fix(ui): Affinity Debug toggle and mobile group-row aria-expanded (#254)

[dborup-agent]

Relates to #254

Follow-up to #248 (#189), findings F1 and F2 from its review. Plan: comment on #254.

1. Node page "Affinity Debug" toggle

Bug: the heading's inline onclick held the Phosphor sprite markup with unescaped double quotes. The browser cut the attribute short, the handler was a syntax error, and the body never opened. The carets were also reversed against the #189 convention (caret-down collapsed, caret-up open).

Fix (public/nodes.js, public/style.css):

• The card comes from renderAffinityDebugCard(). The heading holds a <button type="button" class="affinity-debug-toggle" aria-expanded aria-controls="affinityDebugBody">. The body uses the hidden attribute.
• Caret: caret-right collapsed, caret-down expanded, from one helper, so the render and the toggle cannot disagree.
• toggleAffinityDebug(btn) flips aria-expanded, the body and the caret. One delegated click listener on #nodeFullBody is wired once in init(). It is not wired in loadFullNode, which re-runs on theme-refresh and would stack listeners. As a <button>, it handles Enter and Space natively.
• No inline handler is left in the card. The button keeps the h4 look (font/color/letter-spacing/text-transform: inherit) and gets a 48 px min-height on coarse pointers. No new colours.
• The /api/debug/affinity fetch and its rendering are unchanged.

2. Mobile group rows: select-hash, no aria-expanded

Bug: under the 600 px breakpoint, mobile-page-actions.js (Kpa-clawbot#1461 #7) rewrites a group-row click to select-hash, and the expand column is hidden. The row still announced aria-expanded="false". Also, Enter/Space on a focused row still went through toggle-select and expanded the group (the redirect only catches click). The E2E confirmed this on master.

Choice: render the row as select-hash under the breakpoint, rather than only dropping the attribute. buildGroupRowHtml derives data-action and aria-expanded from one flag:

• aria-expanded is present if and only if the row toggles;
• on mobile, tap, screen-reader activation, Enter and Space all select, which matches what a tap already did;
• dropping only the attribute would have left a row that expands on Enter with no announced state.

Details:

• One breakpoint: mobile-page-actions.js exports window.MobilePageActions = { isMobile }, so the 600 px value stays in one place. Without that module nothing redirects, so the row toggles.
• Tufte v2: mobile packets view — kill chrome, kill duplication (follow-up to #1458 / #1459) Kpa-clawbot/CoreScope#1461 API: harden configured-scope contract for Local Mesh consumers #7 redirect is unchanged. It still covers a row rendered before a breakpoint crossing.
• Resize: a debounced (150 ms) resize listener, added in the page init and removed in destroy, re-renders the visible rows only when the breakpoint is actually crossed (e.g. a phone rotating 390 ↔ 844 px).
• Desktop is unchanged from fix(packets): collapsed groups show a right caret, expanded a down caret (#189) #248: toggle-select, caret and aria-expanded.

Perf:

• groupRowSelectsOnActivate() is one innerWidth comparison per rendered row, and only the visible slice is rendered (virtual scroll).
• A breakpoint crossing costs one visible-slice rebuild (tens of rows). Nothing runs per packet, and there are no new API calls.

Tests

Tests come first: two red commits, then the fix.

• test-issue-254-affinity-debug-toggle.js (new, in test-all.sh) runs the real nodes.js in a vm sandbox. It checks:
  • the card markup (button, aria-expanded="false", aria-controls on the hidden body, one caret-right, no on*= attribute);
  • the toggle opening and closing;
  • the delegated handler toggling, and ignoring other clicks.
• test-packets.js gets 7 new cases:
  • at 390 px and at 600 px, a group row is select-hash without aria-expanded, also when expanded;
  • at 601 px and 1400 px it stays the test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 toggle;
  • single-observation rows are unchanged;
  • without mobile-page-actions.js, a row toggles.
• test-issue-254-affinity-toggle-mobile-aria-e2e.js (new, in deploy.yml after the test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 caret E2E):
  • node page with meshcore-affinity-debug=true, and /api/debug/affinity answered by page.route, so no request reaches the server and no API key is sent;
  • click, Enter and Space, in light and dark;
  • packets: 1400 px toggle; 1400 → 390 → 1400 resize; 390 px touch, where tap and Enter open the detail sheet and do not expand the group.

SCREENSHOT_DIR saves screenshots.

Gates

• scripts/check-xss-sinks.sh --diff origin/master is clean.
• git diff --check is clean.
• No hardcoded colours.
• Fork guards are unchanged: 9 in deploy.yml, 1 in release-fast-path.yml.

🤖 Generated with Claude Code

[dborup-agent]

Rapport — CS-pve-agent3 PR#255 #254 — head 6f09368

Status: Both #254 items are fixed on a draft PR. Every requirement has a unit and/or E2E test, all 8 mutants are killed, local suites are green and CI is green on attempt 1. The PR stays a draft.

Evidence tags: [T] test or run output, [A] assessment or inference, [K] checked in code, diff, git or CI.

Branch

codex/issue-254-affinity-toggle-mobile-aria from origin/master c12aa60b, three commits, test-first. All have author and committer dborup <kontakt@meshview.dk>. No rebase, amend or force-push; only explicit git add. [K]

1. 883c80b3 unit tests (red on master).
2. defbefaa E2E, registered in deploy.yml right after the test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 caret E2E (red on master).
3. 6f09368f the fix. It also fixes a helper in the new unit test that kept the ph- prefix.

The plan is posted on #254. Part 2 uses the "render select-hash on mobile" option. The reason, verified on master: the Kpa-clawbot#1461 #7 redirect only catches click, so Enter/Space on a mobile row still expanded the group. Dropping only aria-expanded would have left a row that expands with no announced state. [T][A]

Requirements

Requirement	Test	Mutant(s)	Result
Card shown only with debugAffinity (flag meshcore-affinity-debug=true)	unit: card renders display:none; E2E: shown with the flag	—	✅ [T]
Heading folds the body in and out	unit: toggle and delegated handler open and close; E2E: click, click, Enter, Space, light and dark	M1, M6	✅ [T]
caret-right collapsed, caret-down expanded	unit: exactly one caret per state; E2E: caret per step	M2, M2b	✅ [T]
aria-expanded follows the state; aria-controls names the body	unit + E2E	M1, M6	✅ [T]
No inline handler with nested markup	unit: no on*= attribute in the card; E2E: no on* attribute on any element in the card	M1	✅ [T]
No calls to /api/debug/affinity in tests, no API key	E2E: page.route answers every request; asserts ≥1 request and an empty X-API-Key. The unit test never fetches	—	✅ [T]
Mobile (≤ 600 px, touch): a group row that selects has no aria-expanded	test-packets.js: 390 and 600 px, collapsed and expanded; E2E 390 px touch: tap and Enter open the detail sheet, the group does not expand	M3, M3b	✅ [T]
Desktop keeps aria-expanded as in #248	test-packets.js: 601 and 1400 px; E2E 1400 px; #189 E2E 3/3	M4	✅ [T]
Breakpoint crossing re-renders	E2E: 1400 → 390 → 1400 resize	M5	✅ [T]
Kpa-clawbot#1461 #7 redirect intact	the code is unchanged; test-issue-1461-mobile-page-actions.js 6/6	—	✅ [K][T]
test-issue-189-group-caret-e2e.js and test-packets.js stay green	3/3; 139 passed, 0 failed	—	✅ [T]
XSS gate, no hardcoded colours, fork guards	check-xss-sinks.sh --diff origin/master exit 0 locally and in CI; no hex or rgb() added; guards: 9 in deploy.yml, 1 in release-fast-path.yml	—	✅ [T][K]
New E2E registered next to the #189 line	deploy.yml line after test-issue-189-group-caret-e2e.js; test-test-all.js 8/8	—	✅ [K][T]

Red on master [T]

• test-issue-254-affinity-debug-toggle.js: 1 passed, 8 failed.
• test-packets.js: 136 passed, 3 failed (the 390 px and 600 px cases).
• New E2E: 3 passed, 9 failed. On master, Enter on a 390 px group row expanded it (aria-expanded="true", 3 child rows).

