Read Riley's two lines. Do not park the follow-up because 20 looked flat, and do not put 75 on the board deck as Harbor's NPS.