Annotation Queue FE — Checkpoint 5: Combined Human + Auto Evaluation

[mmabrouk]

Scope

Third checkpoint: annotation queues with automatic evaluation integration.

• Blocked by backend loops (evaluation engine cleanup by JP)
• When loops ready: human evaluations run in backend (not frontend)
• Eventually "human evaluation" vs "automatic evaluation" distinction goes away

What's included

• Combined human + auto evaluator in same evaluation run
• Queue dispatch for mixed evaluation
• Comparison view: human vs auto eval results

Dependencies

• Checkpoint 1 and 2 must ship first
• Backend loops cleanup (JP) must be done
• Design must cover this flow

[linear]

AGE-3710