Mutants [T]

Each mutant ran against both unit files, the new E2E and the #189 E2E, with a server restart per mutant. Each was then restored (git checkout -- public/, tree clean).

Mutant	nodes unit	test-packets.js	#254 E2E	#189 E2E
M1 old inline onclick markup back	7 failed	green	5 failed	green
M2 carets swapped (down collapsed / right expanded)	3 failed	green	5 failed¹	green
M2b master's down/up carets	3 failed	green	5 failed¹	green
M3 aria-expanded back on mobile rows	green	3 failed	3 failed	green
M3b toggle-select back on mobile rows	green	3 failed	4 failed¹	green
M4 aria-expanded dropped on desktop	green	4 failed	2 failed	3 failed
M5 resize re-render removed	green	green	1 failed	green
M6 delegated listener not wired in init()	green	green	3 failed	green

All 8 are killed by at least one layer. ¹ The log line was truncated, so the count is the ✗ markers visible.

Suites (local Go server on a copy of e2e-fixture.db, prepared as in CI: freshen, seed SQL, corescope-migrate, seeds 2073 and 199; stopped by port) [T]

Suite	Result
sh test-all.sh	219/219 files
node test-frontend-helpers.js	707 passed, 0 failed
node test-packets.js	139 passed, 0 failed
New #254 E2E	12/12
Packets E2Es: #189, Kpa-clawbot#1486, Kpa-clawbot#1122 Details clamp + filter UX, Kpa-clawbot#1128 layout + multi-viewport, #96, #147, #180, Kpa-clawbot#1692, slide-over Kpa-clawbot#1056 / Kpa-clawbot#1168, filter UX, Kpa-clawbot#1758, touch gestures, packet-trace alignment, mobile VCR Kpa-clawbot#1267	all exit 0
Nodes E2Es: Kpa-clawbot#2073, Kpa-clawbot#1147, #199, nodes export, favourite re-render, node reach, Kpa-clawbot#2027, node liveness	all exit 0
Touch targets (test-touch-targets.js, Kpa-clawbot#2052)	exit 0, 11/11
test-a11y-axe-1668.js	0 violations, 120 cells
test-e2e-playwright.js	132/135, 3 skipped

Screenshots (local, Chromium, SCREENSHOT_DIR; not attached, because the CLI cannot upload images) [T]

• Light, 1400 px, collapsed: the card heading "AFFINITY DEBUG" keeps the muted uppercase h4 look, with a single right caret before the magnifying glass and no body below.
• Light and dark, expanded: one down caret. The body shows the mocked data ("No affinity edges for this node", Graph Stats, "Cache age: E2E-254 mock"). The dark theme uses the card's normal tokens, and no colour is added.
• 390 px (touch): the seeded group row has no expand column. A tap opens the bottom detail sheet ("Observations (3)"), and the row stays collapsed.

CI [K]

Run on head 6f09368f, attempt 1, conclusion success. No reruns.

• ✅ Go Build & Test: test-all.sh 219/219, including test-issue-254-affinity-debug-toggle.js; the PR-only XSS --diff preflight ran.
• ✅ Playwright E2E Tests: the Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254 E2E 12/12 and the test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 E2E 3/3; the bug(packets): filter UX disaster — help panel overlaps table, toolbar chaotic, path chips spill rows Kpa-clawbot/CoreScope#1122 Details clamp (test: Details-clamp E2E 'advert links in Details stay visible and clickable' is time-dependent and flaky #244) passed 18/18.
• ✅ Build & Publish Docker Image.
• 📦 Release Artifacts, 🚀 Deploy Staging, 📝 Publish Badges: skipped (fork-guarded).

