test: run test-all.sh in CI with a summary (#174 follow-up)

[adminopenclaw8-sketch]

Relates to #174, #173, #177, #178

test-all.sh now runs every file and prints a summary. CI runs it as the JS unit step, and a self-test makes sure that no root test file can go unrun again. It also covers the review nits from #173, #177 and #178, and an inventory of the root tests that no runner registered.

No production code changed (git diff origin/master...HEAD -- public/ is empty), and no npm dependencies were added. Fork guards are unchanged: 9 × github.repository == 'Kpa-clawbot/CoreScope' on both base and head. No guard lines, triggers, permissions or concurrency were touched.

Base: cbd562cd, with origin/master (bedbe1f1, #167) merged in through a normal merge commit (535d6e88). The only conflict was the one-line test-issue-147-packets-url-detail-params.js registration in test-all.sh.

1. test-all.sh: every file runs, with a summary at the end

• set -e is gone. Each file runs through run <file> (POSIX sh). The script counts passes and failures, lists the failed files at the end and exits 1 if any file failed.
• test-live-anims.js is registered. It was repaired in test: issue 174 part A — emoji gate, live anim, naive banner, observers headings #177 but never registered.
• test-test-all.js is the self-test. It runs the real test-all.sh under /bin/sh and dash (when present) in a temp dir, with its file list swapped for fixture files. It checks that:
  • a failing file does not stop the run;
  • a thrown error, a non-zero exit and a missing file are all listed;
  • the exit code is non-zero, and an all-green run exits 0;
  • every registered file exists and is registered only once;
  • no registered file loads playwright, launches Chromium or reads BASE_URL.
• test-privacy-page.js asserted the literal set -e line and node test-privacy-page.js. It now asserts the run line and the exit-1 path, which is the property it was guarding.

Demo in a copy, with two deliberately failing files (throw at the top of test-aging.js, and process.exitCode = 1 in test-hash-color.js): all 200 files ran, and the run ended with exit 1:

═══════════════════════════════════════
  198 passed, 2 failed (200 files)
  Failed:
    test-aging.js
    test-hash-color.js
═══════════════════════════════════════

2. CI: one list, sh test-all.sh replaces the JS unit lines

The "Run JS unit tests" step in Go Build & Test listed 88 files. 44 of them were also in test-all.sh, and 44 were only in deploy.yml. The 64 other test-all.sh files never ran in CI. I moved the 44 deploy-only files into test-all.sh and replaced the step body with node --version and sh test-all.sh. The step fails if any file fails: sh exits 1, and the job shell is bash -e.

Timings with Node 22.22.1, run locally, 3 runs each:

Variant	Files	Time
old step (deploy.yml lines)	88	~7.9 s
test-all.sh before this PR	108	~8.9 s
chosen: one test-all.sh step (after moving the deploy-only files)	152	~12.2 s
alternative: keep the old lines and add a test-all.sh step	88 + 108	~16.8 s, 44 files twice
final test-all.sh (with the inventory below)	200	~14.7 s

On master, CI took about 5 s for the old step (run 37105041262, timestamps at second resolution). The chosen variant has no double runs and leaves one list to maintain. The CI unit job has no npm ci, so I also ran test-all.sh without node_modules: 200/200.

3. Inventory: root tests that no runner ran

On cbd562cd, 70 root test-*.js files were run by neither test-all.sh nor deploy.yml, and 8 of them have "e2e" in the name. A 71st, test-rx-coverage-mobile-nav-e2e.js, appears in deploy.yml only in a comment. The non-E2E files were run with Node 22, without node_modules (as in the CI unit job), with a 120 s limit.

test-test-all.js now also fails when a root test-*.js runs in neither test-all.sh (run <file>) nor deploy.yml (node <file>), unless it is listed in KNOWN_UNREGISTERED with a reason. An entry that is deleted, or that gets registered, has to be removed, so the list can only shrink.

