The Research & Source-Credibility Gate turns your agent into a skeptical reviewer that pressure-tests research before anyone acts on it — and then stops, leaving the verifying and the final call to you.
It runs five fixed passes over a brief, summary, or AI-generated answer: a claim-support check (which load-bearing claims are actually backed by a source versus merely asserted), a source-quality check (primary vs secondary, authoritative vs anonymous, peer-reviewed vs promotional), a recency-and-relevance check (stale sources behind current-data claims, and sources that don’t actually say what the claim attributes to them), a bias-and-completeness check (cherry-picking, one-sided framing, omitted caveats and conflicts of interest), and a fabrication-risk check (vague or unverifiable citations, suspiciously perfect quotes, statistics with no traceable origin).
Every run returns a structured verdict: TRUST, VERIFY, or REJECT, plus a 0-100 confidence rating, a breakdown of each load-bearing claim as supported, unsupported, or misattributed, the specific weaknesses with the passage quoted, a must-verify list, and a one-line recommendation. Because it separates load-bearing claims from filler and never invents sources, it focuses your attention on what actually matters for a decision instead of being fooled by confident tone.
Who it’s for: anyone who relies on research they didn’t personally verify — analysts, consultants, founders, students, journalists, and teams using AI to generate answers and briefs. It is model-agnostic and read-only: it has no network access and cannot browse, open links, or confirm a source exists. It identifies where research is weak and what a human must check. It is a credibility gate, not a fact database or an oracle.