Inclusion and Fairness Gates
Last updated: November 20, 2026
The second rule family in Rules, Gates and Guardrails: How We Constrain Our AI and Ourselves answers a question that comes before sourcing: should we write about this person at all?
A profile of a private citizen can do real harm, so we decide who is eligible before we spend research on them. This page lists the tests and says where a program enforces them and where an editor decides.
Step one: classify the subject
Every profile subject is classified before anything else happens. The pre-screen record uses six classifications: public official, nominee, public figure, private citizen, organization and removed official. The rule is that when in doubt, we classify at the more protective level. Someone who attended a rally and posted about it is a private citizen unless documented evidence shows they voluntarily took a public leadership role.
Private citizens need a non-speech anchor
A private citizen may be included only if documented conduct, not speech, meets at least one of five anchors:
| Anchor | What qualifies |
|---|---|
| A | A criminal conviction, indictment or formal charge tied to the subject matter, with a court record |
| B | A documented founder, officer, director, chapter-leader or operational-coordinator role in an organization with documented harmful conduct |
| C | Documented, significant, targeted financial support that materially enabled specific harmful conduct, backed by a campaign-finance or court record |
| D | Voluntarily and publicly taking a leadership or coordination role, through press conferences, interviews or public organizing (not merely attending events or posting online) |
| E | Using a documented official, non-elected capacity, such as an election official, law enforcement officer or government contractor, to advance the harmful conduct |
These are never enough alone: political positions, affiliation, social posts, rally attendance, membership without documented leadership, or appearing in someone else’s taxonomy.
The Speech Taint Strip Test comes next. An editor removes all speech-based evidence from the profile and looks at what is left. If the remaining non-speech evidence still justifies inclusion, the anchor stands. If it does not, the profile is removed or reduced to network mapping with no individual findings. A borderline result counts as a no.
Even a private citizen who passes faces hard limits: priority capped at P1 absent a criminal conviction, strictly descriptive language, scope limited to the anchor conduct, and no personal photographs or home or workplace details. The full test is in the published skill Citizen Analyst — Private Citizen Inclusion Gate, and a plain-language explanation is in Accountability Profiles — Standards FAQ.
The pre-screen, and what the publish gate does with it
The test above is written policy. Two programs turn it into a record and a block:
- Pre-screen (
scripts/ca_prescreen.py). It runs before any research budget is spent, with no network access. It checks identity and category, and for a private citizen it requires an anchor from A to E and an evidence link. It writes aninclusionrecord into the draft with one of four results: pass, fail, excluded or policy-question. It runs first because in one 2026 bulk run, none of 20 gate failures was rescued by research. - Inclusion gate at publish (
scripts/lib/inclusion_gate.py). A document whose result is pass publishes. Any other result blocks publication, and so does a malformed record. The one exception is a recorded ruling by Peter that names the specific reason and the specific subject. Then the document publishes with a visible warning. Free-text notes never clear a block, and no ruling clearsexcluded(a removed official).
The gate runs on knowledge-base publishes, the Substack export and timeline stories. Generated tracker pages carry no front matter and are exempt. A profile with no inclusion record is not judged by the gate, so the pre-screen is what puts the record there. The gate exists because a profile once went live with a failed pre-screen: no publish path had asked. See Publish-Time Gates.
Basis for Inclusion
Every accountability profile must carry a Basis for Inclusion block after the overview and before the documented actions. It states the subject’s classification, the anchor met, what is not the basis for inclusion, and how speech in the profile is characterized. For person profiles, the draft check blocks a draft that lacks it. See Draft Checks and Human Review.
The federal-judge threshold
Independent courts are a structural defense against abuse of power, so judges get their own gate. A federal judge appears in a personal profile only on personal misconduct, in one of five forms:
- J1: ex parte contact with a party or the party’s political operation
- J2: undisclosed financial interests tied to a party before the court
- J3: gifts, travel or benefits from parties with matters before the court
- J4: a formal misconduct finding by the circuit judicial council
- J5: impeachment proceedings begun by the House
Erroneous rulings, rulings that favor the appointing president, appointment itself, case-management patterns and unpopular legal theories never clear it. Rulings that matter structurally can be covered in a Judicial Impact Analysis, which examines their effect rather than the judge. This is policy applied by editors. No program checks it.
Priority tiers have a ceiling
The priority tier (P0, P1, P2) must reflect what the cited evidence supports. P0 needs a court record or congressional finding, and P1 needs a documented pattern of conduct. make check-draft computes a ceiling from the sources a draft cites, and a hand-set tier above it is an error that blocks the draft. A tier below it is an editorial choice. Network affiliation or an outside list alone never justifies a high tier.
The people-search ban
Results from people-search sites and data brokers are leads only. We never cite them, publish anything taken only from them, or store them in the knowledge base or committed research records. We never run such a lookup on a private citizen before the inclusion gate passes. A lead is confirmed from a primary record, such as a court filing, and the primary record is what we cite.
The same rule covers police scanner audio. A human listens to it, and it is never ingested automatically. We never publish names, plates or addresses heard on air, or identify victims, minors or suspects from radio, and never reveal the location or timing of a live enforcement operation. The rule file says plainly that no code gate checks broker results or scanner content. It is policy. Our AI commitments are in How Citizen Analyst Uses AI.
The retribution tracker’s naming rule
The retribution tracker names officials as responsible for specific harms, so it has its own rule. A person’s name may appear as the actor only when responsibility is documented, meaning a primary record, court filing, signed order, on-record admission or judicial finding. Below that, we store the role only, such as “the agency head,” and never render an inferred person as the actor. The data schema enforces this: a record that names someone without documented responsibility, or an inferred party with a name, fails validation. Low-confidence rows are flagged for review, not published, and publishing is a human step after a claims-integrity review.
What a human always decides
A program does not decide who is included. Editors classify subjects, run the strip test and decide whether a judge clears the threshold. The rulings that can clear a blocked screen are Peter’s decisions, recorded as entries in a registry in our code. For the full catalog of rule files, see The Written Rules: Every Rule File and What It Governs. For how these gates fit the wider scoring and inclusion toolset, see Scoring & Inclusion Systems — Complete Inventory and Legal Framework and First Amendment Protections.
Sources
- Judicial Conduct and Disability Act, 28 U.S.C. §§ 351-364 (the basis for gate J4).
- Citizen Analyst rule files
patriot-accountability-profile-standards.mdc(May 20, 2026),people-search-guardrail.mdc(November 14, 2026) andpatriot-retribution-tracker.mdc, as read November 20, 2026. - Citizen Analyst code:
scripts/ca_prescreen.py,scripts/lib/inclusion_gate.py,scripts/lib/priority_tier.pyandscripts/lib/draft_checks.py, as read November 20, 2026.
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.
