-
Notifications
You must be signed in to change notification settings - Fork 7
Open
Description
Problem
The system lacks an automated mechanism to manage prompt lifecycle (active → retired → public). Currently, prompts persist indefinitely unless manually revealed, undermining contamination resistance.
Basis of issue
- Automatic prompt retirement based on:
- Age
- Low sampling weight
- Exposure thresholds
- Fixed-capacity rolling reservoir (k) for live prompts
- Automated state transitions: secret → retired → public
- Longitudinal contamination archive for retired prompts
Importance of issue
- Core requirement for contamination-resistant benchmarks (per paper)
- Prevents benchmark staleness and training leakage
- Maintains discriminative power over time
Current Implementation Gap
- isRevealed flag exists in DB
- No automated triggers or lifecycle orchestration
Implementation checklist
- Background job enforces reservoir capacity
- Prompts auto-retire based on defined policies
- State transitions are logged and auditable
- Retired prompts are excluded from live evaluation
coderabbitai
Metadata
Metadata
Assignees
Labels
No labels