TUI prepends paginated history before fork lineage on resumed forked threads

What version of Codex CLI is running?

• rust-v0.148.0-alpha.1

What subscription do you have?

• ChatGPT Pro 20x

Which model were you using?

• gpt-5.6-sol

What terminal emulator and version are you using (if applicable)?

• Alacritty with Codex running inside tmux

What issue are you seeing?

• Resuming a forked thread initially renders Thread forked from ... before the child transcript, but loading an older paginated history page inserts those messages before the lineage marker.
• The marker then appears inside the child transcript instead of remaining its ancestry boundary, even though the persisted parent metadata is intact.
• This makes older child messages look as if they precede the fork relationship and obscures where the resumed thread came from.

What steps can reproduce the bug?

• Create a fork from a thread with enough inherited or child history to require more than one history page.
• Resume the forked thread in the TUI and confirm that Thread forked from ... appears before the initially loaded messages.
• Scroll to the beginning or otherwise trigger loading the preceding history page.
• Observe that the newly loaded messages are inserted before Thread forked from ....

What is the expected behavior?

• The fork lineage marker should remain before every transcript item belonging to the child thread as older pages are prepended.
• Initial resume and later pagination should preserve the order fork lineage, older child history, newer child history.

Additional information

• Fork lineage is currently emitted before the initial history replay buffer begins and is represented as an ordinary transcript cell.
• The non-overlay pagination path chooses an insertion point only after the last session-info cell, defaulting to index zero when that cell is absent.
• The lineage cell is therefore not recognized as a structural insertion boundary.
• This issue report was written with assistance from Codex CLI.