Registered: 49 unit files and 4 E2E files

• 45 green unit files, registered as-is. Three full test-all.sh runs were green.
• test-live-anims.js (see section 1).
• 3 stale unit files, each corrected in a few lines in its own commit with mutants (see the mutants table):
  • test-map-clustering.js: the pills read R3/C2/M1 since 40aa02b4 (regression(map): #1357 cluster role pills lost the count number — show letter+count not letter-only Kpa-clawbot/CoreScope#1360);
  • test-panel-corner.js: the sandbox did not load payload-labels.js before live.js (b74a64cc, Label vocabulary drift across packets / live / packet-filter views Kpa-clawbot/CoreScope#1799);
  • test-customizer-v2.js: the fixture lacked home, which computeEffective has filled in since 54fab055 (Bug: Customizer v2 home section empty + adding FAQ kills steps Kpa-clawbot/CoreScope#525).
• 4 E2E files, registered in the Playwright step. All were green 3/3 locally (Chromium 1208), and none uses the fixture server:
  • test-a11y-1705-subpath-hop-prefix-e2e.js;
  • test-a11y-1719-contrast-root-causes-e2e.js;
  • test-nodes-favorite-rerender-e2e.js;
  • test-packet-trace-alignment-e2e.js.

The 45 green unit files

test-1108-region-hide-nodes.js, test-a11y-1715-dark-role-swatches.js, test-analytics-distance-view-path.js, test-analytics-wardriving-tab.js, test-anl1-tooltip-render.js, test-channel-color-picker.js, test-channel-decrypt-ecb.js, test-channel-decrypt-m345.js, test-channel-live-decrypt-userprefix.js, test-channel-live-decrypt.js, test-color-picker-ux.js, test-compare-flood-filter.js, test-compare-overlap.js, test-embed-mode-1369.js, test-geofilter-draft.js, test-hash-color.js, test-issue-1166-first-seen-column.js, test-issue-1189-composed-cell.js, test-issue-1189-live-iata-badge.js, test-issue-1415-packets-layout.js, test-issue-1488-marker-stroke-vars.js, test-issue-1496-reset-all-complete.js, test-issue-1563-aggregate-and-inflight.js, test-issue-1574-live-map-max-nodes.js, test-issue-1606-pagination.js, test-issue-1644-redesign.js, test-issue-1648-followup-phosphor-leaks.js, test-issue-1648-m5-emoji-scan.js, test-issue-1668-m2-contrast.js, test-issue-1825-observer-node-cross-links.js, test-issue-1836-crossnav-case-normalization.js, test-issue-1843-node-qr-quiet-zone.js, test-live-dt-cap-1524.js, test-live-legend-helper.js, test-node-reach-coverage-debounce.js, test-observer-detail-view-path.js, test-observer-naive-clock-1478.js, test-path-inspector.js, test-payload-labels-content.js, test-payload-labels-namespace.js, test-perf-anomaly.js, test-perf-render-1258.js, test-pull-to-reconnect.js, test-rx-coverage-config-race.js, test-slideover-1056-rowsel-strict.js

Red unit files: not fixed here, proposed for a follow-up issue

The causes were checked against git history. Where the table says "test passes at X^ and fails at X", that was run.

File	First failure	Likely cause
test-channel-colors.js	getRowStyle … assigned GRP_TXT channel: should have left border (19/2)	68a4628e (Kpa-clawbot#675) changed the row style from a 4px border plus tint to a 3px border only. Intentional; stale test.
test-channel-ux-followup.js	new copy points at the ✕ button for individual key removal (28/1)	b812a98a (Kpa-clawbot#1648 M3) reworded the hint to "Use the close button…". Stale.
test-channel-ux-round2.js	iconBtn('ch-share-btn', ...) is called with '📤 Share' (13/1)	b812a98a (Kpa-clawbot#1648 M3): the label is now the #ph-share-network sprite plus " Share". Stale.
test-drag-manager.js	Escape during drag reverts to corner position (18/1)	2b45f787 (Kpa-clawbot#1567) switched to removeAttribute('data-dragged'). The test's mock does not update dataset; a real DOM does. Mock gap.
test-fluid-scaffolding.js	spacing token --space-xs declared with clamp() (2/13)	f0addfda (Kpa-clawbot#1668) added a palette :root block first, and the test only reads the first :root. The tokens still exist.
test-hop-resolver-affinity.js	Should pick NodeA (geo-closest) — got: NodeB (16/1)	2b9f3056 (Kpa-clawbot#874), haversine centroid. The test passes at 2b9f3056^ and fails at 2b9f3056. With haversine, NodeB really is closer (70.88 vs 71.06 km), so the fixture geometry is wrong.
test-issue-1470-card-bg-contrast.js	[data-theme="dark"] block sets --card-bg…: not found (9/1)	e2212f50 (Kpa-clawbot#1627) added a style.css comment containing [data-theme="dark"], and indexOf lands on the comment. Passes at e2212f50^.
test-issue-1646-compare-polish.js	.compare-vs font-size < .compare-select font-size: could not parse (13/1)	d954ea74 (Kpa-clawbot#1668 M5) added a comment with font-size:10px inside .compare-vs, and the parser reads the comment.
test-packets.js	getDetailPreview handles CHAN type (115/13)	30627454 (Kpa-clawbot#1648 M2): 13 emoji assertions; the code now renders #ph-* sprites. Mechanical, but 13 edits.
test-perf-disk-io-1120.js	WAL >100 MB fires ⚠️ flag (10/5)	30627454: ⚠️ is now #ph-warning. The "no ⚠️" negatives pass vacuously. The backfill cases target the old single-snapshot detector, which a26a412c (Kpa-clawbot#1593) replaced. Needs rework.
test-table-sort.js	Cannot find module 'jsdom'	jsdom is not a dependency. Port to vm or delete.
test-marker-outline-weight.js	ERR_MODULE_NOT_FOUND for the Playwright Test runner package	A Playwright Test spec (ESM, page.goto). The runner package is not a dependency, and the spec also needs a server.
test-touch-targets.js	.nav-btn: rendered 44.0x44.0 … expected >=48x48 (5 failures, run with node_modules and Chromium, no server)	The test is from eaf14a61 (Kpa-clawbot#1060, 48px targets), and the current style.css sets 44px minimums. Whether 48 or 44 is the intended target needs a decision. Not classified as stale or as a regression.

E2E files that stay unregistered

All were run against a local server built with CI's fixture recipe: freshen-fixture.sh, the Kpa-clawbot#1486/Kpa-clawbot#1791 seed block from deploy.yml, corescope-migrate, and seed-2073-route-adverts.sql.

File	Result	Should it go in the Playwright step?
test-channel-modal-e2e.js	12/2: "Add Channel button is visible" (button text is now "+ Add"), and "sidebar renders three sections" times out	Yes, after updating it for the chat-app sidebar (Kpa-clawbot#1376)
test-node-reach-e2e.js	TimeoutError waiting for #nqMap .leaflet-container	Unclear. test-node-reach-coverage-e2e.js already runs in CI, so check for overlap first. Cause not investigated.
test-issue-1522-trace-url-sync-e2e.js	needs the Playwright Test runner package (not a dependency)	Only if ported to plain playwright
test-path-inspector-e2e.js	same	Only if ported (test-path-inspector-coverage-e2e.js already runs)
test-rx-coverage-mobile-nav-e2e.js	SKIP: clientRxCoverage is disabled in the fixture config	No; the existing deploy.yml comment already says so

4. Review nits

	Change	Test that catches it
a	test-pr-1490-live-map-gpu-animations-e2e.js: drain bound 5000 → 2500 ms. The fade-pane step now counts Leaflet renderer canvases per pane before and after the burst: exactly one on the animations pane, and none added elsewhere. Before, the contrail alone satisfied the check.	<vp>: the animations drain to 0, <vp>: the fading polylines render on the animations pane only
b	test-rx-coverage-escape.js case 5: mounts the real page, runs createMap's deferred drawCoverage(), answers /api/rx-coverage with hostile node names and an unnamed node with a hostile prefix, and checks the exact escaped labels in the hex tooltip (coverageNodeRow). The cases share a mountPage harness. public/rx-coverage.js is unchanged.	5. hex tooltip: hostile node names and prefixes are escaped
c	test-packet-path-map.js: all four role icons and the first-to-hear #ph-flag are aria-hidden sprite icons at the start of their tooltips, and no tooltip icon lacks aria-hidden="true"	role icons and the first-to-hear flag render as aria-hidden sprite icons
d	test-issue-1407-cb-preset-propagation.js §H: document.body = null, getComputedStyle(null) throws (as in browsers), and the value exists only on documentElement	document.body null: ROLE_COLORS.repeater resolves on documentElement
e	test-issue-1648-m3-emoji-scan.js: 🏓 and 🤖 moved to ALLOW_EXACT['channels.js']. Each must occur exactly once; only that occurrence is cut out, and the rest of the line is still scanned. The 🤖 entry is now the full avatar element.	channels.js clean … / app.js clean …

Note for (a): the commit message of b4c96772 gives 665–682 ms for "10/10 runs against the instrumented frontend". Those 10 runs actually hit a server that still served plain public/, because my local start script stopped the wrong process. I reran with the instrumented build: the served live.js contains __coverage__. All 10 runs were green, at 668–684 ms per viewport. The bound (2500 ms) is unaffected.

Mutants

Each mutant ran in a copy of the tree (rsync, never in the worktree). The E2E mutants were served from a copy of public/ on a second server.

#	Mutant	Old test	New test	Caught by
a1	live.js ~4339: pane: removed from the main fade line only	green	red (desktop + mobile)	the fading polylines render on the animations pane only ("added a renderer canvas to overlay")
a2	pane: removed from the contrail only	green	red	same
a3	animation 660 → 3000 ms (drain ~3 s)	green (5 s bound)	red	the animations drain to 0
b1	coverageNodeRow: name not escaped	–	red	case 5
b2	prefix not escaped	–	red	case 5
b3	label always the prefix	–	red	case 5
b4	escapeHtml without &	–	red	cases 1, 2, 5
c1	phIcon without aria-hidden	–	red	role icons / first-to-hear case
c2	first-to-hear flag removed	–	red	same
c3	flag swapped for another glyph	–	red	same
c4	flag inlined without aria-hidden	–	red	same
c5	sensor icon dropped / client icon dropped	–	red / red	same
d1	getComputedStyle(document.body) only	–	red	§H
d2	documentElement only (pre-#178)	–	red	§G
e1	extra ✅ on the pong line	green	red	channels.js clean
e2	<div>🤖</div> in app.js	green	red	app.js clean
e3	📡 after the 🤖 avatar on the same line	green	red	channels.js clean
e4	second avatar in channels.js / pong text in app.js / 🏓 removed (stale exception)	–	red each	channels.js clean / app.js clean
t1	test-all.sh: old semantics (bare node under set -e)	–	red	failures do not stop the run (sh + dash)
t2	exit 1 removed / failure counted as a pass / failed file not recorded	–	red each	same
t3	duplicate / missing / E2E file registered	–	red each	no file is registered twice / every registered file exists / no registered file needs a browser or a server
t4	new unregistered root test / registered file dropped / E2E line dropped / allowlisted file registered or deleted	–	red each	every root test-*.js runs … / KNOWN_UNREGISTERED has no stale entries
t5	set -e added in front of the new run()	–	green	Equivalent mutant: node runs inside an if condition, which set -e ignores
p	test-all.sh without the exit-1 path / old node registration line	–	red	test-privacy-page.js is registered exactly once in test-all.sh
m	room count off by one / letter swapped / pill without letter	–	red	test-map-clustering.js pill case
pc	occupied corners not skipped / feed default bl → tl / cycle reversed	–	red	test-panel-corner.js
cv	server home always replaced / config dropped / not deep-copied	–	red	test-customizer-v2.js

Possible production bug found, not fixed (public/packets.js is on the do-not-touch list)

When 30627454 (Kpa-clawbot#1648 M2) migrated ▶/▼ to sprites, the collapsed packet-group toggle became #ph-caret-up (public/packets.js:2359), while the expanded one is #ph-caret-down. Collapsed toggles elsewhere use ph-caret-right, which is in the sprite. Collapsed group rows probably show an up-caret. Found while diagnosing test-packets.js, checked in the source and the diff, not checked in a browser.

Other notes

• The emoji regex in the 1648 scans covers U+1F300–1FAFF and U+2600–27BF only, so ⭐ (U+2B50) and other U+2B00 symbols are not caught. Noticed during the e4 mutants; not changed here.
• test-issue-1697-mqtt-mobile-e2e.js has "e2e" in its name but is a vm unit test (it was already in the unit step). It moved into test-all.sh, so the E2E check looks at content (playwright, chromium.launch, BASE_URL), not names.

Verification (local, macOS, Node 22.22.1)

Run	Result
sh test-all.sh (/bin/sh and dash), with and without node_modules	200/200, exit 0
sh test-all.sh with two deliberately failing files (copy)	exit 1, both files listed, 200 run
test-pr-1490-…-e2e.js, plain public/	14/14 (11 runs)
test-pr-1490-…-e2e.js, public-instrumented/	14/14, 10/10 runs, 668–684 ms
4 newly registered E2E files	green, 3/3 each
Fork guards	9 → 9

Not verified

• The packets.js caret: source only, no browser.
• The other E2E files in the Playwright step were not run locally; CI runs them.
• test-touch-targets.js: whether 44px or 48px is the intended target.
• test-node-reach-e2e.js: the cause of the timeout was not investigated.
• Linux /bin/sh (dash) locally: only macOS /bin/dash was tested. CI's ubuntu runner covers it.

🤖 Generated with Claude Code

[dborup]

Review — CS-MacBook review PR#187 test-all i CI — head b23d8d9

Dom: APPROVE med nits (approve with nits)

Independent, read-only review of bedbe1f1..b23d8d9f (14 commits including the master merge 535d6e88, 12 files). I worked on git archive copies, changed nothing in the repo, pushed nothing and touched no shared environment. The head was b23d8d9fd5a03ae99c28350d030a22ec05482716 before and after (checked with git ls-remote).

Tags: [F] browser run, [T] test or CI run, [A] analysis of code, [K] not checked.

No blocking findings. Everything the description claims held up. The nits below are small and none needs to block the merge.

Findings

Prio	Where	Finding
P3	test-test-all.js:157	The registry check finds deploy.yml references with /\bnode (test-…\.js)\b/, which also matches a commented-out line. A new root test whose only "registration" is # node test-x.js in deploy.yml passes the guard (mutant G3). Suggestion: skip lines that start with #. [T]
P3	test-test-all.js:131	KNOWN_UNREGISTERED only gets a "no stale entries" check (:167). A new file can be added to the list and the guard stays green (mutant G4). "The list can only shrink" holds by convention and comment, not by test. A pinned length, for example assert(Object.keys(KNOWN_UNREGISTERED).length <= 18), would enforce it. [T]
P3	.github/workflows/deploy.yml:137	The unit step runs in the Go job, and actions/setup-node appears only in the E2E job (:243, Node 22). The unit step therefore uses the runner's preinstalled Node. The CI log shows v22.23.3 today, but it is not pinned, so a runner image update can change it. Adding setup-node@v5 with node-version: '22' to the Go job would pin it. [T][A]
P3	.github/workflows/deploy.yml:524-525	test-a11y-1705-subpath-hop-prefix-e2e.js and test-a11y-1719-contrast-root-causes-e2e.js use no browser and no server: both exit 0 with PLAYWRIGHT_BROWSERS_PATH pointing nowhere. CHROMIUM_REQUIRE=1 is a no-op for them (they never read it, and neither does test-packet-trace-alignment-e2e.js on :527). They could go in test-all.sh and fail in about 12 s in the Go job instead of after the E2E job starts. The guard would accept them there. [T][A]
P3	test-all.sh:13-22	No per-file timeout, and a file that calls process.exit(0) early counts as passed. Both existed before this PR (each node X line had the same property), so this is not a regression. For a hang, the Go job's 40 minute limit is the only stop. [T][A]

Answers to points 1-6

1. CI step.

• All 88 removed node test-*.js lines are in test-all.sh: 88 unique removed lines, 0 missing, no duplicates. The old test-all.sh files (106) are all still there too. 44 files were in both lists and are now registered once. [A]
• The new step is node --version then sh test-all.sh. The Go job's default shell is bash -e {0} (visible in the log), so the step's status is sh's exit status. test-all.sh ends with exit 1 when any file failed, and exits 0 otherwise. No pipe is involved, so pipefail is irrelevant for this step. [T][A]
• Node: CI ran the step on v22.23.3 (see the P3 on pinning). node_modules is not present when the step runs, and 200/200 pass on a clean git archive without it. [T]
• The suite is order independent: the file list reversed gave 200/200 and exit 0. A full run leaves no stray files in the tree. [T]
• The four new Playwright lines (:524-527): test-nodes-favorite-rerender-e2e.js starts its own server and test-packet-trace-alignment-e2e.js serves its own page. The two contrast gates need neither a server nor a browser, so no BASE_URL is needed. The Playwright job uses shell: bash, which in Actions is bash -eo pipefail, so a failing file fails the step through the | tee. All four pass locally and in CI (PASS lines for 1705 and 1719, all favorite-rerender checks, 17 checks for the trace alignment). [F][T]
• Fork guards: 9 × github.repository == 'Kpa-clawbot/CoreScope' on both base and head. [A]

2. test-all.sh.

• POSIX: the script ran identically under macOS sh (bash in sh mode) and under dash, both 200/200 and exit 0. It only uses $(( )), a function and if/for. [T]
• Every file runs after a failure. Copy with two failing files (one throw, one process.exitCode = 1): 198 passed, 2 failed (200 files), both files listed under Failed:, exit 1. [T]
• No file runs twice or not at all: 200 unique run lines, all files exist. 338 root test-*.js = 200 unit + 120 in deploy.yml + 18 KNOWN_UNREGISTERED, and no overlap between the unit and deploy lists. [A]
• A file that exits 0 early passes (confirmed with process.exit(0) at the top of test-aging.js: 200/200). There is no per-file timeout. See the last P3.

3. test-test-all.js.

• It fails on a new empty root test (9 passed, 1 failed, message names the file) and on a file removed from test-all.sh (mutants G1, G2). [T]
• It does not catch a commented-out deploy.yml reference (G3) and does not stop the exception list from growing (G4). See the P3 findings.
• I ran all 18 KNOWN_UNREGISTERED files. The 10 "red" unit tests exit 1 and the first failing assertion matches the stated reason in each case. The jsdom and @playwright/test entries fail with module-not-found, and package.json has playwright but not @playwright/test or jsdom. [T]
• Not verified: the reasons for the five E2E entries and test-touch-targets.js ("red in Chromium", "skips while clientRxCoverage is off"), because they need a running app and its config. [K]

4. The 45 newly registered tests and the three repaired ones. The block is 48 files: 45 new plus the three repaired files. Each repair follows an intended change in the production code:

• test-map-clustering.js: role pills read letter + count since regression(map): #1357 cluster role pills lost the count number — show letter+count not letter-only Kpa-clawbot/CoreScope#1360. public/map.js:2960-2968 builds letter + n. [A]
• test-panel-corner.js: index.html:177 loads payload-labels.js before live.js at :227, and live.js throws without it. [A]
• test-customizer-v2.js: computeEffective fills in DEFAULT_HOME when the server config has no home, so the fixture now carries one. [A]
• None of them weakened: the assertions are the same or stricter (>R3< instead of >3<). Mutants U1, U2, U3 are caught by the repaired files themselves. [T]

5. Nits 4a-4e. Each described mutant is caught, and the test that catches it is named in the table below.

• 4a: the canvas-per-pane check catches a fade line that leaves the animations pane even though the contrail stays (E1a). The old count >= 1 check would have passed that. The 2500 ms bound catches an engine that drains in about 4 s (E2), which the old 5000 ms bound would have passed. [F]
• Stability of 2500 ms against public-instrumented/: 10 of 10 local runs pass, drain 667-680 ms on both viewports. 6 more runs with 20 busy loops on a 10-core machine: 663-691 ms. In CI the drain was 742 ms (desktop) and 675 ms (mobile). That is a margin of about 3.4x. The server was confirmed by port, process and working directory, and the served roles.js was instrumented. [F][T]
• 4b, 4c, 4d, 4e: see U4-U10.

6. Merge. 535d6e88 merged master bedbe1f1 cleanly apart from the one registration line. Master has since moved to ae595682 (a Go-only change that touches no file in this PR). git merge-tree of the head against it exits 0 with no conflicts, and no new root test-*.js came in, so the guard still holds. [A]

Test counts

• sh test-all.sh on head, Node v26.0.0 (Node 22 is not installed locally): 200 passed, 0 failed, exit 0 (also under dash, and in reversed order). [T]
• Same suite in CI, Node v22.23.3: 200 passed, 0 failed (200 files), about 12 s. test-test-all.js: 8 passed in CI, 10 locally (it also runs under dash). [T]
• Four new E2E lines: 4 of 4 pass locally and in CI. test-pr-1490-live-map-gpu-animations-e2e.js: 10 of 10 (14 checks each), plus 6 of 6 under load, and 14 of 14 in CI. [F][T]

Mutants

#	Mutant	Caught by
U1	map.js pill drops the role letter	test-map-clustering.js, test-issue-1360-pill-letter-count.js
U2	live.js corner cycle skips a corner	test-panel-corner.js
U3	computeEffective drops nodeColors	test-customizer-v2.js
U4	hex tooltip node name not escaped	test-rx-coverage-escape.js (case 5)
U5	hex tooltip prefix not escaped	test-rx-coverage-escape.js (case 5)
U6	roles.js has no documentElement fallback when <body> is null	test-issue-1407-cb-preset-propagation.js (H)
U7	first-to-hear icon changed from flag	test-packet-path-map.js
U8	tooltip icon loses aria-hidden	test-packet-path-map.js
U9	second emoji added on the allow-listed pong line	test-issue-1648-m3-emoji-scan.js
U10	allow-listed pong text copied into another file	test-issue-1648-m3-emoji-scan.js
U11	test-all.sh exits 0 on failure	test-test-all.js, test-privacy-page.js
U12	set -e added to test-all.sh	survives, equivalent: the run function calls node inside if, so set -e changes nothing
U13	failure not counted	test-test-all.js
U14	failed file not listed in the summary	test-test-all.js
U15	a file registered twice	test-test-all.js
U16	an E2E file added to the unit list	test-test-all.js
E1a	fade line leaves the animations pane (contrail stays)	test-pr-1490-…-e2e.js "animations pane only"
E1b	both fade lines leave the pane	same test
E2	animation engine drains in about 4 s	same file, drain step at 2500 ms
G1	new empty root test	test-test-all.js registry check
G2	file dropped from test-all.sh	test-test-all.js registry check
G3	commented-out deploy.yml reference	not caught (see P3)
G4	new file added to KNOWN_UNREGISTERED	not caught (see P3)

CI on head (run 37109895646)

• Go Build & Test: pass.
• Playwright E2E Tests: pass.
• Build & Publish Docker Image: pass.
• Release Artifacts, Deploy Staging, Publish Badges & Summary: skipped, as expected for a PR.

Not verified

• Node 22 locally (v26 only). The CI result on v22.23.3 stands in. [K]
• Whether every one of the 120 deploy.yml root tests sits in an unconditional step. [K]
• The reasons for the five E2E KNOWN_UNREGISTERED entries and test-touch-targets.js. [K]
• CI timing on slower runners than the one used here, beyond the single CI run above. [K]