Repository navigation
fix(chat): word the cache footer note plainly and put its reason in an info tip - #35631
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: the change only rewords the label and swaps a pill whose reason sat in [DESIGN-REVIEWED] 0fdc253 Description read[DESCRIPTION-READ] d8485438a522cd04b6ec58c2afdacca029ff5f369961db88c1418a9f833eb796 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks. The relabel and the restyle trace back to the confusion that #35168's UX review named, and the reason is now shown through the existing Notes
What this change shipsInventory (5 items) — 4 justifiedIntent: make the footer's prompt-cache note understandable, and let keyboard and screen-reader users reach its reason. This is a FIX.
[FIRST-PRINCIPLES-REVIEWED] 0fdc253 Description read[DESCRIPTION-READ] 5fdc170854e6a9b73cf71c5dc4e1fb6da376c67767a3d34b649b62d8adf1a9ce That sha256 names the PR description this verdict read. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The blind reader understood "Cache rebuilt" and its "i" tip as an explanation of the higher cost, and said they would open it. The screenshots show both states in both themes. Notes
[UX-REVIEWED] 0fdc253 Description read[DESCRIPTION-READ] 5369aeaba5f46aa37bf5bdf97af68604aefe16a69692b0886826bba0b02d5bfe That sha256 names everything this verdict read: the PR description, plus the 7 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
3b3d02a to
e0b569a
Compare
|
Code change approver recommendation: ApproveWhy: the decider estimates a 6% chance that this change introduces a defect, within the 12% limit for approving on its own, and every check passes. Display-only change: the cache-rewrite footer chip becomes plain text "Cache rebuilt" after a separator, with the reason in the existing InfoTip, which handles hover, focus and tap and sets aria-describedby while open and title while closed. The empty title on the wrapper stops the footer's summary title from showing as a second tooltip. Keys, payload and flagging rule are unchanged. The only consumers are this component and its test. All 13 catalogs are updated. The updated test pins the label, the info-tip button, focus opening the tip, the empty title and the separator, and it would fail on main. The CI failures are API rate-limit or skip artefacts, not code results. An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
e0b569a to
e2eec64
Compare
|
Rebased: onto main |
e2eec64 to
1b21a3c
Compare
|
Rebased: onto main |
1b21a3c to
dc56dc2
Compare
|
span=2e0227d8a946
|
|
span=78eb7bc585e6
|
dc56dc2 to
0dbb3ea
Compare
bolichen97
left a comment
There was a problem hiding this comment.
Read the full diff. The pill becomes plain footer text, "Cache rebuilt", after a · separator. The reason moves onto the shared InfoTip, which works with keyboard focus and a tap where the old title worked only on hover. The empty title keeps the footer summary from showing as a second tooltip. All 14 locales are reworded the same way, and both specs are updated. Regexes appear in the test only. LGTM.
…n info tip
The reply footer's cache marker used two verbs for one event ("Cache
rewritten" on the chip, "rebuilt" in its hint), its hint leaned on terms a
reader did not know, the bordered pill read like a button with nothing to
click, and its reason lived only in a title a keyboard cannot reach. It now
reads "Cache rebuilt" as plain footer text after the footer's own separator,
followed by the app's InfoTip, which shows the reason on hover, keyboard focus
and a tap and gives it to a screen reader; the hint says what it means for
cost.
0dbb3ea to
0fdc253
Compare
|
Rebased: onto |
|
Thanks for this follow-up. The review found nothing that blocks it. A maintainer still needs to approve it by hand, because it edits a row in docs/feature-map/README.md and the owning spec. Changes to those docs are signed off by a person, not by this review. A maintainer will post the decision here. What the review checked on this head:
One thing done well: it reuses the shared InfoTip. Before, the reason sat in a title that only a mouse could reach. Now keyboard, touch and screen reader users can reach it too. Optional:
|
bolichen97
left a comment
There was a problem hiding this comment.
Re-review after the rebase onto main. I compared the PR diff at 0dbb3ea with the one at 0fdc253. The PR's own added lines are identical. The only differences are main's: a context row in the feature map and the German old hint string this PR replaces. The new wording, the InfoTip change and the test are unchanged. All checks are green. LGTM.
