Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →nwave-ai/nw-der-review-criteria
Evaluation criteria and scoring for data engineering artifact reviews
Passed 22 of 24 test cases with the skill, and 9 of 24 without it.
and 10 more in the full report
13 fixed9 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
A proposal features verified SQL dialect syntax, correct B-tree index selection for range scans, and appropriate write-heavy streaming patterns. What score is assigned for Technical Accuracy?
| Checked | Without skill | With skill |
|---|---|---|
| Awards a score of 10 for Technical Accuracy | Missed | Passed |
The instructions your model reads when the skill loads.
Evaluation criteria for each review dimension. Load when performing reviews.
Evaluate whether recommendations trace to specific evidence.
Checks: Each major recommendation cites a specific research finding | Citations are accurate (finding number matches content) | Vendor-specific claims have multiple independent sources | General best practices distinguished from research-validated guidance
Scoring: 10: All cited, verified | 7: Most cited, 1-2 missing on non-critical points | 4: Major recommendations lack citations | 0: No citations
Evaluate defense-in-depth for data layer.
Checks: Encryption at rest (TDE) | Encryption in transit (TLS) | Access control model (RBAC/ABAC) | SQL injection prevention (parameterized queries) | OWASP/NIST standards referenced | Credential handling (no hardcoded secrets)
Scoring: 10: All 6 checks with standard references | 7: 4-5 checks | 4: 2-3 checks, missing encryption or injection prevention | 0: Security not mentioned
Evaluate balanced presentation of alternatives.
Checks: Multiple technology options (minimum 2) | Pros/cons for each | Context factors identified (scale, consistency, latency, cost) | Recommendation justified by context fit | Limitations acknowledged
Scoring: 10: Comprehensive trade-offs with context-driven justification | 7: Trade-offs present, some alternatives missing | 4: Single recommendation without alternatives | 0: Prescriptive with no analysis
Checks: SQL/NoSQL syntax correct for specified DB | Architecture patterns appropriate for use case (OLTP vs OLAP, write-heavy vs read-heavy) | Optimization strategies valid for target DB | Normalization level appropriate for workload | Index type matches query patterns (B-tree for range, hash for equality) | CAP trade-offs correctly applied
Scoring: 10: All technical claims verified | 7: Minor syntax/edge-case issues | 4: Significant errors affecting recommendations | 0: Fundamentally incorrect guidance
Checks: Scaling strategy (vertical, horizontal, sharding, replication) | Performance characteristics (query patterns, bottlenecks) | Data governance when applicable (lineage, quality, MDM) | Compliance when personal/regulated data involved (GDPR, CCPA, HIPAA) | Backup/recovery for production designs | Monitoring/observability
Scoring: 10: All applicable aspects covered | 7: Core covered, 1-2 peripheral missing | 4: Major gaps (missing scaling or governance for production) | 0: Only immediate question, no broader context
test cases passed, vs 9 / 24 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.