The Written Rules: Every Rule File and What It Governs
Last updated: November 20, 2026
Most of what we promise about how we work is written down in plain text files. This page lists every one of them. As of November 20, 2026 there are 33 rule files in our .cursor/rules/ folder, and this catalog covers all 33.
It is one part of Rules, Gates and Guardrails: How We Constrain Our AI and Ourselves. Some rules are also backed by a check in code. Others are written policy that an editor applies by judgment. The family articles say which is which, and we do not claim more enforcement than exists.
How rule files work
Each rule file has a header that says what it covers and, usually, which files it applies to, written as a pattern such as “every article under the people folder.”
When one of our AI coding or writing tools opens or edits a file that matches a rule’s pattern, the tool loads that rule automatically and the AI is expected to follow it. A few rules are marked “always apply” and are loaded in every session whatever file is open: as of November 20, 2026, 5 of the 33 are marked that way. They are the shared-working-tree rule, the proactive-issue-surfacing rule, the knowledge-graph rule, the skill-routing rule and the local-services rule.
Loading a rule is guidance, not enforcement. The AI can still get something wrong, which is why the most important rules also have a check in code, a human review step, or both.
Who may change a rule
Rule files are protected. A hook in our AI coding tool refuses writes to them, and our project instructions say that changing one takes an explicit instruction from Peter, in his own words, naming the change. Many rule files carry a dated edit record saying what changed and what instruction authorized it. When an AI thinks a rule is stale or wrong, it flags the problem for a human instead of fixing it.
The catalog
The “Applies to” column is a plain-English reading of each file’s pattern list.
Evidence, sourcing and language rules
These are covered in Evidence, Sourcing and Language Rules.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
patriot-accountability-profile-standards.mdc |
The evidence tiers, wording for criminal claims, citation format, priority tiers, “speech is not a crime,” the Basis for Inclusion block, the federal-judge threshold, the factcheck notice marker, and the prohibited-language list | Accountability profiles of people and organizations | Evidence, Sourcing and Language Rules, Inclusion and Fairness Gates |
patriot-numbers.mdc |
Every percentage names its base, every “X times” gives both figures, every rate has a denominator and a year, and cross-year dollars are adjusted or say they are not | Any knowledge-base article | Evidence, Sourcing and Language Rules |
archive-before-cite.mdc |
Capture a web page at the moment it is cited for a contested claim; archive.today is never the capture of record; a missing archive snapshot is not proof a page never existed | Any knowledge-base article | Evidence, Sourcing and Language Rules |
patriot-editorial-framework.mdc |
The Civic Accountability Format for articles, briefs, tracker entries and explainers | The editorial-framework skill file | Evidence, Sourcing and Language Rules |
patriot-opinion-voice.mdc |
When Peter’s first-person opinion voice applies, for op-eds and anything marked Opinion | The opinion-voice skill file | Evidence, Sourcing and Language Rules |
patriot-substack.mdc |
The email-first Substack format and which voice sits underneath it | The Substack and coaching-voice skill files, and Substack articles | Evidence, Sourcing and Language Rules |
patriot-constitutional-law-advisors.mdc |
For legal topics, choose at least two legal sources from different methodological or ideological starting points instead of the single most visible commentator | Know-your-rights, voting, people, organization, background and founding-document content | Evidence, Sourcing and Language Rules |
patriot-obsidian-wiki.mdc |
The Markdown subset and the exact front-matter format that keep our editing tool from corrupting publish-ready files | Any knowledge-base article | Evidence, Sourcing and Language Rules, Draft Checks and Human Review |
Inclusion and fairness rules
These are covered in Inclusion and Fairness Gates. The profile-standards file above also holds the inclusion rules.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
people-search-guardrail.mdc |
People-search and data-broker results are leads only: never cited, never published, never stored in the knowledge base, never run on a private citizen before the inclusion gate passes. Also sets rules for scanner audio | The knowledge base, plus research, batch-run and evidence records | Inclusion and Fairness Gates |
patriot-retribution-tracker.mdc |
The retribution tracker’s data rules, including the naming rule: a person is named as responsible only when responsibility is documented. Publishing is human-gated | The tracker’s data file, schema, code and rendering scripts | Inclusion and Fairness Gates |
Scoring-instrument rules
These are covered in Scoring Instrument Rules. Each one says when its instrument must be run and what the section must contain.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
patriot-malice-evaluator.mdc |
When a Democratic Malice Assessment is required, the order of checks, the ideology-versus-malice gate and the closing legal disclaimer | Accountability profiles | Scoring Instrument Rules |
patriot-spokesperson-misinformation-assessment.mdc |
When a Spokesperson Misinformation Assessment applies, the tenure-window guard and the advocacy-versus-misinformation gate | Accountability profiles | Scoring Instrument Rules |
patriot-dangerous-rhetoric-assessment.mdc |
When a Dangerous Rhetoric Assessment applies, the public-figure-only guard and the three rhetoric axes | Accountability profiles | Scoring Instrument Rules |
patriot-legislative-impact-scoring.mdc |
The two-track Legislative Impact Assessment of one bill, order or rule, and the rule that the two tracks are never merged into one number | Any knowledge-base article | Scoring Instrument Rules |