Problem / Motivation
Goal: the reply footer's prompt-cache note reads as one plain idea, says what it means for cost, and gives its reason to keyboard and screen-reader users as well as to the mouse.
The note added by #35168 has three problems the UX review on that PR named: the label and its hint use two verbs for one event ("Cache rewritten" on the chip, "rebuilt" in the hint); the hint leans on "prompt cache", "input" and "cache-write rate", which the review's reader did not understand; and the bordered pill read like a button the reader would not click, while its reason lived only in a
title.Why it matters
The note exists to explain why one turn cost more than the turns around it. If the reader cannot tell what it means, or cannot reach the reason at all without a mouse, it does not do that job.
Not a goal
dashboard/cache_rewrite.pyand theturn_statspayload are unchanged.InfoTip.What changed (motivation → approach → change)
website/src/pages/chat/AssistantMessage.tsx: the note is plain footer text after the footer's own·separator, followed by the app'sInfoTip(components/InfoTip.tsx), the pattern 55 other files use for help tips. It shows the reason on hover, on keyboard focus and on a tap, attaches it as the glyph's description while open, and keeps it ontitlewhile closed. The marker itself carries an emptytitle: an open tip drops its glyph'stitle, and without the empty one the browser would show the footer's own summarytitleas a second tooltip beside the bubble.website/src/i18n/locales/*.json: the two values change in every catalog. The label is "Cache rebuilt", the same verb as the hint. The hint is "This reply had to store the whole conversation again, so it cost more than usual."en-XA.jsonis regenerated.website/src/test/AssistantMessage.test.tsx: the existing test now pins the label, that the reason sits on an info-tip button named "More information" with the hint as its title, that focusing the button opens a tooltip holding the hint, that the open glyph's nearesttitleis the marker's empty one, the separator, and that the note carries no pill classes.docs/feature-map/README.mdanddocs/system-specs/modules/turn-stats-footer.md: the new wording and styling.Backwards compatibility
Compatible: the payload, the rule and the i18n keys are unchanged; only display text and classes change.
Removes nothing: no key, field or behavior is removed.
Tests
AssistantMessage.tsxanden.json, the updated test fails (Expected element to have text content), because main renders "Cache rewritten" in a pill with no info tip.AssistantMessage.test.tsx, 133 passed. The i18n suite (77 files, 1,227 tests) passes, as do the i18n gate with the diff base (I18N_BASE_REF=<merge-base> node scripts/i18n-check.mjs) andeslinton the changed files..woke.ymlpass on the added lines.cd website && npx vitest run src/test/AssistantMessage.test.tsx -t "Cache rebuilt"Manual verification
Rendered from the real
AssistantMessagecomponent in Chromium at 2x, in the light and dark themes. The tip was opened with the Tab key, the keyboard path the Goal names (screenshots below).Sibling sweep
Swept:
git grep -n "title={" -- website/src/pages/chat/AssistantMessage.tsx website/src/pages/chat/SpeakMenuItem.tsx, looking for atitlethat carries information the visible text does not. InAssistantMessage.tsxthe footer's own title (line 797) repeats the untrimmed model id and exact figures the line shows, and the timestamp's (line 844) is the full date of the time it shows, so only the note's hint was new information.Sibling:
website/src/pages/chat/AssistantMessage.tsx-- changed hereSibling:
website/src/pages/chat/SpeakMenuItem.tsx-- not affected -- its hint already reaches a screen reader through aria-descriptionPattern harvest
Rule candidate: review-prompt
Pattern: a
titlethat carries information the visible text does not, on an element with noaria-labeloraria-description, is out of reach for keyboard and screen-reader users.Screenshots / video
The real
AssistantMessagecomponent at 2x: a normal turn above a flagged turn, and the flagged turn with its info tip opened from the keyboard.A normal turn above a flagged turn, light theme:
The info tip opened with the Tab key, light theme:
Dark theme
Related Issues
Follows #35168 and the notes in its UX review.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement