test(runner): sweep the example1 grid before asserting its numeric cells

[danielzytohoc]

Context

The nightly starter compatibility matrix failed on 2026-09-09 (run 34299201725): matrix: example1 @ 17 [sandpack] and @ 18 red on all three attempts with numericFormat rendered 87% at 17/18, everything else green, and the Sep 8 run fully green.

This is a test defect, not a product one. The assertion added in c447344 reads the <td>s present in the DOM at the initial scroll position. #307 then removed width: 800 from example1 so the grid fills its container, which in the suite's Desktop Chrome viewport is a 624px preview pane holding four of the starter's 150px columns. Item Quality (the 87% cell) now sits past the rendered edge, and Handsontable's column virtualization decides whether its <td> exists at all:

• 17.1.0 and 18.1.0: viewportColumnRenderingOffset: "auto" is a fixed 1 since 17.1.0, so only Item Name → Category are in the DOM. 87% is absent until the grid is scrolled.
• 15.3.0 and 16.2.0: the older proportional auto still renders one more column, so 87 is found — as the last rendered column, by a hair.

Reproduced locally against prod with the Desktop Chrome device: at 18.1.0 the rendered set lacks 87% at scrollLeft: 0 and contains it after scrolling the master holder. The formatting itself is correct at every major.

Fix (spec only): sweep .ht_master .wtHolder across its full scrollWidth, accumulate the rendered cell text along the way, and assert over that union inside expect.poll, so a slower re-render costs a retry instead of a false failure. Putting a fixed width back in the starter is not an option: pipeline/starter-grid-width.test.mjs forbids it and it would bring the clipped-grid defect from #307 back. Nothing in examples/ or in Handsontable itself needs to change.

Types of changes

• New example
• Update to an existing example
• README / documentation change
• Demo runner (runner/) change
• CI / tooling change

How was this verified?

• The red evidence is the spec itself: without this change it fails at 17 and 18 in the linked run (3/3 attempts each) and in a local reproduction against prod.
• With the change, the same tests against prod (sandpack engine, no container sessions involved):

  cd runner && E2E_BASE_URL=https://demos.handsontable.com E2E_STARTER_MATRIX=1 \
    pnpm exec playwright test e2e/starter-matrix.spec.ts -g "example1 @ (16|17|18)" --workers=2 --retries=0
  

  → 3 passed (9.7s). The sweep adds under 300ms per test.
• No tsconfig in the repo covers e2e/, so an ad hoc tsc --noEmit with the tsconfig.base.json options was run on the spec: no errors inside the changed block. The file's pre-existing errors (missing @types/node, noUncheckedIndexedAccess on the version parsing) are untouched.
• PR CI skips this spec by design (E2E_STARTER_MATRIX gate); after merge, a workflow_dispatch of E2E — starter compatibility matrix confirms it ahead of the 01:00 UTC nightly.

Checklist

• New/renamed example: added to runner/config/frameworks.json — n/a, no example change
• New example: added a row to the tables in README.md — n/a
• Ran pnpm build (and pnpm dev) in the affected example/server-example locally — n/a, spec-only change

Related issue(s):

1. Failed run: https://github.com/handsontable/examples/actions/runs/34299201725
2. Follow-up to fix(examples): let the JavaScript and example1 starter grids fill their containers (DEV-2794) #307 (starter grid sizing) and the per-major numeric assertion introduced in c447344

---

Note

Low Risk
E2E test-only change in an opt-in matrix suite; no runtime or product behavior changes.

Overview
Fixes flaky example1 checks in the opt-in starter compatibility matrix (E2E_STARTER_MATRIX) that were failing on Handsontable 17 and 18 with missing 87% even though formatting is correct.

Instead of collecting cell text from whatever columns are in the DOM at scrollLeft: 0, the spec horizontally scrolls .ht_master .wtHolder across its full width, unions rendered values (with a short post-scroll wait for Walkontable), and uses expect.poll + toContain for each expected formatted value from example1NumericCells. That avoids false negatives when the preview pane is narrower than the grid and column virtualization omits off-screen columns.

Spec-only; no starter or Handsontable product changes.

^{Reviewed by Cursor Bugbot for commit f91d9cb. Bugbot is set up for automated code reviews on this repo. Configure here.}