After removing window two's fourth lane, ac_layout_read still reported B9 in dispatch.focusedSessionId, although B9 was no longer selected in any lane. Restoring lane two to B7 still left B9 in that field.
Evidence: revisions a6ffd4a39cd846be and b1a877467d4d1da5; tiled focusedLane was 1. Window one likewise retained C18 in the field while its selected lanes were C17/B7, then only C17 after removal.
Expected/acceptance: expose the effective focused row/lane/session consistently with workspace observation after selection, removal and focus changes. If the field intentionally remembers classic Dispatch selection, name/document it accordingly and expose effective tiled focus separately.
Impact: operators can mistake a legacy selection for the input target. This does not establish actual input misrouting. Source lead: workspace/control/layout.ts spreads stored dispatchMode alongside normalized tiled state. Add a behavioral regression check for these transitions.
Environment: macOS, development Electron, two Agent Code windows, external MCP plus computer use, observed 2026-09-05. Exact running build revision was not verified. Refs #793, #795.
After removing window two's fourth lane, ac_layout_read still reported B9 in dispatch.focusedSessionId, although B9 was no longer selected in any lane. Restoring lane two to B7 still left B9 in that field.
Evidence: revisions a6ffd4a39cd846be and b1a877467d4d1da5; tiled focusedLane was 1. Window one likewise retained C18 in the field while its selected lanes were C17/B7, then only C17 after removal.
Expected/acceptance: expose the effective focused row/lane/session consistently with workspace observation after selection, removal and focus changes. If the field intentionally remembers classic Dispatch selection, name/document it accordingly and expose effective tiled focus separately.
Impact: operators can mistake a legacy selection for the input target. This does not establish actual input misrouting. Source lead: workspace/control/layout.ts spreads stored dispatchMode alongside normalized tiled state. Add a behavioral regression check for these transitions.
Environment: macOS, development Electron, two Agent Code windows, external MCP plus computer use, observed 2026-09-05. Exact running build revision was not verified. Refs #793, #795.