audit-response-validator

name: audit-response-validator description: > Validates AiVIS Cite Ledger audit response payloads against the canonical server-side CITE LEDGER contract. Use this skill when reviewing, generating, or repairing AI audit JSON that must pass deterministic validation in server/validators/validateAuditResponse.ts. It enforces the exact seven- dimension model, blocks legacy labels, checks score consistency, and ensures every recommendation and remediation path is evidence-backed.

Audit Response Validator

Use this skill to verify that an AI-generated audit payload is structurally valid, deterministic, and traceable to the AiVIS Cite Ledger ten-family (avs-v3) CITE LEDGER model.

Validation order

Follow this order exactly.

1. Run base schema validation first.
2. Reject the payload immediately if the schema fails.
3. Validate CITE LEDGER structure and dimensional integrity.
4. Validate score consistency and summary-layer counts.
5. Validate recommendation and evidence traceability.
6. Return a concise pass/fail result with exact error messages.

Canonical source of truth

Treat server/validators/validateAuditResponse.ts as the authoritative contract.

Do not invent extra sections, rename dimensions, or soften validation rules.

Required audit structure

The payload must contain:

• visibility_score
• ai_platform_scores
• summary
• key_takeaways
• topical_keywords
• keyword_intelligence
• brand_entities
• primary_topics
• faq_count
• category_grades
• content_highlights
• recommendations
• crypto_intelligence
• cite_ledger

The cite_ledger block must contain:

• cite_ledger_score
• data_confidence
• dimensions
• brag_entries
• hard_blockers
• verdict
• top_fixes
• citation_probability

Seven-dimension enforcement

The only valid dimensions are:

• Schema & Structured Data
• Content Depth
• Technical Trust
• Meta Tags & Open Graph
• AI Readability
• Heading Structure
• Security & Trust

Apply these checks:

• Exactly 7 dimensions must be present.
• No duplicate dimension names.
• No legacy labels are allowed.
• Every required dimension must exist.
• Every required weight must match the canonical value.
• Total dimension weight must equal 100.

Score consistency checks

Enforce these invariants:

• visibility_score must equal cite_ledger.cite_ledger_score.
• category_grades must contain exactly 7 entries.
• cite_ledger.brag_entries cannot be empty.
• cite_ledger.top_fixes must contain exactly 5 entries.

Recommendation checks

Every recommendation must be evidence-linked and measurable.

Validate that:

• priority is valid.
• description starts with Measured:.
• difficulty is one of the allowed values.
• implementation is concrete and sufficiently detailed.
• evidence_ids is non-empty.
• every evidence_id begins with ev_.
• estimatedVisibilityLoss matches a percentage range such as 10-20%.

Evidence and remediation quality gates

Treat these as completion criteria for a valid audit payload:

• Findings must be grounded in observable page or ledger evidence.
• Remediations must point to a specific fix path.
• No hallucinated dimensions, labels, or sections.
• No legacy terminology from older scoring systems.
• No partial schema acceptance.

Failure handling

If any rule fails, return:

• valid: false
• a complete list of exact validation errors

If the schema passes but a business rule fails, report the business rule error separately and do not downgrade it to a warning.

Output format

Always produce one of these two outcomes:

Pass

• valid: true
• errors: []

Fail

• valid: false
• errors: [exact messages]

Example use cases

• validating a new AI audit response before saving it
• repairing a payload that changed dimension names or weights
• checking that remediation items still point to evidence IDs
• blocking legacy category drift before it reaches the client