patriot-policy-winners-losers.mdc |
The Policy Winner-Loser Map of a market, with two gates: benefit is not corruption, and no investment advice | Any knowledge-base article | Scoring Instrument Rules |
patriot-agency-damage-assessment.mdc |
The Agency Damage Assessment of one federal institution, including the coverage gate, privacy limits on the personnel ledger and the rule that stubs carry no score | Any knowledge-base article | Scoring Instrument Rules |
AI and research rules
These are covered in AI and Research Runtime Guardrails.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
patriot-investigative.mdc |
Conventions for writing investigative skills and tool articles: cross-reference by name, use the evidence tiers, never present AI analysis as documented evidence, keep personal data out | All skill files and investigative-tool articles | AI and Research Runtime Guardrails |
scraping-conduct.mdc |
Honor robots.txt and site terms unless an editor records a public-interest exception, throttle requests, never scrape behind a login, store only the fields a story needs | The code that fetches from outside services | AI and Research Runtime Guardrails |
transcription-confidentiality.mdc |
Audio from a confidential source is transcribed on one machine only and never sent to a cloud service; public-record audio may use one | The transcription tool, and source material folders | AI and Research Runtime Guardrails |
ai-generated-media-disclosure.mdc |
Every AI image carries a visible “AI-generated illustration” label and a content credential, and no AI image shows a real, identifiable person in an invented scene | Image-generation code, the paths that carry images to the site and Substack, and knowledge-base articles | AI and Research Runtime Guardrails |
Publishing and content-architecture rules
These are covered in Publish-Time Gates.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
patriot-wordpress-content-architecture.mdc |
The site holds two kinds of content that must never be confused; the publish manifest is the source of truth and is never cleared and republished | The publishing scripts, the manifest files and knowledge-base articles | Publish-Time Gates |
patriot-timeline-repository.mdc |
When a knowledge-base article changes, the matching entry in the master timeline repository is updated too | Any knowledge-base article | Publish-Time Gates |
Engineering and change-control rules
These are covered in Engineering and Change-Control Guards. They govern how people and AI tools build and maintain the platform rather than what we publish.
| Rule file | What it governs | Applies to | Family article |
|---|---|---|---|
shared-working-tree.mdc |
Several tools work in one checkout at once: never revert changes you did not make, never sweep them into your commit, never switch branches without asking | Always loaded | Engineering and Change-Control Guards |
proactive-issue-surfacing.mdc |
When work reveals a structural problem, name it, explain the cause and propose a fix instead of working around it silently | Always loaded | Engineering and Change-Control Guards |
graphify.mdc |
How to use the code knowledge graph, with accuracy cautions | Always loaded | Engineering and Change-Control Guards |
patriot-skills-routing.mdc |
Route authoring and research tasks to the right skills through the shared router instead of guessing names | Always loaded | Engineering and Change-Control Guards |
patriot-local-services.mdc |
How the locally run development services are managed | Always loaded | Engineering and Change-Control Guards |
plan-integrity.mdc |
Plans must ground every stated count in a command, trace moved files through every loader, and surface conflicting requirements instead of silently choosing | Plan documents, reorganization records and the scripts | Engineering and Change-Control Guards |
patriot-api-conventions.mdc |
Coding conventions for the public chat-service backend | The backend’s Python code | Engineering and Change-Control Guards |
patriot-ekb-theme-integration.mdc |
How our site theme overrides the knowledge-base plugin’s styling | The theme’s PHP, CSS and JavaScript | Engineering and Change-Control Guards |
patriot-design-brand-sync.mdc |
The design-system contract and the brand guide must agree in both directions | Theme design files, brand assets, icons, design docs and design scripts | Engineering and Change-Control Guards |
patriot-admin-docs-freshness.mdc |
Keep the administrator, editor and user manuals current whenever platform components change | Scripts, console code, theme, rules, skills, the knowledge base and the manuals themselves | Engineering and Change-Control Guards |
patriot-shortcode-manual-freshness.mdc |
Keep the shortcode manual current whenever a site shortcode is added, removed or changed | The WordPress shortcode and block code | Engineering and Change-Control Guards |
How the counts add up
The tables above hold 8, 2, 6, 4, 2 and 11 rules, which sum to 33. We counted the files with ls .cursor/rules/*.mdc | wc -l. The grouping is ours: a few rules touch more than one family, and we placed each where it does most of its work.
What a rule file cannot do
A written rule does not stop anything by itself. Where a rule has a matching check in code, such as the legal-disclaimer requirement, the family articles say whether the check blocks or only warns. Where a rule is policy that editors apply, such as the ban on people-search results and the scraping-conduct rule, the rule file says so itself, and so do we.
For the plain-language version of our editorial standards, see Rules & Editorial Standards — Overview. For our public commitments on AI, see How Citizen Analyst Uses AI.
Factual correction requests: If you believe information in this page is incorrect, please contact factcheck@citizenanalyst.ai with your name (optional), the specific claim, and any supporting documentation. We review all submissions and correct verified errors promptly.
