Repository navigation
Lead Overview with what is wrong, and add a jump-to palette - #11
Merged
Merged
Conversation
Two things taken from an IA study for another console. Most of it does not
apply here and was left: it is built around promoting config between
environments, and this console has no environments, no write API and a byte
budget. Its list-with-a-side-panel would also be a regression against "the
URL is the artifact", since the panel is not addressable and the detail
route already is.
The attention banner. Job one in PRODUCT.md is "is the federation healthy,
and if not which upstream?", and this page answered it with a count in the
Federation card, leaving the operator to reach /upstreams and work out the
filter. One sentence now names the condition and links to the list already
filtered to it, and renders nothing at all when there is nothing to say.
It is deliberately not role="alert", which is why Banner grew `announce`.
The alerting banners — unauthorized, skew, a failed read — describe the
console being degraded; they arrive unbidden and the page underneath is not
what was asked for. Federation health is the opposite: it is the answer the
operator navigated here to read, and hijacking a screen reader to deliver
requested content is noise dressed as urgency. Keeping the role reserved
also keeps getByRole('alert') meaning one thing, which two tests were
relying on without saying so.
The palette. Ctrl/Cmd-K, navigation only. The existing filters are good and
stay in the URL, but neither gets you to a page from somewhere else, and the
catalog runs to several hundred names. The hints use words and the two
arrows the shipped subset actually carries; a command glyph or a return
arrow would be drawn by whatever the machine fell back to, which is the
defect this repo already fixed once for the arrows. The trigger sits after
the skip control in DOM order, which polish.spec.ts already pins.
Both cost about 13 KiB, taking the bundle to 76% of budget.
One thing the a11y suite learned: scan() settled animations once, which was
enough while every banner rendered with the route. This one renders when the
data lands, after <main> became visible and after the first settle resolved,
so axe intermittently sampled it mid-fade and reported a composite as a
contrast failure. It now settles, waits two frames, and settles again — the
same gap applies to anything arriving with a query rather than a route.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
blakebauman
added a commit
that referenced
this pull request
Aug 15, 2026
Only dist/app.js conflicted, and build output is never hand-merged: the source merged cleanly, so this is a rebuild of it. Both sides survived and are independent. #11's attention banner and palette live on Overview; this branch's change is which URL a DocsLink points at. The one place they touch — overview.tsx — took both without either side losing anything: the banner leads the page, the docs link still sits in the header action, and it no longer carries a version prop that does not exist. 91 e2e tests and 8 docs-link checks pass on the merged tree. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
blakebauman
added a commit
that referenced
this pull request
Aug 15, 2026
The merge commit shipped a stale dist/app.js. Resolving the conflict, I ran `git checkout --ours` on it after the rebuild, which put the pre-merge build back over the correct one — so the committed bytes were missing everything #11 added. Worse than the slip was the check: `git diff --quiet -- dist` compares the working tree to the index, and both held the same stale file, so it reported agreement. CI's check is `git diff --exit-code -- dist` *after* a build, which is the only form that means anything, and it is what caught this. Build output has no "ours" side. The resolution for a conflict in dist/ is always to fix the source and rebuild. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two borrowings from an IA study written for another console (EdgeVault's). Most of it does not apply here and was left alone — it is built around promoting config between environments, and this console has no environments, no write API, and a byte budget. Its list-with-a-side-panel would also be a regression against the URL is the artifact, since the panel is not addressable and the detail route already is.
The attention banner
Job one in
PRODUCT.mdis "is the federation healthy, and if not which upstream?" This page answered it with a count in the Federation card, leaving the operator to reach/upstreamsand work out the filter. One sentence now names the condition and links to the list already filtered to it.It renders nothing when the federation is healthy. Half-open breakers are not counted (neither proven nor failed), and a disconnected upstream whose breaker is also open is one finding, not two.
It is deliberately not
role="alert", which is whyBannergrew anannounceprop. The alerting banners — unauthorized, version skew, a failed read — describe the console being degraded: they arrive unbidden and the page underneath is not what was asked for. Federation health is the opposite. It is the answer the operator navigated here to read, and hijacking a screen reader to deliver requested content is noise dressed as urgency.Keeping the role reserved also keeps
getByRole('alert')meaning one thing across the suite — two version-skew tests turned out to be relying on that without saying so, and went red the moment a second alert existed.The jump-to palette
Ctrl/Cmd-K. Navigation only; there is no write API to reach.
The existing filters are good and stay in the URL, but neither gets you to a page from somewhere else, and the catalog runs to several hundred names. Success here is defined as an answer in under thirty seconds.
Two details are constraints rather than taste:
U+2190-2193, so the hints read↑ ↓ to move, Enter to open, Esc to close. A command glyph or a return arrow would be drawn by whatever the machine fell back to — the exact defect this repo already fixed once for the arrows. There is a test asserting those code points stay out.polish.spec.tsalready covers that, so I removed the duplicate assertion I had written; it kept failing on the trap that file is documented around.Budget
249 KiB of 320, from ~236. Both features together are about 13 KiB.
dist/rebuilt in this commit.One thing the a11y suite learned
scan()settled animations once, then scanned. That was enough while every banner rendered with the route; this one renders when the data lands — after<main>became visible and after the first settle resolved — so axe intermittently sampled it mid-fade and reported a composited colour as a contrast failure. It now settles, waits two frames, and settles again. The same gap applies to anything that arrives with a query rather than a route, so this is a fix for the suite generally, not just for this banner.Branched from
mainrather than the currentdocs-linksbranch, so it does not carry #10.🤖 Generated with Claude Code