vcsdd-grading-criteria

name	vcsdd-grading-criteria
description	Use this skill when evaluating VCSDD grading dimensions during adversarial review. Provides concrete evaluation checklists for each of the 5 dimensions and convergence signal detection.
origin	VCSDD

VCSDD Grading Criteria

When to Activate

• During Phase 3 adversarial review
• Evaluating convergence signals in Phase 6
• Writing evidence blocks for verdicts

Dimension Evaluation Checklists

spec_fidelity

• Extract all REQ-XXX IDs from behavioral-spec.md
• For each REQ-XXX: find the corresponding test(s) in test files
• For each REQ-XXX: find the corresponding implementation
• Check for unspecified features in implementation

edge_case_coverage

• Extract edge case catalog from behavioral-spec.md
• For each edge case: find the corresponding test
• Identify edge cases the spec MISSED (fuzzer thinking)
• Check boundary values (off-by-one, MAX_INT, empty collections)

implementation_correctness

• Trace each function's logic against its spec requirement
• Check all error paths (what happens when dependency fails?)
• Look for null/undefined dereferences
• Check concurrent access assumptions
• Look for resource leaks (unclosed files, connections)

structural_integrity

• Functions are focused (single responsibility)
• Names match their purpose precisely
• No copy-paste duplication
• Abstraction level consistent with surrounding code
• No dead code or commented-out blocks

verification_readiness

• Pure functions take only primitives/data, return Result/Either
• Side effects confined to effectful shell modules
• Functions are small enough to reason about formally
• Proof obligations from verification-architecture.md are reflected in code structure

Evidence Requirements

Every PASS verdict must include evidence blocks with:

• type: file | test-output | coverage | proof-result
• location: real file path (must be verifiable with ls)
• description: what was verified at that location