Neither known flake fired (#244 Details clamp, #250 TestStatsFileHasNoCredentials_118).

Remainder

• A group expanded on desktop keeps its child rows after shrinking to mobile, and has no collapse path there. This dead-end existed before (Tufte v2: mobile packets view — kill chrome, kill duplication (follow-up to #1458 / #1459) Kpa-clawbot/CoreScope#1461 API: harden configured-scope contract for Local Mesh consumers #7). The row now announces no state. [A]
• The Tufte v2: mobile packets view — kill chrome, kill duplication (follow-up to #1458 / #1459) Kpa-clawbot/CoreScope#1461 API: harden configured-scope contract for Local Mesh consumers #7 click redirect now only matters for rows rendered before a breakpoint crossing (inside the 150 ms debounce). It is deliberately left unchanged; a later cleanup could drop it. [A]
• Open/closed state of the Affinity Debug card resets when loadFullNode re-renders (theme-refresh). This happened before too. [K]
• The card still fetches /api/debug/affinity on page load, even while collapsed. Unchanged. [K]
• aria-expanded on role="row" in a plain table: screen-reader output has not been checked (carried over from fix(packets): collapsed groups show a right caret, expanded a down caret (#189) #248). [A]
• Not verified: Firefox, Safari, a real screen reader, staging or production (by design). No customizer values were added.

[adminopenclaw8-sketch]

Review — CS-Macmini PR#255 affinity-toggle — head 6f09368

Dom: APPROVE med nits

Independent, read-only review. I made no changes to the PR. Local runs used git archive of the head, of origin/master d05b0db5 and of the merged tree (git merge-tree --write-tree origin/master 6f09368f… → dcec26c5, no conflicts), in scratch directories. git ls-remote showed the head at 6f09368f before and after the review. During the review, master moved to 310c501a (#252). That change touches only test-issue-1122-details-row-clamp-e2e.js, which does not overlap this PR, and merge-tree against 310c501a is also clean. I did not rerun the suites on it. [K]

Evidence tags: [T] test or run output, [A] assessment or inference, [K] checked in code, diff, git or CI.

Findings

#	Severity	Finding	Evidence
F1	nit (test gap)	The new unit test does not prove the card actually uses the renderer. It exercises renderAffinityDebugCard() in isolation. My mutant M1, which puts master's inline onclick card back into the loadFullNode template and leaves the renderer exported but unused, passes test-issue-254-affinity-debug-toggle.js 9/9. Only the E2E kills it (5 failed). Relatedly, "no inline event handler anywhere in the card" is green on master (1 of the 9 master cases passes), because the renderer is missing there and the test scans an empty string. CI runs the E2E, so this is a gap in the fast layer only. A one-line source assertion would close it: loadFullNode's template contains ${renderAffinityDebugCard()}, or nodes.js has no onclick= near node-affinity-debug.	[T] M1 · [K]
F2	nit, pre-existing (follow-up)	An expanded group that crosses to ≤ 600 px stays expanded, with no state and no collapse path. At 1400 px I expanded the seeded group, then resized to 390 px. The row re-rendered as select-hash without aria-expanded, but it keeps class="expanded" and its 3 child rows stay visible. The expand column is hidden, so there is no way to collapse it there. Before this PR the same row announced a stale aria-expanded="true", so this is no regression. The author lists it under "Remainder". A follow-up could collapse expandedHashes when the mode flips to mobile, or keep the children but drop them from the visible slice.	[T] reviewer check · [K]
F3	nit (test gap)	The E2E covers Enter on a mobile row but not Space. The handler treats both the same (packets.js, the keydown branch with Enter/' '). My own check pressed Space on the focused 390 px row: the sheet opened and the group did not expand.	[T]
F4	info, pre-existing, out of scope	nodes.js init() still adds a document-level keydown listener (nodesEsc) per node-page init. It is only removed when Escape is pressed, so navigating A → B → C stacks three. Escape then sets the same hash three times, which is harmless. This PR does not touch it, and the new #nodeFullBody click listener does not leak (see 1).	[K]

No blocking findings.

