WBS Section 4 of 7
LLM Rubric Scoring
AI Step 2 — Score structured resume data against the role rubric using Claude.
Handles failures gracefully. Supports re-scoring on rubric changes.
Start / End
Process
Decision
Error State
Success / End
Parsed candidate data (from §3)
Construct scoring prompt
Structured data + role rubric
→ Claude Skill (version-controlled)
Deal-breaker
triggered?
Yes
Auto-flag: Score = 1
Rationale explains which
deal-breaker triggered
No
Send to Claude API
Generate: score (1–10) +
2–3 sentence rationale
Valid response?
Int 1–10 + rationale
Fail
LLM Failure
Timeout / refusal /
malformed output
1st
attempt?
Yes
Auto-retry
w/ backoff
No
Flag: "Manual
Review Required"
Reason logged
→ Separate queue,
not in ranked shortlist
Pass
Validate output schema
Score: int 1–10, rationale: non-empty
Save score to record
Score + rationale + timestamp
More in
batch?
Next
No
→ Scored · Ready for Shortlist (§5)
↻ Re-score path
Leo can trigger re-score on rubric change.
Previous score preserved as history.