Research: Run the Brave/macOS gesture test matrix

[Millbot-AI]

Goal

Collect real two-finger trackpad measurements in Brave on macOS using the Phase 1 proof of concept.

Acceptance criteria

• Test horizontal swipes in both directions on a page without horizontal scrolling.
• Confirm that vertical scrolling does not produce a back or forward candidate.
• Test a horizontally scrollable table, carousel, and code area.
• Record raw and normalized deltas, sequence duration, velocity, cancellation, and momentum behavior.
• Repeat representative cases with natural scrolling enabled and disabled.
• Add anonymized findings to docs/gesture-research.md.
• Do not implement tab-closing or history behavior.

[Millbot-AI]

First real Brave/macOS measurement recorded in commit ed6f5de.

Four physical back swipes reached the content script as complete pixel-mode wheel sequences (84-142 events, 682-1370 ms, net X from -2326 to -2910 px). On this Mac/configuration, the physical back direction mapped to negative deltaX. Each sequence crossed the research threshold once.

Critical finding: only the first event in each sequence was cancelable; all remaining 83-141 events were non-cancelable. This means threshold-time cancellation would be too late. The next controlled test must determine whether canceling the first horizontal event suppresses Brave native navigation when real internal back history exists.

The full measurements and the still-open test matrix are in docs/gesture-research.md. Phase 2 remains blocked.

[Millbot-AI]

Core Brave/macOS measurements are documented in commits 917a82a and 0b52e3e. Slow and fast vertical scrolling produced no candidate. The local horizontal fixture was recognized in both directions for every wheel event, and remained scrollable with root overscroll containment active (scrollLeft 717 to 0). The issue stays open for its extended acceptance criteria: representative real tables, carousels and code areas, natural scrolling enabled and disabled, and broader device/browser coverage.