# Retest result — local training fixture

The training fixture reproduced the expected clear-search failure at both 1440 × 900 and 390 × 844: after a no-result query, Clear search left the input empty but restored 0 of 15 cards. This is a deliberately introduced training defect, not a finding in the production site.

The minimal correction adds update() immediately after input.value = ''. The original before.html and its SHA-256 remain intact; after.html and minimal-fix.diff record the correction separately.

The same browser script then restored all 15 cards after clearing, at both sizes. Searching meeting returned 1 card; a no-result query returned 0; coding returned 3. Keyboard Tab reached the copy button and Enter activated it. Clipboard readText matched the full frozen template and displayed text, including 12 newline characters. Focus was visible and neither viewport overflowed horizontally.

This run used desktop Chromium with two viewport sizes. The 390 × 844 run is a narrow viewport test, not a claim about physical iOS Safari or Android devices. The fixture is local training material; it makes no production performance or accessibility certification claim.