This issue is staged for the repository's GitHub tracker cutover. Existing task instructions remain authoritative until the verified handoff.
What
Observe a real application restart against the shipped project-scoped roster and conductor persistence, using a disposable test project rather than any in-use workspace, since the merged implementation explicitly states this was not done.
Why
The change that shipped this (keying the pane roster and conductor identity to the selected project, so switching projects cannot read back another project's panes) passed a large automated suite, but its own pull request lists explicitly what it does not establish: "The app was not launched... No live-app render test... Freeze and binding are proven at the decision and storage layers, not through App." The task's own acceptance criteria describe a successful restart reopening the same pane ids, brains, isolation flags, saved worktrees and conductor as fresh processes, which only a real restart can confirm.
Scope
In scope: launching the real application against a disposable test project, opening panes, restarting, and confirming the same panes, ids and conductor identity are restored; then switching to a second disposable test project and confirming no cross-contamination. Out of scope: any code change; this is a verification pass over already-shipped code. Out of scope: performing this against any live, in-use project's saved roster.
Done when
A real restart is observed to reopen the same pane ids, brains, isolation flags, saved worktrees and conductor as fresh processes for an isolated test project, and a project switch is observed not to restore or overwrite another project's panes; the result is recorded against this task.
Validation
Direct manual observation of a real restart and a real project switch against disposable test projects; this is exactly the evidence category the shipping PR's own text says is still missing.
Dependencies and decisions
None; the shipped code is not in question, only whether its restart behavior has actually been observed running.
Links
Pull request #48, which shipped the underlying change and discloses this limitation directly.
This issue is staged for the repository's GitHub tracker cutover. Existing task instructions remain authoritative until the verified handoff.
What
Observe a real application restart against the shipped project-scoped roster and conductor persistence, using a disposable test project rather than any in-use workspace, since the merged implementation explicitly states this was not done.
Why
The change that shipped this (keying the pane roster and conductor identity to the selected project, so switching projects cannot read back another project's panes) passed a large automated suite, but its own pull request lists explicitly what it does not establish: "The app was not launched... No live-app render test... Freeze and binding are proven at the decision and storage layers, not through App." The task's own acceptance criteria describe a successful restart reopening the same pane ids, brains, isolation flags, saved worktrees and conductor as fresh processes, which only a real restart can confirm.
Scope
In scope: launching the real application against a disposable test project, opening panes, restarting, and confirming the same panes, ids and conductor identity are restored; then switching to a second disposable test project and confirming no cross-contamination. Out of scope: any code change; this is a verification pass over already-shipped code. Out of scope: performing this against any live, in-use project's saved roster.
Done when
A real restart is observed to reopen the same pane ids, brains, isolation flags, saved worktrees and conductor as fresh processes for an isolated test project, and a project switch is observed not to restore or overwrite another project's panes; the result is recorded against this task.
Validation
Direct manual observation of a real restart and a real project switch against disposable test projects; this is exactly the evidence category the shipping PR's own text says is still missing.
Dependencies and decisions
None; the shipped code is not in question, only whether its restart behavior has actually been observed running.
Links
Pull request #48, which shipped the underlying change and discloses this limitation directly.