Problem
When viewing samples from non-chat evaluations (e.g., embedding similarity), the Transcript and Messages tabs are empty. This is confusing — users click expecting content and find nothing.
Context
Inspect was designed for LLM chat evaluations where each sample has a conversation transcript. For embedding-based evaluations (like the Grounding Measure G1/G2/G3/G4 metrics), samples are word pairs with similarity scores — there are no messages or tool calls.
Proposed fix
In the sample detail view, conditionally hide or disable tabs that have no data:
- If
sample.messages is empty, hide/grey out the Messages tab
- If
sample.transcript (or equivalent events list) is empty, hide/grey out the Transcript tab
This would benefit any non-chat evaluation displayed in the viewer.
Upstream potential
This is a reasonable improvement for the upstream Inspect AI project — non-chat evaluations are a valid use case.
🤖 Generated with Claude Code
Problem
When viewing samples from non-chat evaluations (e.g., embedding similarity), the Transcript and Messages tabs are empty. This is confusing — users click expecting content and find nothing.
Context
Inspect was designed for LLM chat evaluations where each sample has a conversation transcript. For embedding-based evaluations (like the Grounding Measure G1/G2/G3/G4 metrics), samples are word pairs with similarity scores — there are no messages or tool calls.
Proposed fix
In the sample detail view, conditionally hide or disable tabs that have no data:
sample.messagesis empty, hide/grey out the Messages tabsample.transcript(or equivalent events list) is empty, hide/grey out the Transcript tabThis would benefit any non-chat evaluation displayed in the viewer.
Upstream potential
This is a reasonable improvement for the upstream Inspect AI project — non-chat evaluations are a valid use case.
🤖 Generated with Claude Code