1. Affinity Debug

• Shown only with debugAffinity: [T]
  • With meshcore-affinity-debug=true the card shows.
  • Without the flag, the card stays display:none and no /api/debug/affinity request is made (reviewer check, route counter = 0).
  • The CLIENT_CONFIG.debugAffinity path is unchanged code. [K]
• Click, Enter and Space toggle it: [T] The PR's E2E passes 12/12 locally and in CI. My own checks also cover a tap at 390 px with touch, in light and dark.
• Exactly one caret per state: [T][K]
  • Collapsed shows ph-caret-right and expanded shows ph-caret-down. One <use> sits in .toggle-icon, from the single helper affinityDebugCaretHtml, so the first render and the toggle cannot disagree.
  • After a theme-refresh while the card is expanded, it re-renders consistently collapsed (false, hidden, caret-right).
• aria-expanded / aria-controls: [T][K]
  • aria-expanded sits on the <button> and follows the body.
  • aria-controls="affinityDebugBody" names the body's id, and the body uses hidden.
  • There is one node page at a time, so the fixed id is unique (#nodeFullBody count = 1).
• No inline on* handler: [T] The reviewer check found 0 on* attributes on the card subtree after several re-renders.
• No listener leak (checked myself): [T] I wrapped EventTarget.prototype.addEventListener in an init script and recorded every registration of onFullBodyClick. Then I navigated node A → B → A → #/packets → A and fired theme-refresh three times. Results:
  • 4 registrations, one per init();
  • all on #nodeFullBody, click;
  • only one on a connected element, because the old bodies are discarded with app.innerHTML;
  • afterwards, one click opens the card and one click closes it.
• Two reasons it cannot leak: [T][K]
  • init() rebuilds app.innerHTML, so every node page gets a fresh #nodeFullBody and the old element is discarded with its listener.
  • Within one page, theme-refresh re-runs loadFullNode, which replaces only the body's content, not the element. Even re-adding the same named handler would be de-duplicated (mutant M3).
• Styling: [T][K]
  • The button inherits the h4 look: the computed color equals the h4 colour in both themes, and background is transparent.
  • On coarse pointers the button is 48 px tall.
  • Keyboard focus shows the global 2 px focus ring.
  • The light-blue flash right after a tap is Chromium's default -webkit-tap-highlight-color, the same as on the back button. It is gone after 2 s.
  • No new colour values.

2. Mobile: select-hash below the breakpoint

• Is the choice right? Yes. [T][A]
  • On master, Enter on the focused 390 px row expanded the group (aria-expanded="true", 3 child rows). The PR's E2E reproduces this on master, and I reran it: 3 passed, 9 failed.
  • The Tufte v2: mobile packets view — kill chrome, kill duplication (follow-up to #1458 / #1459) Kpa-clawbot/CoreScope#1461 API: harden configured-scope contract for Local Mesh consumers #7 redirect only listens for click. So dropping only the attribute would have left a row that expands without announcing it.
  • Rendering select-hash makes tap, Enter and Space all select. That matches what a tap already did, and the row's markup now describes its behaviour.
• Breakpoint: [K][T]
  • groupRowSelectsOnActivate() calls MobilePageActions.isMobile() (innerWidth <= 600).
  • The CSS that hides #pktTable td.col-expand is @media (max-width: 600px). Both include the scrollbar, so they agree.
  • Touch contexts at 600 / 601 px: 600 px is select-hash with the column hidden; 601 px is toggle-select with aria-expanded and the column visible.
  • Mutant M4 (own < 600 check) is killed by the unit test.
• Breakpoint crossing: [T]
  • The PR's E2E covers 1400 → 390 → 1400.
  • My checks add a rotation from 390×844 to 844×390 and back. Each side renders correctly.
  • A fast 1400 → 390 → 1400 flip inside the 150 ms debounce also ends as the desktop row: the timer is cleared, and the mode compare is a no-op.
  • destroy removes the listener and clears the timer. [K]
• Touch device wider than 600 px: consistent. [T]
  • Touch contexts at 601, 620, 768 (tablet portrait) and 1024 (tablet landscape) all render toggle-select with aria-expanded="false" and a visible expand column.
  • A tap on the caret expands (true, children shown), and a second tap collapses (false).
  • At 601–640 px the detail opens in the mobile sheet, and at 641–1023 px in the slide-over. Both cover the table until they are closed. This is unchanged from master and is not caused by this PR. [A]

