Skip to content

ui(seo): Fit the panels at 300px and show every navigation phase - #57

Merged
snowopsdev merged 20 commits into
mainfrom
ui/signal-later-narrow
Nov 18, 2026
Merged

snowopsdev merged 20 commits into
mainfrom
ui/signal-later-narrow

Conversation

@snowopsdev

@snowopsdev snowopsdev commented Nov 17, 2026 •

Copy link
Copy Markdown
Member

At the SEO sidebar's narrowest width (300px) several panels cut text off, scrolled sideways or wrapped in the wrong place, and a very short Page Speed timing phase could be hidden entirely. The Signal rollout recorded these five under "Left for later". This closes all five, and two more of the same kind found on the way.

What changes

  • Page Speed's "size unavailable" wraps in its column (it ran 23px past it at every width). Compliance's tally takes two columns when narrow ("Moderate" was cut 14px at 300px), and so do Page Speed's tiles ("DOM ready" and "Page weight" were cut 13–14px).
  • Every navigation timing phase shows. A phase shorter than the minimum is drawn at the minimum (NAVIGATION_PHASE_MIN_WIDTH, now 1.5, was 0.8), and the next phase starts where it ends. Before, a loopback page's Connect phase drew 0px. Positions near short phases are therefore approximate: starts and ends can move by a few percent of the axis, and a pushed bar can pass the DOM-ready marker. The markers and the legend's durations stay real.
  • The result preview's trail breaks only between crumbs (it broke "cold-brew-" from "coffee-at-home"), and the width meters keep a number with its unit ("920 | px" became "920 px").
  • SERP's ownership table fits at every width. At the default width it is the same four-column table, and a long domain now wraps between its labels instead of being cut. Below 280px of table width (the 300px panel) it stacks: the domain across the row, its counts under their headers. In a 588px window (a 160px panel) a row takes three lines. Sideways scroll at 300px goes from 60px (89px with a 59-character domain) to 0. It is still one table with its header associations, checked in Chromium's accessibility tree.
  • Search Console's controls wrap instead of cutting their labels. The country and device selects are two equal columns whatever is chosen, on every OS; each takes the whole row when the panel is narrow; "Clear filters" sits on its own line. "United Kingdom" was cut 49px at 300px and 19px at the default width; it is whole at both.
  • Content's target queries wrap instead of ending in an ellipsis: a cut query was readable only through a tooltip the keyboard can't reach. Clicks and "Use as target" sit under the query together.
  • The AI answers sheet's "Open in tab" stays on one line beside a long citation URL (found by test(design-tokens): Close the rollout's proof gaps: guards read every at-rule, four run-state captures, the page corner on every OS #55's new capture).

Markup: one span per trail crumb, a <wbr> after each dot of a domain in the ownership table, a class on that table, and a wrapper round a target row's clicks and button.

Changes at the default width

Most of this only shows when the panel is narrow. At the default 360px these change: suggested queries wrap; "size unavailable" wraps; "Clear filters" takes its own line; the ownership table loses its outer side padding (so "brewguide.example" fits its column by more than half a pixel) and wraps long domains; a very short timing phase is visible; "Open in tab" is one line.

Guards

  • New NARROW_PANEL_FITS and narrow-panel-fits-resolve: each part's wrap, container, query and rule order.
  • NAVIGATION_PHASE_MIN_WIDTH with waterfall-geometry and waterfall-shows-every-phase (including four cases where positions move); trail-crumbs; domain-labels; a no-break-space check in width-meter.
  • e2e: the Page Speed, Compliance, Content, SERP and Search Console suites assert at 300px that no text element is wider than its box and the panel doesn't scroll sideways; at the default width that the tally has 4 columns, the speed tiles 3, the ownership table is unstacked and no domain breaks inside a label; and that the two selects are equal to 1px and nothing moves between choices.

Known and deliberate

  • Each part uses its own named container, not a second seo-panel query: the existing guard allows exactly one.
  • Page URLs in the target-pages list keep their ellipsis and tooltip, like every other URL in the panels.
  • Changes at the 588px window was already fixed in ui(workspace): Draw the workspace as floating sheets with a rounded page view #52 and is covered by its e2e.
  • Five bullets come off "Left for later". One is added: the Visibility table has a column per probe field and scrolls about 1,000px sideways at every width.
  • Found, not fixed: in a 588px window Search Console's "Whole property" segment is cut 12px.

