Interviewer Prompt
Version Management
How interviewer prompt versions are tracked and stamped.
How Versioning Works
Every time you save the interviewer rubric:
- A modal asks for change type, name, and description
- A new version is created with a semver number (e.g., 1.2.0)
- The full rubric config is snapshot — immutable, never modified
Version Stamping
When an interview starts:
- The system fetches the current interviewer version
- This version is stamped on the transcript row as
interviewer_version - When the interview ends and scoring happens, the
interviewer_versionis carried to the scorecard
This means you can always trace back: "This candidate was interviewed with version 2.1.0 of the interviewer rubric."
Why This Matters
- A/B comparison: Compare scorer outcomes between different interviewer versions
- Regression detection: If scores drop after a rubric change, you can identify the version
- Audit trail: Legal defensibility — you can show exactly what prompt was used for each interview
- Rollback safety: If a change causes problems, rollback creates a new version (forward-only, no history rewriting)
Trends
The interviewer prompt page shows version trends:
- Average duration per question (normalized by depth)
- Completion rate
- Downstream scorer outcomes (average score, recommendation breakdown)
- Delta indicators: green ↑ = improvement, red ↓ = regression