arrow_back Back to Dashboard

About FashFinder

With the rise of fascist rhetoric and authoritarian actions in the United States—particularly under the Trump administration and among aligned political movements—there is an urgent need to systematically document and analyze these patterns. FashFinder is an automated monitoring system that tracks government statements, executive actions, and investigative journalism to identify and classify rhetoric and policies that align with established frameworks of fascism.

The Framework

FashFinder's classification system is based on Jason Stanley's seminal work "How Fascism Works: The Politics of Us and Them", which identifies ten pillars of fascist politics:

  • Mythic Past – Appeals to a fabricated glorious past requiring restoration
  • Propaganda – Systematic undermining of shared truth and reality
  • Anti-Intellectualism – Attacks on expertise, education, and critical thinking
  • Unreality – Conspiracy theories about hidden enemies
  • Hierarchy – Rigid social divisions based on group identity
  • Victimhood – Claims of persecution to justify hostility
  • Law and Order – Selective legal application favoring in-groups
  • Sexual Anxiety – Enforcement of rigid gender hierarchies
  • Sodom and Gomorrah – Moral panic requiring purification
  • Lebensraum – Territorial and demographic displacement narratives

Each category contains detailed subcategories and specific indicators. See the full taxonomy reference for complete details, including a visual diagram of how these categories interact.

How It Works

FashFinder operates through a fully automated pipeline that runs every 12 hours:

1

Scraping

Collects content from 6 sources: White House, Executive Orders, ProPublica, The Intercept, Reveal, The Guardian

↓
2

Screening (AI - Fast Model)

Quickly filters out benign content (~80% filtered): infrastructure announcements, routine diplomacy, holiday messages

↓
3

Classification (4-Stage Pipeline)

Stage 1: Identify categories (1-3 per article)
Stage 2-3: Deep analysis of each category with quotes and evidence
Stage 4: Cross-category pattern analysis and severity rating (1-4)

↓
4

Publication

Automatically commits to repository, triggers deployment, updates dashboard (2-3 minute latency)

↓
5

LLM Pattern Analysis

AI generates narrative analysis of detected patterns: identifies systemic trends, escalation risks, rhetoric→action progressions, and provides comparative context with historical fascist movements. Uses Tavily web search to gather academic sources and historical precedents for deeper analysis.

Full pipeline runs every 12 hours (12 AM & 12 PM Pacific) via GitHub Actions
Analysis updates immediately after each scraping session

Important: Official government sources capture rhetoric (what they say publicly), while investigative journalism reveals actions (what they do, including undisclosed policies). Both are essential for a complete picture.

The Dashboard

The FashFinder dashboard organizes data into four main tabs for easy exploration:

  • Overview – Summary statistics, AI-generated narrative analysis, key findings, and recent classifications (sorted by publication date)
  • Classifications – Full searchable list with filtering by category, severity, confidence, source, and date range
  • Analytics – Interactive charts showing severity distribution, top categories, rhetoric vs. actions, and trends over time
  • Insights – AI narrative analysis with systemic patterns, risk assessments, and historical context

Each classification card shows direct quotes with explanations, corroborating sources when available, and links to the original source material. The interface uses a modular TypeScript architecture with components organized by concern (UI, visualization, data management).

Multi-Source Strategy

Authoritarian governments often use euphemistic language in official statements while implementing harsh policies through bureaucratic channels. FashFinder addresses this by:

  • Official Sources – Capture public rhetoric and stated positions
  • Investigative Sources – Reveal hidden actions, leaked documents, and unreported implementations

Each classification shows its complete evidence chain, allowing you to verify claims and distinguish between rhetoric and documented actions.

Severity Calibration

FashFinder uses a violence-based severity framework calibrated to historical fascism from the 1920s through 1940s. This calibration is critical for two reasons:

  • Analytical Headroom: By calibrating to early-to-mid stage fascism, we can track current patterns while preserving space to measure actual late-stage atrocities if they occur
  • Violence as Threshold: The key distinction between severity levels is whether state violence has been deployed, not just the scale of oppression

Four-Level Severity Scale

Severity 1 → Early 1920s

Rhetoric Without Implementation
Authoritarian rhetoric, appeals to mythic past, propaganda, conspiracy theories, victimhood narratives. Policy proposals that haven't been enacted.