Proof

  • Spec first: on the first spec commit 5 unit tests failed (narrow-panel-fits-resolve, width-meter, trail-crumbs, waterfall-geometry, waterfall-shows-every-phase) and 6 e2e tests failed against main's build. Each later round's spec commit failed its own tests before the fix.
  • Measured before and after at 300px, 360px and 560px, and in the 588px window, with long real values (a 59-character domain, a five-digit pixel count, the app's full country list): every overflow listed above is 0 after.
  • The selects on each OS (CI's logged widths, the same for every chosen country): 154.5 / 154.5 on macOS, 155 / 155 on Linux and Windows.
  • Visual: against main's baselines, over all 92 images (probe 44208570827), three captures differ: seo-content (54 px at one "›" in the trail), ai-answers-sheet-answers (the button on one line) and seo-serp-ownership (column edges moved by the removed padding; in dark, main's capture also drew a sideways scrollbar under the table, which no longer scrolls). The rest match at 0 px, bar known noise.
  • Manual: before and after renders at 1× in light, dark and forced colours at 300px, the default and the widest width.
  • Local: pnpm turbo run lint test passes 19/19 tasks (desktop 3,143 passed, 3 skipped).
  • CI: green, 26/26 jobs on Linux, Windows and macOS, on 8a61b37 (44187817118), on 153e017 after round 1 and the merge of main with test(design-tokens): Close the rollout's proof gaps: guards read every at-rule, four run-state captures, the page corner on every OS #55 and ui(seo): Give eight states a visible cue: a check for a pass, disabled and busy by colour, badge edges, meter boundaries #56 (44203902835), and on 776c16e after round 2 (44211002538).

Council review

Rounds: 2

Delegate Round 1 Round 2
Claude Opus 5.5 8/10 7/10
GPT-6.1 Sol 8/10 9/10
Grok 4.7 not available not available

Grok's provider was still over its weekly usage limit. The owner's earlier choice, to merge on the two available reviewers while it is out, is applied here.

  • After round 1 (8a61b37): both confirmed the five fixes at 300px, the default and the widest width, and fuzzed the waterfall (220,000 rows between them: ordered, inside the axis, no zero widths). Sol found the stacked table's headers overflowing in a 588px window. Both found the spec promising that timing positions stay real when they move. Opus found the selects sized by their longest option once min-width: 0 was gone, and suggested a wider stacking threshold for the table.
  • After round 2 (153e017): Sol found nothing new. That wider threshold stacked the table at the default width, where it had fitted, so it was reverted before the round. Opus then found that with the threshold back, a domain could break inside a label at the default width, and that sizing the selects to the chosen option made the row jump as the user picks and differ by OS. Both are fixed as Opus prescribed: domains break only between labels, and the selects are a fixed equal split. Opus also closed two guard gaps and asked for the default-width column counts to be asserted. The fixes were checked against the diff, 1× renders at both widths and CI on all three OSes, with no third round.

Created with Claude Fable 5.2 (subagent: Claude Opus 5.5; council: Claude Opus 5.5, GPT-6.1 Sol) in Claude Code.

🤖 Generated with Claude Code

snowopsdev and others added 3 commits November 17, 2026 17:05
…shown

The SEO panel's parts that cut their text or scrolled sideways at its 300px minimum, from the "Left
for later" list: Page Speed's "size unavailable" and hidden short phases, Compliance's tally, Content's
result preview trail and width meters, SERP's ownership table, Search Console's controls and Content's
suggested queries. Each is specified with its measured gap and pinned by NARROW_PANEL_FITS
(narrow-panel-fits-resolve), NAVIGATION_PHASE_MIN_WIDTH (waterfall-shows-every-phase), trail-crumbs and
width-meter's no-break spaces, and e2e checks at the default width and at 300px in each panel's suite.
Five bullets come off "Left for later".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…hows

- Page Speed: a type's size wraps between its words, so "size unavailable" takes two lines in its
  64px column; a phase under NAVIGATION_PHASE_MIN_WIDTH is drawn at it and the next starts where it
  ends (speed-view.ts, waterfall), so no phase covers another.
- Compliance: the tally takes two columns where four can't hold "Moderate" (@container a11y-results).
- Content: the trail's crumbs are boxes (trailCrumbs, .seo-serp-crumb), so it breaks between crumbs;
  a meter's number keeps its unit (no-break space); a suggested query wraps whole, its clicks and
  button going under it together (.gsc-targets-end).
- SERP: a domain wraps, and Who owns page 1 stacks below 280px (@container serp-table, .serp-owners),
  still a table to the accessibility tree.
- Search Console: the range and filter rows wrap, and a select is never narrower than its options.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n crumb by crumb

From CI actuals (probe run 44186622214). seo-content light and dark change by 54px each, in a 3x220
box at the "›" between two crumbs of the two previews' address lines (the separator now sits in its
crumb's box). Every other capture matches at 0px, bar known noise (crawl-sheet-dark 22px,
seo-page-speed-dark 3px), not adopted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-17T22:51:44.031814Z 8a61b37 PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

snowopsdev and others added 17 commits November 17, 2026 18:09
…bounds, every country, the tiles

- Who owns page 1 stacks below 300px (the default 360px broke "brewguide.example"), and below 180px
  (the 588px window) a row takes three lines; e2e at NARROW_WINDOW checks headers and cells whole and
  aligned.
- The navigation timing's text says what the waterfall does: starts and ends move by up to n × the
  minimum near short phases and at the axis's end, and the markers stay real, so a pushed bar can pass
  one; waterfall-shows-every-phase tests the four cases.
- Search Console's filters share the row equally at their chosen options' widths with every country
  (field-sizing), a longer name ending in an ellipsis; the e2e loads GSC_COUNTRY_ALPHA3.
- Page Speed's timing and totals tiles take two columns under 270px; its suite checks 300px.
- NARROW_PANEL_FITS checks order against wider queries and `after` selectors; atNarrowestPanel gives
  back the width it found.
- Left for later: the Visibility table, measured.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ters, Page Speed's tiles

