Clarity Gate — Pre-ingest Cognitive Quality Verification Tool for RAG Systems

Skill Overview

Clarity Gate is an open-source pre-ingest verification system designed to ensure documents entering a RAG knowledge base carry correct cognitive-quality markings, preventing LLMs from presenting unverified guesses as confident hallucinations.

Use Cases

1. RAG corpus quality control: When your knowledge base contains draft documents, tickets, meeting notes, or user-provided content, Clarity Gate can automatically check and add missing uncertainty markers before documents are ingested.

2. Scalable HITL verification: When you need to go beyond spot-checking in human-in-the-loop (HITL) workflows, Clarity Gate intelligently splits verification requests into "data confirmation" and "ground-truth verification" rounds, focusing human attention on the statements that truly require verification.

3. Automated ingestion gateway: When you need to enforce quality standards automatically before document ingestion, ensuring statements are correctly classified as fact, assumption, or prediction, and requiring human verification when markers are missing.

Core Features

1. 9-point verification system: Covers cognitive-quality checks (hypothesis vs. fact tagging, enforcement of uncertainty markers, hypothesis visibility, verification of authoritatively appearing data) and data-quality checks (data consistency, implicit causality, mislabeling of future states, temporal coherence, routing of externally verifiable claims).

2. Two-round HITL verification workflow: Intelligently routes statements requiring human verification to Round A (quick confirmation for statements with found sources) and Round B (statements needing true verification), optimizing human attention allocation.

3. Verify and Annotate modes: Verify mode generates a verification report and initiates a HITL workflow; Annotate mode directly produces a repaired Clarity-Gated Document (CGD) that can be safely ingested into a RAG system.

FAQ

How is Clarity Gate different from traditional fact-checking tools?

Traditional fact-checking tools focus on whether content matches sources (accuracy), whereas Clarity Gate focuses on whether a statement is properly qualified (cognitive quality). For example, treating "revenue is expected to reach 50 million" as a factual assertion might pass accuracy checks, but would fail cognitive-quality checks because a prediction is being presented as a fact. Clarity Gate fills the gap of enforcing cognitive constraints before ingestion.

Can Clarity Gate automatically verify the truth of facts?

No. Clarity Gate verifies form rather than veracity — it checks whether statements are properly marked as uncertain, but it cannot determine whether the statements themselves are true. This is why the two-round HITL verification is mandatory — LLMs can fabricate facts and add source markers to pass checks. HITL ensures real human verification before statements enter the knowledge base.

How do I use Clarity Gate in Claude?

There are several ways: 1) upload the dist/clarity-gate.skill file in the claude.ai web app; 2) use the same skill file in Claude Desktop; 3) when using Claude Code, clone the repository — it will auto-detect skills in the .claude/skills/ directory. To run it, simply say "Run clarity gate on this document" to start verification.