# Parent comparison: image 029

Parent: /root, local task 01a0e067-c27c-77f2-a3f9-ec52a44b62cf. Fresh describer: /root/blind_029, gpt-6-luna high, forked with no conversation context. Its only evidence input was this image. Parent read the completed description and verified the image checksum.

[Actual screenshot](../inputs/029.png) · [Independent description](../descriptions/029.md)

Variant A. Captured action: Dirty proposal restored when switching away and returning. URL: https://calendar-stage-20260923.cfd-staging.com/panel/proposals/?proposal=p_9b7954858811812947a65a4b721cbdf5505942bbd71ee49751577d0db7b3a2f7. Screenshot SHA256: `9eec66847998df0af30a8e01c37d65c5ff36933d6784cee6671a84b90eb20e98`.

- C23: When a save fails or another session changes the same proposal, then the user sees a clear unsaved/conflict state with a recoverable path and their typed values are not silently lost.

**Parent comparison:** The description reads the restored-but-unsaved draft notice after returning to A's proposal, with local values and customer panel retained. The actual dirty-switch journey verifies the typed field values survived changing contacts and returning. It is deliberately not labelled saved. Later explicit save/readback, error recovery and conflict merge are separate steps in the same completed scenario.

[Completed GUI journeys and exact behavioral evidence](../../deployed-gui/scenario-matrix.md). Full criterion acceptance also requires the other applicable states and nonvisual checks. A screenshot alone does not prove persistence, timing, arithmetic or downloaded bytes.
