mirror of
https://github.com/stablyai/orca.git
synced 2026-09-28 16:02:45 +00:00
Closes the limit the migration-pair suite stated: that one measures the re-key function, not the write path. This drives both builds' real Store over a real file and answers the three rollback questions separately - old build reads new state, old build writes back and the new build still accepts it, and neither deletes a top-level key it does not understand. Both same-version round trips run as controls. They earned their keep: a first draft asserted on three hand-picked row maps and BOTH controls failed, because the store round-trips neither sleepingAgentSessionsByPaneKey nor terminalSurfaceTombstonesByPaneKey in either build. Comparing whole sessions read from one file measures the question instead of the author's guess about the schema. tabsByWorktree[id] === [] is pruned to absence by both builds alike, so rollback does not cause it. Pinned literally so a build that starts preserving that intent turns the cell red on purpose.