Methodology

Version

v1 — Initial release. Locked 2026-05-17.

Decision tree — how a claim becomes a verdict

1. Transcription — Yentl converts audio or media into transcript text with timestamps. Biometric speaker tagging is disabled by default in v1 for privacy compliance.
2. Claim extraction — Yentl identifies discrete factual claims in the transcript segment. Pure opinions, jokes, rhetorical questions, and statements with no checkable proposition are not treated as factual verdicts.
3. Scope screen — Each claim is checked for whether it is appropriate to assess, needs caution, or should be left without a factual verdict. See the scope rules section below.
4. Initial check — Yentl searches for evidence, retrieves sources, and evaluates whether those sources support, contradict, or complicate the claim while deeper review continues.
5. Reviewed verdict — After source reconciliation, the visible label becomes Supported, Mixed, False, No reliable backing, or Opinion.
6. Bias / fallacy / rhetoric markers — In parallel, Yentl identifies cognitive biases, logical fallacies, and rhetorical patterns from the Yentl taxonomy (123 entries, CC-BY-4.0).

Reputation tier definitions

High

Peer-reviewed journals, major national news organizations with documented editorial standards, government statistical agencies, established international institutions (WHO, UN bodies, etc.). Strong track record of corrections.

Medium

Credible regional outlets, specialized trade publications, think tanks with disclosed funding, official organization statements. Some editorial standards present; possible institutional bias.

Low

Blogs, opinion sites, advocacy organizations without editorial independence, sources with documented history of inaccuracy or partisan framing, social media posts, anonymous sources.

Marker taxonomy

Yentl detects 123 markers across three categories: cognitive biases (55 + 28 extras), logical fallacies (25 extras), and rhetorical patterns (15). The full machine-readable taxonomy is available at /taxonomy.json (CC-BY-4.0). Primary source: Cognitive Biases & Logical Fallacies Used by Antisemites by Israel B. Bitton (2024).

Scope rules

Yentl checks claims when there is a public factual proposition and enough context to search for evidence responsibly. In summary:

Check

Verifiable factual claim about public facts, public figures in their public roles, scientific consensus, or historical record.

Check carefully

Contested empirical matters where reasonable experts disagree; Yentl includes confidence level, named dissenting positions, and extra source requirements.

Treat as opinion or context

Pure opinions, jokes, rhetorical questions, claims with no verifiable proposition.

Do not verdict

Private-individual harassment vectors, doxxing, hate speech, extremist or threatening content, CSAM, defamation-trap setups. Yentl avoids turning these into verdict-like output.

Prompt-version log