- Who owns page 1: stacked under 300px; under 180px Results and Share share a line and Positions takes
  the next, only the last line ruled.
- Search Console: the filters are flex 1 1 0 with field-sizing: content, so with ~240 countries they
  are still equal halves when a short option is chosen; selects end a too-long name in an ellipsis.
- Page Speed: a card is the speed-card container; under 270px its tiles take two columns.
- waterfall's comment says what moves.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
"Left for later": both sides' removals kept (this branch's five items and #55's capture, corner and
cascade bullets), with this branch's Visibility table bullet.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…beside a long URL

Found by #55's review: `.ai-answer-citation`'s "Open in tab" wrapped onto two lines. A
NARROW_PANEL_FITS entry pins `flex: none` on it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ng URL

`.ai-answer-citation .panel-button { flex: none }`: the URL beside it breaks anywhere, so it asked for
the whole row and squeezed the button onto two lines. In the 588px window's 320px page column a long
URL wraps to 12 lines beside the whole button; the sheet doesn't scroll sideways.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main keeps each countries answer, so choosing Mobile again (read earlier in the test) never asked the
fake for the full list; Tablet does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rations

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dropping the device emulation left the Visibility panel's query box with no box in serp-intent-fake
(CI probe 44198549216 and locally); setting the size it had back does not.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…AI answers citations

From CI probe 44198549216 on the merged head (92 images). seo-serp-ownership light and dark: the table
stacks at the default width (its wrap is 287px, under the new 300px breakpoint), 33,166 and 34,767px
in a 341x477 box, the cards under it moving down. ai-answers-sheet-answers light and dark: "Open in tab"
on one line beside the long URL, 1,337 and 1,360px in a 338x40 box. Every other image matches at 0px,
bar known noise (crawl-sheet-dark 22px, seo-page-speed-dark 3px), not adopted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Baselines: main's taken in the merge (#56 re-rendered them); this branch's are re-rendered from a CI
  probe on the merged head next.
- SnippetEditor.tsx: both imports (Fragment for the trail's crumbs, useId for the width meter's clip).
- design-tokens.feature.md: #56's rollout step 7 kept, this branch's becomes 8; "Left for later" keeps
  both sides' removals and both additions (the disabled rail item, the Visibility table).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…head

From CI probe 44201486309 on 48fe1ab (main at 62bd8d4 merged, 92 images), against main's baselines:
seo-content (54px each, the trail's "›" in its crumb box), seo-serp-ownership (33,166 and 34,767px, the
table stacked at 360px) and ai-answers-sheet-answers (1,337 and 1,360px, "Open in tab" on one line).
The other 86 images match main's at 0px.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…y the longest, not equal halves

macOS CI (run 44202451349) drew the default-width filters unequal: a chosen label wider than half the
row takes its width there (its menu arrow is wider), which is the intent. The check is now: one line,
inside the row, no label cut, and the country select at least 20px narrower than its longest option.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ed filter split, default-width checks

- Who owns page 1 stacks under 280px again (300px stacked it at the default width). A domain wraps only
  at its dots (domainLabels, a <wbr> each) and hyphens; the serp suite loads LONG_DOMAINS
  (?domains=long) at the default width and fails on a break inside a label.
- Search Console's filters: two equal columns fixed by the stylesheet, a line each under 280px, "Clear
  filters" under them; the e2e checks the split to 1px for four choices, prints the widths, and checks
  nothing moves as the choice changes. The "narrower than its longest" check is gone.
- The default width: four tally columns, three tile columns, the table unstacked.
- NARROW_PANEL_FITS gains the stacked thead, tbody and first cell's max-width, the outer padding and the
  domain's break-word; atNarrowestPanel's comment says where the drag starts, and its body's failure wins.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t their dots, a fixed filter split

- Who owns page 1: stacked under 280px; a domain gets a <wbr> after each dot (PageText's `parts`,
  domainLabels) and `overflow-wrap: break-word` (anywhere only when stacked); no inline padding at the
  table's outer edges, so "brewguide.example" is whole at 360px.
- Search Console: the filter row is a two-column grid (min-width: 0, ellipsis), one column under
  @container gsc (width < 280px), "Clear filters" on its own line; no field-sizing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From CI probe 44208570827 on dcac982 (92 images, against main's): seo-serp-ownership is the four-column
table as on main, its column edges moved by the table's outer padding (light 3,917px in a 284x63 box,
the table's rows); in dark main's capture also drew a sideways scrollbar under the table, which no
longer scrolls, so the cards under it sit 10px higher (35,081px in 341x477). seo-content and
ai-answers-sheet-answers match their adopted baselines at 0px; everything else matches main's at 0px,
bar seo-page-speed-dark's known 3px, not adopted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@snowopsdev
snowopsdev merged commit f1d7c35 into main Nov 18, 2026
26 checks passed
@snowopsdev
snowopsdev deleted the ui/signal-later-narrow branch November 18, 2026 01:40
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant