LearnNewsExamplesConnect
Frontmatter
assigneesneo-opus-grace
createdAtSep 3, 2026, 9:53 PM
updatedAtSep 4, 2026, 10:06 AM
githubUrlhttps://github.com/neomjs/neo/issues/18231
authorneo-opus-grace
commentsCount0
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]
closedAtSep 4, 2026, 10:06 AM

A click inside the portal's scroll container never becomes actionable

Closed v13.2.0/archive-v13-2-0-chunk-18 enhancementaitestingportal
neo-opus-grace
neo-opus-grace commented on Sep 3, 2026, 9:53 PM

Context

Splitting out the one open item from #17877, whose own prescription shipped in PR #17859 and which is now closed. The residual is not a link-rewrite question and never was — it is a harness-capability question that blocks a whole tier of portal coverage.

Measured by @neo-opus-ada on 2026-08-30 (comment) while chasing a click-through arm:

goto  #/learn/guides/uibuildingblocks/DockLayouts   → h1 "Dock Layouts: One Application, Many Windows"
location.hash = '#/learn/benefits/body/Speed'       → h1 "Extreme Speed"

Hash navigation works — router, exact-id store.get, record fetch and re-render all run in about four seconds. What fails is locator.click, which exceeds the 90 000 ms ceiling at test/playwright/playwright.config.e2e.mjs:95. Playwright never satisfies its visible / stable / receives-events checks for content inside the portal's scroll container.

The Problem

The decisive datum is the control: clicking the sidebar tree item fails identically. That element is the app's own navigation and owes nothing to the link rewrite. A red assertion whose control is also red measures the harness, not the subject — so the click arm was written and correctly reverted rather than shipped as a false "routing is broken" signal.

That leaves a documented hole, in prose, at test/playwright/e2e/portal/LearnLinkRoutingNL.spec.mjs:51-61:

Not covered here, and deliberately not asserted: that CLICKING one of these links arrives at the destination.

The cost is not one missing assertion. No portal e2e arm can currently assert that clicking anything arrives, so every future arm in this tier inherits the same gap and the same temptation to paper over it with force: true. Ada declined to file it at the time because the seats that would pick it up were rate-limited — that constraint has expired, and the measurement should not die with the ticket that hosted it.

The Architectural Reality

  • Content renders into .neo-app-content-component (src/app/content/Component.mjs), inside the portal's scrolling content region.
  • Neo applies DOM deltas from the app worker on the main thread. An element that is repeatedly re-written by delta application can fail Playwright's stable check (two consecutive animation frames at the same box) without ever being invisible — a distinct failure from occlusion.
  • The three actionability checks fail for different reasons and are not distinguishable from the timeout message alone. Which one fails is the whole finding, and nothing has measured it yet.
  • Sibling precedent for a shared harness helper is test/playwright/util/ — learnLinkNavigability.mjs, DeltaCapture.mjs, flickerCensus.mjs. A new helper belongs there, not beside a single spec.

The Fix

  1. Determine which actionability check fails, against the sidebar-tree control rather than the rewritten anchor — the control is the honest subject, because it isolates the harness from the feature.
  2. Fix the cause at the layer that owns it: a scroll/stability helper in test/playwright/util/, or an e2e config/use adjustment if the cause is global.
  3. Only then add the click arm to LearnLinkRoutingNL.spec.mjs and delete the prose gap at :51-61.

Acceptance Criteria

  • The failing actionability check is named — visible, stable, or receives-events — with the measurement that identifies it, not inferred from the timeout text
  • The control passes first: clicking the portal sidebar tree item navigates, in the e2e harness, before any assertion is added about an in-content link
  • A click on a rewritten in-content link arrives at its destination (destination h1 asserted), and the deliberate-gap comment at test/playwright/e2e/portal/LearnLinkRoutingNL.spec.mjs:51-61 is deleted in the same PR
  • The arm is proven able to fail: deleting the rewriteLinks invocation or the portal's contentRoute reddens it — the mutation the sibling href arms already survive
  • No force: true, no dispatchEvent, and no raised timeout: the arm exercises what a reader's pointer does
  • If the cause is general to the tier, the remedy is a shared helper under test/playwright/util/ rather than a per-spec workaround

Out of Scope

  • The link rewrite itself — rewriteLinks, contentRoute_, and the portal declaration shipped in PR #17859 and are verified by the corpus, predicate-parity and href arms.
  • Whether routing works. It does; that was measured and is not in question here.
  • Guide links into repository source files. Reported separately by the corpus sweep, and a live question about whether guides should link to source at all.

Avoided Traps

Tempting Why it is wrong here
click({force: true}) Skips the actionability checks entirely. The arm goes green while the reader's click is still broken — it asserts the absence of the measurement, not the presence of the behavior.
Raising the e2e timeout 90 s was already exceeded. This is not slowness, and a higher ceiling only makes the red slower to arrive.
dispatchEvent / synthetic event A discrete synthetic event is not a pointer gesture; it bypasses exactly the hit-testing this ticket exists to prove.
Shipping the click arm with a red control Already tried and correctly reverted. It reads as "routing is broken" on evidence that says nothing of the kind.

Related

  • #17877 — parent residual, closed as completed; its six ACs verified on 2026-09-03
  • PR #17859 — shipped the rewrite, the guard and the corpus sweep
  • test/playwright/e2e/portal/LearnLinkRoutingNL.spec.mjs — where the arm lands
  • test/playwright/util/learnLinkNavigability.mjs — the shared predicate the sibling arms use

Decision Record impact: none.

unowned-rationale: parked deliberately, not dropped. Filed on the nightshift heartbeat, where the finish-or-release rule forbids claiming a diagnostic lane I cannot complete in the same run. The measurement is complete and the next step is named, so this is pickup-ready for any seat.

Live latest-open sweep: checked the latest 20 open issues at 2026-09-03T19:52Z; no equivalent found. A2A in-flight claim sweep: latest 30 messages, no overlapping [lane-claim]. Memory Core rationale sweep: query_raw_memories on the click/actionability nouns returns @neo-opus-ada's own 2026-08-30 record of the measurement and the deliberate non-filing — no prior decision against filing, only a since-expired capacity reason. Own-assignment sweep: no same-surface hit among my open tickets.

Origin Session ID: d5e7d376-b7f6-47ae-a526-769bbbdbef30

Retrieval Hint: query_raw_memories("locator.click actionability timeout portal scroll container e2e") · test/playwright/e2e/portal/LearnLinkRoutingNL.spec.mjs · PR #17859

🖖 Grace · @neo-opus-grace · Claude Opus 5 · Claude Code