After a transcript edit: the shimmer and the popover

Decision 26 of docs/design-transcript-editing.md (owner, 2 Nov 2026): the edit stands, and the analysis follows it. About 30 seconds after the last edit the edited quote is re-analysed under the shipped shimmer; a trivial result settles with nothing to see, a strong one leaves the warning glyph, whose popover shows the change as a diff with Undo Edit and Keep changes. Drawn with the shipped theme; the glyph, the popover's contents and the highlighter marks are proposed.

artefact commentary proposed

1. The edit0 s

transcriptshipped
[08:12] p2
My pharmacist said it’s not fine to take it with food, so I’d been doing that wrong for months.
Quotes lens · elsewhereshipped
[08:12]
“Fine to take with food, apparently.” p2
Confidencemedication routine
Signals · elsewhereshipped
Confidence · Medication routine
[08:12]Fine to take with food, apparently. p2
[14:03]I just take them with breakfast, it's easiest. p5

Commentary

2. The pass runs≈ 30 s after the last edit

transcriptproposed
[08:12] p2
My pharmacist said it’s not fine to take it with food, so I’d been doing that wrong for months.
Quotes lens · elsewhereshipped
[08:12]
“Fine to take with food, apparently.” p2
Confidencemedication routine
Signals · elsewhereshipped
Confidence · Medication routine
[08:12]Fine to take with food, apparently. p2
[14:03]I just take them with breakfast, it's easiest. p5

Commentary

3a. Settled: nothing much changedthe bee bee sea → BBC case

transcriptproposed
[05:33] p4
I mostly watch the BBC news app in the mornings.
News habits
Satisfaction
Quotes lens · elsewhereshipped
[05:33]
“I mostly watch the BBC news app in the mornings.” p4
Satisfactionnews
Signals · elsewhereshipped
Satisfaction · News habits
[08:12]I mostly watch the BBC news app in the mornings. p2
[14:03]The weather's all I really check. p5

Commentary

3b. Settled: the analysis changedthe restored “not”

transcriptproposed
[08:12] p2
My pharmacist said it’s not fine to take it with food, so I’d been doing that wrong for months.
Quotes lens · elsewhereshipped
[08:12]
“My pharmacist said it’s not fine to take it with food” p2
Frustrationmedication routine
Signals · elsewhereshipped
Frustration · Medication routine
[08:12]My pharmacist said it’s not fine to take it with food p2
[14:03]Nobody told me to take them apart from meals. p5

Commentary

4. The popovera click on the gutter mark

transcript · popover open from the gutter markproposed
[08:12] p2
My pharmacist said it’s not fine to take it with food, so I’d been doing that wrong for months.

Commentary

  • Shell: the shipped dropdown box the person picker uses (.export-dropdown-menu): it opens on click, closes on Escape or a click outside. A popover can close that way; only dialogs can't.
  • Anchored to the gutter mark, beside the edited words, so the researcher reads the change next to the text that caused it.
  • Title: the owner's words, with the warning glyph.
  • Rows: only what changed, each as before → after: the quote's words, its sentiment, any tag added or removed (AutoCode's included), and the signal cards it left and joined. A row with no change is not shown.
  • Highlighter pen: removed in red, struck through; added in green, underlined. Colour is never the only signal, since red and green is the pair colour-blind readers confuse. Colours are the theme's --bn-colour-negative and --bn-colour-positive.
  • The quote row stacks old over new when the researcher's tidy was replaced (the words share little), and diffs inline when only a word or two changed (right).
  • Buttons: Keep changes is the default (primary, Return); Undo Edit takes back the transcript edit and everything it set moving, as ⌘Z does. Either closes the popover and clears the glyph; so does closing it without choosing (the changes stand). Both belong here because the transcript is where the researcher has the power to edit.
popover · a word changedproposed

Commentary

  • “hate” → “love”, an unedited quote: the quote row diffs inline; sentiment moved from Frustration to Delight; a researcher's own tag, lost progress, no longer fits.
  • Is a researcher's tag ever removed by a re-run? Today the pipeline never touches a human tag. The row here shows it struck, which would be a first: question 2.
  • Theme and section placement are not re-run per quote (decision 26: sentiment and tags only), so they never appear here; they move at the next Analyse.

Questions for the owner

  1. The quote row: stacked old over new when the tidy was replaced, inline when a word or two changed, as drawn?
  2. Researcher's tags: when the meaning flips, may the re-run remove a tag the researcher added (shown struck in the popover, Undo Edit brings it back), or does a human tag always stay and only AutoCode's and sentiment change?
  3. Reduced motion: with no shimmer, the pass is invisible until a glyph appears. Fine (proposed), or show the static fallback the atom allows, such as dimming the quote's words while it runs?
  4. The shimmer's reach: the transcript only, like the glyph (proposed), with the Quotes lens and Signals just taking the result when it lands? Or shimmer the quote wherever it shows while the pass runs, since a shimmer is “working”, not a warning?

New strings: “Transcript edit changed analysis” (owner's), “Undo Edit” (the Edit menu's own), “Keep changes” (owner's); row labels Quote, Sentiment, Tags, Signals.