QA run · items unblocked by the R-1/R-2 fixes · 4/4 now pass · 19f98fa1 · 2026-08-20

[os-elon]

Follow-on to #10454. Once the two metadata read-door defects were confirmed fixed on 17.1.0, the clauses they had been blocking became testable for the first time. This round drives exactly those. All four items now pass; no fail.

Companions: #10224 · #10225 · #10233 · #10245 · #10248 · #10257 · #10342 · #10454. Anchor: #10236.

Trigger

Re-test trigger #2 — a fix landed. #10454 confirmed R-1 (runtime-authored dashboard served by no read door) and R-2 (same on view) are fixed. Those two defects carried four downstream verdicts in the 2026-08-18 wave that were never about the items themselves:

• dashboards.chart-type-matrix c0/c4/c7 — blocked(product-bug → R-1)
• dashboards.strict-widget-rejects-stray-keys c5 — fail, via R-1
• dashboards.global-filters-rescope c6 — fail, via R-1's overlay variant
• studio-authoring.view-authoring-live c2/c4 — blocked(dependency) on its own failed c3, which was R-2

Environment

framework 19f98fa1 (17.1.0) · objectui pin + console stamp 9a3daf8d37ad, check:console-sha ✓ · showcase on 3456, fresh file DB · 2026-08-20 21:16–21:45 UTC. Browser via executablePath=/opt/pw-browsers/chromium. Execution mode: SEQUENTIAL, declared.

Results

item	rev	prior (e4e5c6e3)	now
dashboards.chart-type-matrix	3	PARTIAL — 16/20 variants	pass — 20/20
dashboards.strict-widget-rejects-stray-keys	2	FAIL (c5)	pass
dashboards.global-filters-rescope	2	FAIL (c6)	pass
studio-authoring.view-authoring-live	1	FAIL (c3)	pass

chart-type-matrix — the matrix closes at 20/20

The four variants that were unreachable (kpi / gauge / solid-gauge / bullet) need a scratch dashboard, which R-1 made unrenderable. Re-authored as qa_single_value_gallery:

PUT /api/v1/meta/dashboard/qa_single_value_gallery → 200 {success:true, seq:1, state:"active"}
GET by-name → 200, item present, widget types = [kpi, gauge, solid-gauge, bullet]
GET /api/v1/meta/dashboard → 200, list includes it
browser /_console/apps/com.example.showcase/dashboard/qa_single_value_gallery
 → no "Dashboard Not Found", 0 pageerror, all four widget titles present

Screenshot oracle consulted, not just the DOM: all four tiles render a value of 10, and showcase_task holds exactly 10 rows — the displayed value is server truth, not a placeholder.

Worth recording so the next runner does not read it as a defect: the four tiles are visually identical (plain numeric cards, no gauge arc or bullet track). The clause calls them "four single-value synonyms", so one shared renderer is what the wording describes.

c7 (compareTo): first attempt answered 422 naming the schema — compareTo.kind accepts only previousPeriod | previousYear, and I had sent {period:'previous'}. My payload, not a defect. With the prescribed shape: 200, read back as {"kind":"previousPeriod"}, list includes it, and the dashboard renders with 0 pageerror.

strict-widget-rejects-stray-keys — c5

Prior: the corrected widget published {success:true, seq:3} and then rendered "Dashboard Not Found". Now:

PUT qa_stray_key_probe (dataset + dimensions:[status] + values:[task_count]) → 200 seq:3
GET by-name → 200, item present, widgets = 1
browser → no "Dashboard Not Found", 0 pageerror, title "Tasks by Status" present,
 5 .recharts-rectangle marks

5 marks = 5 distinct statuses, matching the wave's own ground truth — so the widget did not merely mount, it bound to the data.

Non-regression control on the same item: re-sent a widget carrying the pre-ADR-0021 inline analytics keys (object + categoryField + valueField + aggregate) → 422 invalid_metadata, still refused. The read-door fix did not loosen the authoring gate that c0–c4 pin.

global-filters-rescope — c6

The overlay half was verified in #10454: re-PUT of the shipped showcase_revenue_pulse with kpi_new_accounts.filterBindings.dateRange → created_at and kpi_invoices.filterBindings → {region:false} now reads back as authored (prior: 200 receipt, re-read still returned the code values). c0–c5 and c7 passed in the prior wave and were not re-driven — no trigger.

Scope honesty: the clause says "Studio inspector authors filterBindings". This round proved the persistence and read-back through the metadata door, not the inspector UI path. The mechanism the clause depends on is fixed; the console-driven half is not re-verified here.

view-authoring-live — c2 and c4 unblocked

Full draft → publish → edit → re-publish cycle, all through the consumer door:

PUT /meta/view/showcase_task.qa_live_<id>?mode=draft → 200 {state:"draft"}
GET by-name while drafted → 404 (c1: no draft leakage, still holds)
publish (PUT without ?mode=draft) → 200
GET by-name → 200, columns = [title, status] (c2)
edit + re-publish (label V2, +priority column) → 200
GET by-name → 200, label "QA Live V2",
 columns = [title, status, priority] (c4)

One item deliberately NOT re-run

platform-core.builtin-apps-nav-render (P0) was this round's originally planned target. After reading #9453 I dropped it, and the reason is worth recording rather than leaving as a silent gap:

That wave walked Setup 43/43 and Account 7/7 destinations — 50/50 ok, screenshot-first then DOM, zero pageerror. Its c0/c1/c2 are blocked(fixture) for one reason only: Studio is not installed on the showcase. rev 2 still names three built-in apps, so the ceiling on stock fixtures is unchanged, and c3–c6 already passed there (I independently reproduced c3 in #10257). Re-running it would consume a browser round to reproduce the same partial. It needs a fixture decision — install Studio, or reword the item for a two-app showcase — not another run. That decision is already on #10236.

Gaps for the anchor

Parked-on: #10236 · global-filters-rescope c6 — mechanism verified via the metadata door;
 the Studio-inspector UI path remains unverified
Parked-on: #10236 · chart-type-matrix — record in the item that the four single-value
 synonyms share one renderer, so identical tiles are expected