look for p_ convo output

[jgieringer]

Description

With the new output defaults, conversations are now stored under p_.../conversations/.txt instead of conversations/p_.../.txt

When passing p_.../conversations as the input to judge, it creates a judge folder called j_...conversations instead of j_...p_... which caused "Unknown" User name and Agent name 💀

This PR addresses looking one level up for the expected p_... dir name to preserve user and agent information

[Copilot]

Pull request overview

Updates transcript folder resolution to support the new default layout where transcripts live under p_.../conversations/*.txt, ensuring judge run folder naming (and derived metadata like user/agent names) stays aligned with the generation run id when users pass p_.../conversations as input.

Changes:

• Add special-case handling in resolve_conversation_input() to “lift” gen_run_root/basename from the parent when the input folder is a nested conversations/ dir under a generation run folder.
• Add unit tests covering the nested conversations/-as-input behavior and the non-generation-parent fallback behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
utils/conversation_layout.py	Adds logic to detect .../conversations inputs and attribute them to the parent p_... run folder for consistent judge output naming/metadata.
tests/unit/utils/test_conversation_layout.py	Adds unit tests validating the new nested conversations/ input resolution behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[emily-vanark]

Hmm... running uv run pytest -m "not live" on this branch still results in the mysterious adhoc outputs... (I guess I was wrong about it being the live tests? Updated the issue.)

Stll getting 69 judge_unscoped files similar to the ones in the issue and an empty single_datetime_conv folder.