AI Evaluation Scorecard
| Title: AI Evaluation Scorecard |
Version: 1.0 |
Owner: [TENANT_CONFIGURATION_REQUIRED — AI Governance Lead] |
Status: Draft |
Last reviewed: 2026-09-07 |
Next review: [TENANT_CONFIGURATION_REQUIRED] |
Reviewers: AI Governance, QA |
Purpose and scope
Defines the scorecard template completed for every skill release, per ai-evaluation-strategy.md.
Scorecard template
| Field |
Description |
| Skill name |
e.g., Candidate Matching |
| Prompt version / Model version |
From prompt_version/model_version tables |
| Evaluation date |
— |
| Dataset(s) used |
Reference to evals/datasets/ |
| Functional accuracy score |
vs. release criteria in ai-evaluation-strategy.md |
| Explainability score |
Rationale specificity/consistency (see evaluation-prompts.md) |
| Fairness result |
Pass/fail vs. disparity threshold (fairness-and-bias-governance.md) |
| Security/red-team result |
Pass/fail per red-team-plan.md |
| RAG groundedness (if applicable) |
Citation validity %, no-answer calibration |
| Cost/latency |
Token usage, p95 latency vs. budget |
| Reviewer sign-off |
AI Governance Lead + domain owner |
| Release decision |
Approved / Rejected / Approved with conditions |
| Conditions/follow-ups |
e.g., “monitor fairness metric weekly for 1 month” |
Example scorecard (fake/demo data)
| Field |
Value |
| Skill name |
Candidate Matching |
| Prompt version |
matching-v1.3 |
| Model version |
model-tier-mid-v2 |
| Functional accuracy |
84% precision@5 (threshold: 80%) — Pass |
| Fairness result |
No significant disparity detected — Pass |
| Security/red-team result |
10/10 test cases blocked — Pass |
| Cost/latency |
p95 1.8s, within budget — Pass |
| Release decision |
Approved |
Storage
Completed scorecards are stored in evals/reports/ and referenced from the evaluation_record table (data-dictionary.md).
Change control
| Version |
Date |
Author |
Change |
| 1.0 |
2026-09-07 |
Documentation package generation |
Initial creation |