This page explains how the Executive Order Analysis Project collects, categorizes, and analyzes presidential executive orders. It is written for researchers, journalists, and analysts who want to understand, evaluate, or build on this work.
This project was built and is maintained by Matthew Tushman, a financial services professional with over 33 years of experience in wealth management and asset management. He is not an attorney, academic, or policy analyst by training. The project began in April 2025 using Python and artificial intelligence, with AI generating the majority of the code. The analytical frameworks, category definitions, and interpretive conclusions are his. AI tools are used extensively, given the volume of text analyzed; the substantive decisions and direction are his own.
The project is strictly nonpartisan and will never be monetized. The methodology is fully reproducible. All analytical limitations are disclosed explicitly in each analysis. When this work is cited, it should be attributed to an independent researcher using computational methods, not to a university-affiliated expert or legal professional.
More background is available on the About page.
The analysis covers the full text of all presidential executive orders published in the Federal Register since January 20, 2025. Orders are collected from the Federal Register after publication, typically two to three days after signing. The project currently covers 244 executive orders, comprising over 292,000 words and more than 5,400 meaningful sentences.
The Federal Register is used as the primary source because it contains the official legal text as published. White House releases are used for supplementary metadata where needed. Only executive orders are included; presidential memoranda, proclamations, and other executive actions are outside the current scope.
Each executive order is assigned a single primary category from a predefined set of 26 categories. Classification uses a dual-model validation approach: an initial categorization is produced by one AI model, reviewed by a second AI model applying the same analytical framework, and the first model independently evaluates that feedback before confirming or revising. The final decision is always made by the first model. A written rationale is saved for every order and is publicly available on the Category Rationales page.
Subject matter first. Classification is always based on what the order is primarily about: the subject it addresses and the entities it directly affects, not the procedural mechanisms it uses to get there. An order that uses budget tools to target a specific policy area is classified by the policy area, not by the budget mechanism.
Specific overrides general. When an order directly targets a specific type of entity that corresponds to a named category, that specific category takes precedence over broad governance categories. An order primarily targeting a law firm is classified as Law Firms, not Government Reform and Efficiency.
Conflict resolution by text coverage. When multiple categories could reasonably apply, the category that receives the most text and attention in the order is selected. If coverage is roughly equal, the category addressed first in the title or opening section is preferred. If still unclear, the more specific category is chosen over the more general one.
Government Reform and Efficiency is a residual category. This category is selected only when the order's primary purpose is to reform government operations or processes and no more specific category applies. It is not used when another category directly matches the subject matter.
National identity, patriotism, citizenship, and cultural heritage. Distinct from Historical Recognition, which covers specific monuments, memorials, and commemorations.
Orders addressing individual rights, civil liberties, free speech, due process, and constitutional protections.
Climate change, greenhouse gas emissions, global warming, and international climate agreements. Distinct from Energy & Environmental Policy, which covers energy production regulations not centered on climate.
Orders specifically addressing pandemic-era mandates, vaccine requirements, public health emergency declarations, and COVID-related policy reversals.
Armed forces structure, military readiness, defense spending, veterans policy, and military personnel matters.
Diversity, equity, and inclusion programs in federal agencies, contractors, and institutions receiving federal funding.
Broad economic strategies, taxation, trade policy, tariffs, and monetary policy. Distinct from Fiscal Management (government spending oversight) and Economic Prosperity (job creation and growth initiatives).
Job creation, economic growth, workforce development, and industry-specific development initiatives. Distinct from broad Economic & Fiscal Policy.
K-12 and higher education policy, accreditation, student loans, and federal education funding.
Voting rights, election administration, voter registration, and electoral process integrity.
Energy production, fossil fuels, renewables, and environmental regulations for energy operations not centered on climate change. Distinct from Climate & Environmental Policy and Energy Security & Infrastructure.
Energy independence, grid security, pipeline projects, energy supply reliability, and strategic energy infrastructure. Distinct from Energy & Environmental Policy, which covers regulations.
Religious organizations, faith-based service providers, religious exemptions, and the role of religious institutions in federally funded programs.
Government spending, budgeting, debt management, and financial oversight of federal funds. Distinct from Economic & Fiscal Policy, which covers broad economic strategy.
Diplomatic relations, international agreements, and responses to foreign government actions. When tariffs or sanctions are used primarily as diplomatic tools, this category applies. Distinct from Economic & Fiscal Policy when the primary purpose is trade management rather than diplomacy.
Reforming government operations, reducing bureaucracy, reorganizing agencies, and improving federal processes. Selected only when no more specific category applies.
Designating monuments, renaming locations, creating memorials, and recognizing historical events or figures. Distinct from Civic Identity, which addresses broader national identity themes.
All aspects of immigration policy: border control, visa processing, refugee admissions, deportation, removal procedures, and immigration screening. A comprehensive category covering the full immigration policy domain.
Criminal enforcement, police activities, prosecution priorities, criminal justice policy, and orders reforming or investigating the justice system. Covers both the exercise of law enforcement authority and oversight of it.
Orders primarily targeting specific law firms. Any order whose primary subject is a named law firm is classified here regardless of the mechanisms used.
Intelligence community operations, national security designations, counterterrorism, and foreign threat assessment.
Extraction rights, resource management on federal and public lands, mining policy, and access to natural resources for economic development.
