CSS testharness corpus chartered and closed in three days; rendering failures down to 757
Six goals closed with final numbers, including css-testharness-compat, chartered on 10-27 and closed on 10-30. As the large encoding and wasm corpus runs were recorded, the full-WPT pass rate rose from 7.41% to 42.11%. Rendering failures fell from 884 to 757.
Overview
The week brought 703 commits and six goals closed with final numbers. The biggest change came from css-testharness-compat: a coverage audit on 10-27 found that the 195,173 testharness subtests under css/* had never run at all (only reftest manifests had been imported).
The goal was chartered the same day, walked through M0 (530 cases, 9,843 subtests, 27.8%) to M4 closure on 10-30, and archived: its 65,501-subtest corpus face stands at 17,781 passing (27.1%), the css coverage snapshot face rose from 0.12% to 9.08%, and the CI coverage loop went green. wasm-compat archived with core 188,280/188,289, jsapi 878/878 and webapi 272/272; memory64 was re-attributed as Chrome-parity. encoding-legacy-mb archived at 360,431/360,431; form-url-encoding-compat was chartered and closed the same day (766,958/766,959); encoding-compat-v2 closed at 1,139,476/1,152,394 (DC-4 criterion 98.879%); net-api-compat closed its phase-2 re-entry at 11,040/12,077.
New and re-opened: editing-compat was chartered and laid down its M1 baseline (128 cases, 106,736 subtests, 2.18%), with a follow-up re-measure cutting run-face failures by 48.5%; dom-expansion was chartered; webcrypto-compat re-entered, archived its M1/M2 milestones, cleared the 9 fetch/Request failures, and re-verified its 33,109-subtest anchor byte-for-byte at the latest tip; selection-compat was re-chartered (M1 not started).
Rendering failures fell from 884 to 757 (R5373 accounting, denominator 16,590): after the two root causes behind the inline-host 710-failure cluster were fixed, that face went from 170 passing to 720 passing/160 failing; svg width-only height fallback, nested float clear-edge prediction, anonymous cells for out-of-flow children inside tables, and parent-height backfill with collapse barriers closed over consecutive rounds, turning background-root-017/018 from ✗3.75%/2.53% to ✓0.00%. html-semantics-compat climbed to 68.32% (M2 slice 183, closing the keyboard click-only activation and SVG shape hit-testing slices); html-syntax-compat re-closed at 61,341/61,411 after a corpus refresh (failures 105→70).
The website fixed its bilingual typography and mobile overflow, and the homepage gained a full-WPT accounting stat — the denominator is the union of the three browsers' wpt.fyi aligned stable runs with uncovered paths counted as zero; the first snapshot on 10-27 read 7.41%, and after the large corpora were recorded it reached 42.11% on 10-31.
Completed
css-testharness-compat went from charter to archive in three days (chartered 10-27, closed 10-30): the 195,173 testharness subtests under css/* started from zero coverage; the corpus face closed at 17,781/65,501 (27.1%), the coverage snapshot face rose 0.12%→9.08%, and the CI coverage loop went green
wasm-compat archived: core 188,280/188,289, jsapi 878/878, webapi 272/272; memory64 re-attributed as Chrome-parity
encoding-legacy-mb archived at 360,431/360,431; form-url-encoding-compat chartered and closed the same day (766,958/766,959); encoding-compat-v2 closed at 1,139,476/1,152,394 (DC-4 98.879%); net-api-compat closed phase 2 at 11,040/12,077
editing-compat chartered with its M1 baseline (128 cases, 106,736 subtests, 2.18%), run-face failures down 48.5%; dom-expansion chartered; webcrypto-compat re-entered (M1/M2 milestones archived, 9 fetch/Request failures cleared); selection-compat re-chartered
Rendering failures 884→757 (R5373 accounting, denominator 16,590): the inline-host cluster went 170P→720P/160F; svg width-only height, nested float clear, table OOF anonymous cells and parent-height backfill with collapse barriers closed over consecutive rounds, turning background-root-017/018 ✗→✓0.00%
html-semantics-compat climbed to 68.32% (slice 183); html-syntax re-closed at 61,341/61,411 after the corpus refresh (failures 105→70)
Website bilingual typography and mobile overflow fixes; homepage gained the full-WPT accounting stat (wpt.fyi three-browser union denominator: 7.41% in the 10-27 snapshot, 42.11% on 10-31)
In progress
rendering-compat has 757 cases left against the 100% target; the R5372 residuals (rtl containing block ignore-left, §10.3.7 equation ordering, static-x +29) and deep-domain parked items hold. editing-compat keeps closing rounds past its M1 baseline (style-with-value cluster and more); webcrypto-compat is progressing through M3; selection-compat's M1 has not started.
Sources
Engineering guides
How ZeroWeb tests AI-autonomous software engineering
A browser gives AI development a long-running test case: architecture, code, debugging and verification must work together.
ZeroWeb Rust browser architecture: from request to pixels
Follow the page pipeline and the boundaries between the browser application, embeddable WebView, engine and rendering infrastructure.
Getting started with ZeroWebView and ZeroBrowser
Prepare a local checkout, launch the desktop browser and inspect the WebView API before embedding the experimental engine.
How to read ZeroWeb WPT and performance results
Understand self-source reftests, assertion-based suites, changing corpus sizes and p95 benchmarks before comparing results.