⊘ FME Framework V19

The Forensic Media
Evaluation Framework

FME is the multi-dimensional model behind every Rhetoric Audit analysis. Version 19 replaces the monolithic LLM pass with a four-stage forensic pipeline, adding span-level annotations, Aristotelian appeals, emotion arcs, and external claim grounding.

Origins

Why a Multi-Dimensional Framework

Most public bias tools collapse a complex piece of writing into a single label — left, right, true, false, fake, real. That compression destroys the very structure a careful reader needs in order to evaluate the work.

FME takes the opposite approach. It profiles a piece across a set of independent dimensions, each answering a narrow, well-scoped question. The dimensions can disagree with one another, and that disagreement is informative.

V19 Architecture

What Changed in V19

V19 is a structural overhaul. The user-facing 13 parameters are preserved for backward compatibility, but the engine underneath is a different tool. Three fundamental rewrites:

Single-pass LLM → 4-stage pipeline

The LLM no longer both observes and scores. Stage 1 annotates; Stage 2 runs deterministic math.

Document-level → span-anchored

Every score traces back to character offsets in the source text. No more opaque numbers.

Closed system → external knowledge

Stage 1.5 queries Google Fact Check Tools, Wikidata, and Wikipedia. Stage 1.6 adds real-time cross-platform signals from X, News, and Web to distinguish loaded rhetoric from false claims.

Cross-platform corroboration (Stage 1.6)

Runs parallel to Stage 1.5. Blends real-time signals into the Factual Grounding Index — FGI = Stage1.5 × 70% + cross-platform × 30% when ≥2 independent signals found.

Backward compatible surface

12 of 13 original parameters kept as-is. 1 replaced (emotion triad → Plutchik-8).

Core Model

The FME Dimensions

Philosophical Frame

The underlying worldview a piece is reasoning from.

Political Inclination

Where the piece sits on the political spectrum based on its rhetorical posture.

Manipulation RiskV19 ↑

How aggressively the text uses persuasion techniques that bypass deliberate thinking.

Fallacy DensityV19 ↑

The concentration of common logical fallacies across the piece.

Evidence ValidityV19 ↑

How well the central claims are anchored in verifiable, attributable evidence.

Emotion ArcV19 NEW

Plutchik-8 emotion vector per paragraph, labeled with a trajectory type.

Intent Transparency

How openly the piece discloses its perspective or agenda.

Source Reliability

An assessment of the cited sources and upstream credibility.

Strategic Silence

Material context or counter-evidence that is conspicuously absent.

Narrative Archetype

The story shape the piece falls into — hero/villain framing, etc.

External GroundingV19 NEW

Claims verified via Fact Check Tools, Wikidata, and Wikipedia (Stage 1.5).

Cross-Platform CorroborationV19.1 NEW

Real-time signals from X, News, and Web corroborate or contradict article claims. Blended into FGI at 30% weight when ≥2 signals found (Stage 1.6).

See FME V19 applied to a real article

Install the Chrome extension or sign up to run your first forensic analysis. Free credits included on signup.