Historical record, labelled October 19, 2026. This check ran on October 1, 2026, before Practical Picks adopted its research-only method (October 3, 2026). It is kept as dated, bounded evidence about the versions named below. We no longer run product checks, and this record did not decide the current Pick. The October 19, 2026 article no longer compares TestivAI Witness. PP-242: local visual-regression fixture, October 1, 2026 The test page was a disposable, self-authored HTML/CSS page with two plan cards. It was served only from 127.0.0.1:4422. We captured 980 x 700 desktop and 390 x 844 mobile views in Chrome for Testing 151.0.7922.34 on Ubuntu Linux x86_64 (kernel 6.8.0-139-generic). Node was 22.23.3. Packages were Playwright Test 1.62.1, BackstopJS 6.3.25, and TestivAI Witness/Witness Playwright 2.0.1. The fixture source and runnable configuration are kept in the private PP-242 research record. For each tool we froze a same-page baseline, repeated it unchanged, then inserted CSS rules that: added 7 px article top padding; changed the heading from 34 to 36 px; hid a badge; and added 8 px article left padding on mobile. The CSS changes left the page DOM unchanged. All three tools passed unchanged desktop/mobile repeats and flagged each deliberate change. Playwright and BackstopJS exited 1 on mismatches. The TestivAI Playwright reporter identified changed snapshots but exited 0. We then captured both desktop and mobile in one run before each gate check. For unchanged captures, `testivai report --fail-on-diff` found two passes, zero missing baselines and exited 0. For the 7 px padding change, it found two changes, zero missing baselines and exited 1. A separate mobile-only unchanged run found one missing desktop baseline; the same gate exited 3 by default. A CI job must capture every expected viewport before running the gate, or treat missing-baseline coverage explicitly. These were local gate commands, not a hosted CI integration. For the desktop padding change, TestivAI reported DOM unchanged, computed styles changed on two article elements, and one shifted region at +7 px. One synthetic local page cannot establish comparative reliability, false positive rates, speed, browser breadth, or performance on dynamic production sites. Mobile was browser emulation, not a physical phone. We did not test hosted CI integrations, masks, animations, Docker, hosted services, or cross-OS baselines. We accepted only this fixture's baselines. A human must inspect real changed screenshots before accepting a site's new baseline.