migrate-rollouts cannot recover an active rollout missing its state metadata

Summary

Codex detects active rollout files absent from the local state database, but codex migrate-rollouts --apply refuses to recover them because their SQLite metadata is missing.

Environment

• macOS on Apple Silicon
• Codex CLI 0.151.0
• Codex Desktop 26.825.41651
• ChatGPT authentication

Observed

codex doctor reports healthy state and thread-history databases plus two active rollout files missing state-DB rows. For either affected thread, the supported recovery command fails closed with:

thread-store internal error: rollout migration failed: thread <redacted> is missing its SQLite metadata

The raw rollout file remains present and unchanged; no direct SQLite repair was attempted. In contrast, the same native migration successfully rebuilt paginated history for 2,228 other active rollouts.

Expected

A supported recovery path should either reconstruct minimal metadata from a valid rollout or explicitly provide a safe, reversible metadata-rebuild operation. The current migration requires the missing record it is meant to repair.

Why this matters

This leaves local sessions present on disk but unable to enter the paginated-history path, contributing to stale/partial history display after resume or Desktop reconnect. This appears related to the broader projection-stall class discussed in #41079, but the immediate failure here is missing state metadata.

Start at the codex migrate-rollouts --apply entry point and trace the thread-store migration failure for rollouts missing SQLite metadata. Compare this path with the successful migration of other active rollouts, then define done as a supported recovery or explicitly safe, reversible metadata rebuild with regression coverage for the missing-row case.