LTL-UI latest30 navigation error review

This static review site contains the 2,358 individual candidates produced by the seven properties added between the frozen latest23 and latest30 Selected-31 checker. The source experiment replayed 31 websites across three previously recorded traces with action policy tier1-wait-tab4-shift1/1.

The site preserves each canonical X#### ID and separates two kinds of annotation:

• Agent first-pass suggestion: evidence-based guidance generated on 2026-09-04. It is not human-confirmed ground truth.
• Manual review: the decision and note entered in the browser. These values remain in localStorage until exported.

Each replay panel uses one representative trigger for that case and replay. “Before” is the trace entry immediately preceding the matched trigger screenshot; “After / trigger” is the screenshot recorded on the matching violation entry. A red rectangle is shown when the trace includes a focus box. Checker snapshots, historical Orca evidence, fresh-route DOM/source leads, and known evidence gaps appear below the screenshots.

Manual decisions are correct, false-positive, duplicate, and unsure. Exported files use schema ltl-ui-latest30-navigation-error-review/1. Import accepts that same schema and ignores IDs outside this fixed 2,358-case dataset.

dataset-summary.json records candidate, property, evidence, image, trace-match, and missing-image counts from the build. cases.js is the generated browser payload. The images directory stores each source screenshot once and allows multiple cases to reference the same asset.