3. Desktop is unchanged from #248

Reviewer check at 1400 px on the seeded group. After expanding it, aria-expanded stays "true" and class="expanded" stays, with the child rows present, through each of: [T]

• theme-refresh (renderTableRows);
• sorting by Time twice;
• a filter expression that matches the group (type == GRP_TXT), and clearing it;
• a simulated live update: a packet message for the same hash pushed through the page's WS listeners. Children went 3 → 4, and the row kept "true".

After collapsing, a further live update keeps "false".

test-issue-189-group-caret-e2e.js: 3/3 locally on the merged tree, and green in CI. [T]

4. Tests

Red on master, green on head/merged [T]

Test	master d05b0db5	merged dcec26c5
test-issue-254-affinity-debug-toggle.js	1 passed, 8 failed	9/9
test-packets.js	136 passed, 3 failed (390 / 390-expanded / 600 px)	139/139
test-issue-254-affinity-toggle-mobile-aria-e2e.js	3 passed, 9 failed	12/12

• The E2E mocks /api/debug/affinity with ctx.route and never sends a key. I confirmed that every affinity request in my own runs carried an empty X-API-Key (≥ 1 request). [T][K]
• The unit tests run the real code. test-packets.js loads public/packets.js and public/mobile-page-actions.js into a vm sandbox and varies innerWidth. The Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254 unit test loads the real public/nodes.js. [K]

Mutants

Each mutant is a copy of the merged tree's public/. For each one I ran the #254 unit test, test-packets.js, the #254 E2E and the #189 E2E against a fresh server.

Mutant	#254 unit	test-packets.js	#254 E2E	#189 E2E	Reviewer checks	Verdict
M1: master's inline onclick card put back into the loadFullNode template (renderer exported but unused)	9/9 green	green	5 failed	green	no button, script aborts	killed by the E2E only (F1)
M1b: renderAffinityDebugCard() returns master's inline markup	7 failed	green	5 failed	green	aborts	killed
M2: aria-expanded back on mobile group rows	green	3 failed	3 failed	green	4 failed	killed
M3: delegated listener added in loadFullNode (every render) instead of init	green	green	12/12	green	1 failed (call count only)	equivalent, see below
M4: own innerWidth < 600 check instead of MobilePageActions.isMobile()	green	2 failed	12/12	green	1 failed (600 px touch)	killed by the unit test
M5: resize listener not added in init	green	green	1 failed	green	2 failed	killed
M6: carets swapped	3 failed	green	5 failed	green	aborts	killed
M7: toggle flips aria-expanded and the caret but leaves the body hidden	2 failed	green	3 failed	green	1 failed	killed
M8: resize re-render without resetting _lastVisibleStart / _lastVisibleEnd	green	green	1 failed	green	2 failed	killed

M3 is behaviourally equivalent. addEventListener with the same function reference, type and capture flag on the same element is de-duplicated by the DOM. So re-adding the named onFullBodyClick on every loadFullNode registers it once. My call-counting check saw 4 calls on the live #nodeFullBody, but one click still toggled exactly once. The code comment "would stack listeners" is therefore stricter than necessary for a named handler, though it would apply to an inline closure. Wiring it in init() is still the cleaner choice. [T][A]

All non-equivalent mutants are killed by the PR's own tests. M1 is killed only by the E2E (F1). [T]

