Settings visual scores — G

All visual scores and defect severities are [INFERENCE], returned by the same verbatim P-settings prompt for every viewport PNG. DOM metrics and bench invariant lines are measurements, not visual-model conclusions.

Round summary

RoundMinimum axisP1P2P3All axes ≥9; zero P1/P2
0309626no
1308338no
2308233no
3308136no
4307735no
5407433no
6408637no

Round 0

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png54664564
settings-group-desktop.png44564554
settings-search-desktop.png54563543
settings-top-mobile.png64574554
settings-group-mobile.png54463543
settings-search-mobile.png43463433

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 1

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65674554
settings-group-desktop.png54564554
settings-search-desktop.png65674554
settings-top-mobile.png65564454
settings-group-mobile.png54465444
settings-search-mobile.png54564433

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 2

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65675665
settings-group-desktop.png54464554
settings-search-desktop.png54463543
settings-top-mobile.png64575454
settings-group-mobile.png54464454
settings-search-mobile.png54474333

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 3

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65574654
settings-group-desktop.png54665564
settings-search-desktop.png54464444
settings-top-mobile.png65574454
settings-group-mobile.png54464444
settings-search-mobile.png54464333

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 4

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65674664
settings-group-desktop.png54664564
settings-search-desktop.png54463433
settings-top-mobile.png65575454
settings-group-mobile.png54464544
settings-search-mobile.png54564433

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 5

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65674664
settings-group-desktop.png54564564
settings-search-desktop.png55574544
settings-top-mobile.png65575454
settings-group-mobile.png54564544
settings-search-mobile.png54564444

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Round 6

PNGHierarchyDensity / rhythmAlignment / spacingContrast / legibilityApple feelDiscoverabilityPolishMinimum
settings-top-desktop.png65574564
settings-group-desktop.png54675564
settings-search-desktop.png55664544
settings-top-mobile.png65665454
settings-group-mobile.png54564444
settings-search-mobile.png54564444

Measured bench check (pass):

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

Fix inputs sent

Every listed JSON preserves the complete P1/P2 element, problem, and fix strings verbatim with its source PNG, plus compatible-at-intent-level P3s. The editor owns contract compatibility and records any deferrals.

Final decision and focus amendment

Converged: false. Best evaluated round: 6. Final evaluated round: 6.

No round qualified. The six-round iteration cap was reached. Rounds 5 and 6 both have minimum axis score 4; the binding latest-round tie-breaker selects round 6 despite round 5 having fewer P2 findings. Severity counts above are per-PNG findings, not deduplicated defects. Vision findings and scores are [INFERENCE].

Main subsequently authorized the CSS-only round-7-focus amendment. It is not a seventh scored visual iteration, and it does not change the non-convergence decision. The six fresh final board captures are in round-final, with matching metric JSON. No critique was run for this amendment. Final source hashes are recorded in verdict.json; the staged install receipt is /tmp/qwfix/settings/round-7-focus/sha256.txt.

Final keyboard proof

The natural Tab walk reached a body cycle after 104 desktop stops and 51 mobile stops, with 36 Settings-owned stops at each viewport. Settings ownership means descendants of .qwizz-settings; core includes all other stops, including WordPress admin-bar/menu controls and the shared stats rail.

Scope / viewportStopsMissing indicators before → afterClipped before → after
Settings / desktop363 → 08* → 0
Settings / mobile363 → 03 → 2
Core / desktop680 → 011 → 11
Core / mobile151 → 17 → 7

*The raw desktop before count included one collector false positive: the opacity-zero Play sound input’s invisible external outline. Its visible proxy was the relevant indicator. Seven other recorded before failures concern visible Settings elements. Do not describe all eight as product defects.

The remaining measured Settings failures are the mobile Player and Advanced navigation anchors, whose focused boxes extend approximately 25.95px and 34.61px beyond the scroll strip’s right edge. A bounded replay found Player still clipped after 600ms, with scrollBehavior:auto and scrollLeft:0; this is not a smooth-scroll timing artifact.

Both viewports passed genuine Preview opening, focus inside the modal, Escape closing, exact opener focus restoration, and search input receiving pdf. The amended search assertion failed: the results panel reported 5 matches, but visible indexed setting rows outside the results panel increased 7 → 85, rather than decreasing. The denominator is the existing search controller’s full [data-search-target] row inventory, not only matched rows and not only rows inside the viewport. The previous walk’s distinct-native-field count, 22 → 106, remains preserved but is not the amended gate.

Both final walks had 0 console errors and 0 page errors. The overall keyboard proof remains FAIL; core failures are explicitly out of scope, but the two mobile Settings clips and the requested row-decrease behavior still fail.

Evidence:

  • keyboard-walk-final.json: final classifications, scope comparison, modal/search assertions, complete ordered stops.
  • keyboard-walk.json: pre-amendment walk with corrected stationary root-overflow geometry.
  • keyboard-walk-initial.json: preserved first walk; its mobile body-overflow clip calculation was wrong.
  • keyboard-walk-final-initial.json: preserved post-amendment output before disregarding the invisible input outline.
  • keyboard-final-artifact-check.json: live bounds and settled mobile clipping evidence.
  • Missing-indicator and mobile clip screenshots remain in this evidence directory. Only actual visible indicator paint is used for final ring clipping; inset shadows do not extend outside their carrier.

Final bench and corpus checks

Every round 0–6 and the final post-keyboard/post-capture check returned the identical lines:

opt_sha=89877cf601ca195bf0ff75c936637e485d8b6de454c301934bff54ab0838ff9f
total=94 owner1=10

No QuizWizz Settings save occurred. The disposable user is 1171 with the required fresh scheme; changing its scheme before baseline was user metadata only.

All round 0–6 capture console/page error counts were zero. The final six captures also reported zero errors; the selected group fit below the toolbar and inside the viewport at both sizes. Closed/expanded toolbar heights remained 52/96px.

The first 42 scored PNGs have correct viewport dimensions and 39 unique SHA-256 values: the desktop top, group, and search images are byte-identical between rounds 3 and 4, consistent with the mobile-only round-4 change. Scores assigned to those identical images differ, demonstrating vision-evaluation variance; do not attribute such changes to CSS. The final complete image/file inventory, dimensions, digests, and duplicate groups are in file-inventory.json.

The official 40-step collector’s unfiltered tiny-text count includes assistive/replaced-span mechanisms; its disabled Import contrast sample must not be presented as an active-control failure. Final active Choose file contrast is 4.26:1. The optional capture timing field is not used as page-performance evidence.

This is bench evidence, not release authorization: releaseEligible: false.