Interpret role-specific AI reviews and configure their evidence window.
Reviewed · 2026-10-12
On this page · 3
1. Read the assessment and evidence
- Read the AI assessment card together with its completion time and any historical-snapshot notice.
- The subject and platform administrators can open the same completed assessment shown on the profile, including observations, next steps and source links. A newer collected version does not hide an assessment when its cited version remains readable.
- Criteria use 0–4 points converted to a 100-point scale. Insufficient evidence is unscored, not zero. A weighted overall score needs at least two scored criteria.
- This is a mentoring draft. Compare the documented work and missing evidence, not just scores across people with different data coverage.
- Expand the card's details for refresh frequency, next attempt and guidance. Scheduled reviews run in turn without repeated profile refreshes.
- Leadership assessment applies only to platform administrators, active lab PIs/managers and assigned grant PIs (grant owners). Responsible researchers, project owners and delegated managers use contributor criteria unless they independently hold one of those PI roles. Their project creation and management permissions remain unchanged. Its four criteria cover research direction and goals, validation oversight, delegation and mentoring, and risk/resource/follow-up decisions. Contributor assessment covers progress, reproducibility, collaboration and next steps.
- The subject's explicit assignments determine the perspective, not the requester’s role. Leaders need not personally execute every experiment or coding task; team results are not automatically personal achievements. Role and policy changes trigger a scheduled reassessment. A previous score with a different perspective is hidden until reassessment finishes. An assessment labeled as predating role-specific criteria retains its original interpretation.
2. Administrator: set the period and frequency
- Open Lab management → AI assessment policy, or use Assessment settings from a detailed review. Only platform administrators can save the policy.
- Enable automatic profile review. Choose a refresh interval of 12, 24 or 48 hours, or seven days.
- Enter an integer from 1 to 365 under Evidence window (days); the default is 30. Seven includes today and the preceding six days. A 24-hour refresh and a 90-day window reviews the latest 90 days each day.
- Background goals and plans may fall outside the window. A longer window does not acquire unconnected or uncollected data, and input limits remain.
- Add guidance and four weights. Each must be 1–100 and the total must be 100. Save, then verify the period and interval.
- Changes apply to subsequent reviews. The previous completed review retains its original criteria while the replacement is queued. AI permission and the administrator's daily allowance still apply.
3. When a review is missing or old
- Check linked GitHub/Notion identities, participation and researcher metadata on collected sources. Insufficient linked evidence may leave a review pending or partially unscored.
- Ask an administrator to confirm lab AI and automatic review settings. Queued and running jobs are processed in turn.
- Temporary failures retry up to twice after an hour; daily quota limits resume after the Seoul date changes. Include the displayed status when requesting support.
- A readable completed snapshot may remain visible during refresh. Check its date and source set instead of treating it as the new period's result.