5. Rules

• No hardcoded colours. The added lines contain no hex, rgb() or hsl() values. The CSS uses inherit and none. [K]
• XSS gate: scripts/check-xss-sinks.sh --diff origin/master on the head exits 0. I ran it in a temporary detached worktree, which I removed afterwards. git diff --check is clean. [T]
• Fork guards: github.repository == 'Kpa-clawbot/CoreScope' occurs 9 times in deploy.yml and once in release-fast-path.yml, the same as master. [K]
• deploy.yml: exactly one added line, the Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254 E2E right after the test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 caret E2E. [K]
• Closing keywords: none in the title, body or commit messages. The PR says "Relates to Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254", closingIssuesReferences is empty, and Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254 is still open. [K]
• Commit author: all three commits have author and committer dborup <kontakt@meshview.dk>. [K]
• No new map[string]interface{}: the PR changes no Go code. [K]

Suites (merged tree, against origin/master d05b0db5)

Suite	Result
sh test-all.sh	219/219 files [T]
node test-frontend-helpers.js	707 passed, 0 failed [T]
node test-packets.js	139 passed, 0 failed [T]
#254 E2E / #189 E2E / test-issue-1461-mobile-page-actions.js	12/12, 3/3, 6/6 [T]
Packets E2Es: Kpa-clawbot#1486, Kpa-clawbot#1122 clamp (18/18) + filter UX, Kpa-clawbot#1128 layout + multi-viewport, #96, #147, #180, Kpa-clawbot#1692, slide-over Kpa-clawbot#1056 / Kpa-clawbot#1168, filter UX, Kpa-clawbot#1758, touch-gestures coverage, packet-trace alignment	all exit 0 [T]
Nodes E2Es: Kpa-clawbot#2073, Kpa-clawbot#1147, #199, nodes export, favourite re-render, node reach, Kpa-clawbot#2027, node liveness; touch targets, Kpa-clawbot#2052	all exit 0 [T]
test-e2e-1267-mobile-vcr.js	not run: it needs /usr/bin/chromium, which this macOS host lacks [T]
test-a11y-axe-1668.js	local net=1 (mobile light /observers colour contrast on "connected") and identical on master. CI on this head reports net=0. Local and time-dependent, not this PR. [T][K]
test-e2e-playwright.js	local fail-fast at "Version info lives on Perf dashboard" (the #navStats wait times out), identical on master. It passes in CI on this head. [T][K]

The local Go server ran on a copy of e2e-fixture.db, prepared as in CI: freshen, the seed SQL from deploy.yml, corescope-migrate, then seeds 2073 and 199. I stopped it by port.

CI: run on head 6f09368f, attempt 1, conclusion success. The log shows the #254 E2E at 12/12, axe at net=0, and Version info ✅. [K]

Screenshots (local Chromium; described, not attached)

• Light, 1400 px, expanded: "AFFINITY DEBUG" keeps the muted uppercase h4 look, with a single down caret before the magnifying glass. The body shows the mocked "No affinity edges for this node" and Graph Stats. [T]
• Dark, 1400 px, collapsed: a single right caret, no body, and the card uses the normal dark card tokens. [T]
• 390 px, touch, light and dark, expanded after a tap: the heading button spans the card at 48 px height with one down caret. The body text wraps cleanly, and nothing overflows horizontally. [T]
• 390 px packets, light and dark: the seeded group row shows Time / Type / Details with no expand column. A tap opens the bottom sheet, and the row stays collapsed. [T]

Not verified

• Firefox, Safari / WebKit, a real phone or tablet, and a real screen reader. In particular, I have not heard how aria-expanded on a role="row" in a plain table is announced; this is carried over from fix(packets): collapsed groups show a right caret, expanded a down caret (#189) #248.
• A real WebSocket stream: the live update was injected through the page's WS listeners.
• test-e2e-1267-mobile-vcr.js locally (no /usr/bin/chromium).
• The full test-e2e-playwright.js locally (fail-fast on an env-only failure that also fails on master); I relied on CI for it.
• Staging and production, by design.