BRAG Evidence Gates: How AiVIS Cite Ledger Uses Proof-First Citation Filtering | AiVIS Cite Ledger Blogs
By R. Mason · · 13 min read · INFRASTRUCTURE
BRAG is the filtering layer that stops AI visibility tools from reporting findings they cannot prove. Here is what the system is, how it was customized for AiVIS, and why it changes citation audit reliability.
Key Takeaways
- BRAG (Based, Retrieval, Auditable, Grading) is a filtering standard that prevents tools from reporting findings they cannot prove.
- AiVIS implements BRAG validation with Python-first, TypeScript-fallback architecture for reliable evidence checking.
- Every finding must be traced to specific evidence before receiving a BRAG ID and appearing in your audit report.
- The evidence-key mapping system links failed rules to crawl observations, ensuring findings are based in page reality.
- Post-fix re-audits validate that changes resolved the finding, closing the loop between detection and correction.
Article
# BRAG Evidence Gates: How AiVIS Cite Ledger Uses Proof-First Citation Filtering
The problem: audit tools that report findings they cannot prove
Most AI visibility platforms operate on a simple model: crawl a page, run a model, return a score.
This creates a structural defect. The model generates plausible-sounding findings, but those findings are not traced to specific, observable page data. When teams attempt to fix the reported issues, the feedback loop breaks—re-auditing produces different results, or no measurable improvement.
This is not a theoretical risk. In production audits, we identified the failure point: tools reporting findings they cannot defend with evidence.
The solution: BRAG evidence gates
BRAG (Based, Retrieval, Auditable, Grading) is a filtering standard that enforces one rule: **if a finding cannot be traced to specific page data, it does not appear in the output.**
In AiVIS, this is implemented through evidence-key mapping: every failed rule must link to a specific evidence item from the crawl. If the evidence key cannot be found in the evidence map, the finding is rejected and does not receive a BRAG ID.
The system validates:
- Rule results against evidence items using RULE_EVIDENCE_MAP
- AI recommendations against evidence items they claim to address
- Severity weighting that determines priority (critical: 15 points, high: 8, medium: 4, low: 1.5)
What BRAG prevents: a real example
**Without BRAG:** A model suggests "add product price to schema" for a SaaS company that has never sold products. The finding is a hallucination—plausible-sounding but not grounded in page reality.
**With BRAG:** The finding is checked against crawl data. No product price field exists in schema. The system checks business type from page content. Finding is dropped as non-applicable. The recommendation never reaches your output.
This is measurable. The BRAG gate rejects findings that cannot be traced to evidence, preventing hallucinated recommen
Enable JavaScript for the full interactive reading experience with related articles and discussion.