Severity 2 → Late 1920s - Early 1930s

Oppressive Policies/Systems WITHOUT Violence
Discriminatory laws and regulations, systematic surveillance and list-making, institutional capture (courts, agencies, media), rights restrictions, funding weaponization. Can be widespread but no direct violence yet.
Examples: Secret terrorist lists, discriminatory bans, Nuremberg-style laws

Severity 3 → Late 1930s

State Violence Begins
Political arrests and prosecutions, extrajudicial killings, detention without trial, concentration camps, organized state repression, police/military violence against civilians.
Examples: Arrests from terrorist lists, political prisoner detentions, violent crackdowns

Severity 4 → 1940s

Mass Atrocities
Genocide, ethnic cleansing, death camps, industrial-scale systematic killing, mass deportations resulting in deaths, widespread crimes against humanity.
Examples: Holocaust, genocidal campaigns, systematic extermination

Critical Distinction: Secret terrorist lists = Severity 2. Arrests and prosecutions based on those lists = Severity 3. The threshold is violence, not scope. This calibration allows us to track early warning signs while preserving analytical space for escalation.

Pattern Analysis

Beyond individual classifications, FashFinder tracks patterns using a three-metric system that provides a comprehensive view of trajectory and risk:

Three-Metric Dashboard

timeline Fascism Stage (1-4)

Current position on the historical fascism timeline based on average severity. Maps to historical periods: Stage 1 (early 1920s), Stage 2 (early 1930s), Stage 3 (late 1930s), Stage 4 (1940s).

local_fire_department Current Intensity

Activity level in the last 30 days (mild/moderate/severe/extreme). Measures recent classification frequency and average severity.

show_chart Momentum

Trajectory and velocity (stable/accelerating/rapid/critical). Combines acceleration rate, temporal clustering, severity trends, and rhetoric→action progressions to assess whether patterns are escalating.

Statistical Pattern Detection (runs every 12 hours with new data):

  • Three-metric dashboard (Stage, Intensity, Momentum)
  • Severity trends over time (rising, falling, stable)
  • Category co-occurrence patterns (how categories cluster together)
  • Rhetoric vs. action ratios (are statements being operationalized?)
  • Rhetoric→action progressions (tracking when rhetoric precedes policy implementation)

LLM Narrative Analysis (runs after each scraping session):

  • Qualitative interpretation using the three-metric framework
  • Assessment of progression toward violence threshold (Sev 2 → Sev 3)
  • Identification of systemic patterns across categories
  • Risk assessment with historical context calibrated to early-to-mid stage fascism
  • Comparative analysis with 1920s-1930s historical precedents

The combination of statistical and narrative analysis provides both rigorous quantitative metrics and meaningful qualitative insights. Statistical patterns catch what's happening, while LLM analysis explains why it matters and what historical precedents exist. The three-metric system ensures we understand not just where we are (Stage), but also how active it is (Intensity) and where it's headed (Momentum).

Why We Display Costs

FashFinder is committed to full transparency about its operations. The cost tracking dashboard publicly displays:

  • Exact API costs for each classification
  • Per-stage breakdown showing which analysis steps are most expensive
  • Monthly cost trends and projections
  • Token usage and processing efficiency metrics
  • LLM pattern analysis costs (weekly narrative generation, ~$0.035 per run)

This transparency serves several purposes:

  • Accountability – Anyone can verify that resources are used efficiently
  • Reproducibility – Other researchers can estimate costs for similar projects
  • Sustainability – Public cost data helps identify optimization opportunities
  • Trust – Openness about methodology and expenses builds credibility

By making our costs visible, we demonstrate that this is a serious, well-designed research tool rather than an opaque "black box" system.

Methodology & Limitations

FashFinder uses state-of-the-art AI language models (Claude Sonnet 4.5) to analyze content against Stanley's taxonomy with explicit prompts calibrated to historical fascism. Each classification includes:

  • Severity levels (1-4) calibrated to historical fascism timeline (1920s-1940s) with violence as the critical threshold
  • Confidence ratings reflecting analytical certainty
  • Direct quotes extracted from source material with exact text
  • Detailed explanations of why each quote matches specific taxonomy patterns
  • Cross-category analysis identifying reinforcing patterns

Full Transparency: See the exact prompts used by the AI classification system on our AI Prompts page. These prompts are automatically extracted from source code to ensure accuracy.

Important limitations:

  • Pattern Detection, Not Prediction: This system identifies rhetoric and policy patterns that align with academic frameworks of fascism. It does not make political judgments, predict outcomes, or claim that any individual or administration is fascist.
  • Calibration Trade-offs: By calibrating to early-to-mid stage fascism (1920s-early 1930s), we preserve analytical headroom for escalation. This means current Severity 2-3 classifications represent serious patterns but are not equivalent to late-stage genocide.
  • AI Limitations: While sophisticated, AI classification can miss context, make errors, or reflect biases in training data. All classifications should be viewed as analytical tools, not definitive judgments.
  • Source Limitations: We rely on publicly available documents and investigative journalism. Undisclosed policies and classified actions may not be captured.

Data Sources: White House, Executive Orders, ProPublica, The Intercept, Reveal, The Guardian
Framework: Jason Stanley, How Fascism Works (2018)
Last Updated: November 2025