Healthcare policy, prescription drug pricing, public health programs, and health agency oversight. Distinct from COVID-19 Policies, which is limited to pandemic-specific actions.
Reducing or restructuring federal regulations across agencies, deregulation initiatives, and regulatory review processes.
Technology development, AI policy, digital infrastructure, and tech sector regulation. Applies only when technology is the primary subject, not an incidental component of another domain.
Transportation networks, physical infrastructure projects, and federal infrastructure investment. Distinct from Energy Security & Infrastructure, which is limited to energy systems.
Several category pairs require careful judgment. These are the most common close calls and how they are resolved.
Climate & Environmental Policy applies when an order's primary focus is climate change, emissions reduction, global warming, or international climate agreements.
Energy & Environmental Policy applies when the primary focus is energy production regulations, renewable energy mandates, fossil fuel policy, or environmental standards for energy operations that do not center on climate.
Foreign Policy & International Relations applies when the primary purpose is managing diplomatic relations or responding to foreign government actions, even if economic tools such as tariffs or sanctions are used to do so.
Economic & Fiscal Policy applies when the primary purpose is managing U.S. trade policy, protecting domestic industries, or addressing trade deficits, even if specific countries are targeted.
Economic & Fiscal Policy: Broad economic strategies, taxation, trade, and monetary policy.
Fiscal Management & Accountability: Government spending, budgeting, and federal financial oversight.
Economic Prosperity & Development: Job creation, workforce development, and industry-specific growth initiatives.
Technology & Innovation Policy applies only when technology development, AI, or tech sector regulation is the primary subject matter. If technology is a component or tool within another domain, such as cyber elements in national security or digital learning in education, the primary domain category is used instead.
Civic Identity & National Heritage applies to broader themes of national identity, patriotism, and cultural heritage.
Historical Recognition & Commemoration applies specifically to orders designating monuments, renaming locations, creating memorials, or recognizing specific historical events or figures.
Sentiment analysis is applied at two levels. Document-level analysis uses a structured four-part AI framework to characterize tone, framing, and rhetorical strategy across each order. Sentence-level analysis uses the VADER (Valence Aware Dictionary and sEntiment Reasoner) framework, developed in 2014 and widely used in computational linguistics research, to score individual sentences numerically.
VADER assigns each unit of text an overall score ranging from −1.0 (most negative) to +1.0 (most positive), based on a dictionary of words and phrases rated for emotional tone. Scores above +0.05 are generally considered positive; scores below −0.05 are generally considered negative. Rather than using the VADER software library directly, scores in this project are generated by an AI model prompted to apply VADER's scoring framework at a fixed setting, meaning the same input always produces the same output on repeated runs. Results closely align with library-computed scores in direction and relative magnitude.
Each executive order receives a structured four-part sentiment analysis:
Characterizes the dominant emotional register of the order and how it changes across sections: for example, from adversarial framing in opening sections to procedurally neutral language in implementation sections.
Identifies positive, negative, and neutral language as framed by the order itself. Separately notes whether major factual claims are supported by citations, evidence, or sourced references within the order.
Tracks how rhetoric evolves through the order's structure, linking tonal choices to stated policy aims in each section.
Evaluates how sentiment relates to the order's substantive goals, potential implications for named stakeholders, comparison to typical executive order language, and the order's character as a political document. Includes explicit discussion of analytical limitations and potential biases.
Every meaningful sentence in each executive order receives an individual VADER overall score, a positive score, a neutral score, and a negative score. An AI-generated commentary explains the primary words and phrases most likely driving each sentence's score, and notes where a word's dictionary meaning may differ from its intended meaning in a policy context. This sentence-level data underlies the annotated insight analyses published in the Insights section.
Analysis of all executive orders has identified several recurring rhetorical structures that appear across most orders, regardless of policy domain:
VADER was designed for social media and informal text. It has well-documented limitations with formal legal and policy language, where words like "eliminate," "deprive," or "threat" may carry negative dictionary-based scores regardless of their constructive policy intent. Automated scores are best treated as a structured lens rather than ground truth.
Sentiment measures words, not intent. A sentence that is substantively protective or cooperative can score as negative because of the vocabulary it uses. The commentary layer in each analysis is designed specifically to surface these gaps.
Category assignments reflect primary subject matter. Executive orders often span multiple policy domains. The single primary category assigned captures the dominant subject; secondary categories are documented in the rationale files but are not used for grouping or ranking.
This is applied analysis, not peer-reviewed scholarship. The frameworks, classifications, and interpretations are those of an independent researcher. They should be treated as high-quality applied analysis and paired with primary sources (the orders themselves as published in the Federal Register) when used in academic or professional work.
This analysis is based solely on the text of each executive order as published. It does not incorporate external context such as court rulings, agency guidance, legislative history, empirical research, or other public sources that may bear on the claims made within an order. Factual assertions in the orders are assessed as written, not against independent evidence.
When referencing analysis from this project, please attribute it to Matthew Tushman, Executive Order Analysis Project, Fetaverse, fetaverse.net/executive_orders/, with the year of the specific analysis cited. The project is licensed under CC BY-NC 4.0: you may share, quote, and adapt this content for non-commercial purposes, including journalism, education, and policy analysis, provided attribution is given.
Questions about the methodology can be directed to the contact information on the About page.