Patriot University — Dangerous Rhetoric Assessment (DRA)
Structured, evidence-tiered classification that scores a public figure’s attributable public speech or writing for the rhetorical markers that empirically raise the risk of real-world harm to identifiable groups or persons. Produces a Dangerous Rhetoric Assessment (DRA) with a Public-Figure-Only eligibility guard, three independent 1–5 scores — Dehumanization (DS), Targeting/Exclusion (TE), Incitement-Risk (IR) — each behind its own gate, mandatory target-class specification (T1–T5), a bipartisan comparison baseline, a two-tier rigor model with an explicit claim ladder, a publication-tier fascist-rhetorical-structure gate, and layered defamation-law protections.
This skill is the rhetoric sibling of malice-evaluator and spokesperson-misinformation-assessment. Where the Democratic Malice Assessment (DMA) scores documented actions for democratic-mechanism subversion and explicitly excludes speech, and the Spokesperson Misinformation Assessment (SMA) scores documented statements for veracity, the DRA scores rhetoric as rhetoric for group-directed danger. It reuses the DMA/SMA gate-then-score architecture and defamation framework and the deviation-from-baseline pattern of spokesperson-misinformation-assessment.
Canonical rules live in docs/dangerous-rhetoric-methodology.md — this skill operationalizes them. If the skill and the charter disagree, the charter governs.
When to Use This Skill
Run the DRA when ANY of the following apply:
- Creating or updating a public-figure accountability profile (
category: trump-administration,federal-legislators,media-figures,political-operatives, etc.) that documents significant, attributable public rhetoric. - A profile quotes 1+ public statements that plausibly clear a construct gate (dehumanizing metaphor, group targeting, or a cluster of incitement-risk factors).
patriot-speech-analyzerescalates a passage as group-directed / dehumanizing / violence-adjacent (a marker class the analyzer’s institutional-erosion taxonomy does not score).- Reviewing public-figure profiles during periodic audits for DRA coverage.
Do NOT run this skill for:
- Ordinary private citizens — categorically excluded by the Eligibility Guard (Section 1). This framework is not, for now, used on citizens’ protected speech.
- Profiles still in
_drafts/status. - Subjects who clear
patriot-private-citizen-inclusion-gateonly on speech — DRA runs downstream of the inclusion gate on a non-speech anchor, never as a substitute for it. - Election-doubt propaganda → route to
election-propaganda-ombudsman. Statement veracity/misinformation → route tospokesperson-misinformation-assessment. Institutional-erosion action markers →patriot-speech-analyzer/malice-evaluator.
Constitutional and Philosophical Foundation
Legal concepts:
- New York Times v. Sullivan (1964) — public figures must prove actual malice. The Eligibility Guard preserves this framing.
- Milkovich v. Lorain Journal (1990) — the DS/TE/IR scores are evaluative judgments about the text, cited separately from the disclosed quote (protected opinion grounded in disclosed fact).
- Brandenburg v. Ohio (1969) / NAACP v. Claiborne Hardware (1982) / Snyder v. Phelps (2011) — even probability-raising, offensive, or hateful public speech is constitutionally protected unless directed to and likely to produce imminent lawless action. DRA scores rhetoric well below that bar and never asserts the speech is unprotected.
Dangerous-speech & dehumanization research:
- Benesch / Dangerous Speech Project — the five-element framework (message, audience, context, speaker, medium) and the recurring “hallmarks” (dehumanization, accusation in a mirror, threat to group purity/integrity, assertion of attack on women/girls, coded language).
- UN Rabat Plan of Action (2012) — the six-part incitement threshold (context, speaker, intent, content-and-form, extent, likelihood-including-imminence): a deliberately high bar that distinguishes protected offensive speech from incitement.
- Kteily/Bruneau blatant-dehumanization program and Haslam’s dual model — the validated category system for dehumanizing rhetoric.
- Byman (Brookings, 2021) — elite hateful rhetoric raises the risk of violence through indirect mechanisms (emboldening, target-naming/direction, amplification), not direct command; the speaker’s power and reach are multipliers.
Civic principle: vigorous, adversarial, even harsh political rhetoric is the lifeblood of democratic contest and is protected. DRA marks the boundary where rhetoric acquires the structure of group-directed dangerous speech — it does not penalize hard-edged politics.
Section 1: The Eligibility Guard (categorical — runs FIRST, before any gate)
Public-Figure-Only Guard. A statement is eligible for DRA only if the speaker was a public official or public figure speaking in a public capacity at the time of the utterance. This is the direct implementation of the product constraint that DRA is not, for now, applied to ordinary citizens’ protected speech, and it is the structural analogue of the SMA Tenure-Window Guard.
Procedure:
- Confirm the subject already qualifies under
patriot-private-citizen-inclusion-gateon a non-speech anchor. If not, stop — DRA does not run. - Classify the speaker:
public_official/public_figure/private_citizen.private_citizen→ not eligible; stop. - Confirm the utterance is public (rally, briefing, floor speech, published writing, official social account, broadcast). Private/leaked/off-record speech is out of scope for v1.
- Drop out-of-scope utterances with a one-line note and never score them.
The reconciliation clause (load-bearing). DRA does not score speech for inclusion and does not treat rhetoric as an action. It is an analytic characterization of rhetoric already attributable to a public figure whose profile stands on independent, non-speech anchors. It never asserts the speech is illegal, never uses rhetoric as an inclusion basis, and never feeds a democratic-mechanism/action score. This is what keeps DRA from colliding with the platform-wide “speech is protected / rhetoric is not action” guardrails in patriot-private-citizen-inclusion-gate, malice-evaluator, democratic-health-monitoring, and civic-world-model.
Section 2: The Three Constructs and Their Gates
The three constructs are independent. A text can score high on one and zero on the others — that is a valid, reportable result, not a lesser one. Each construct’s gate must be passed before scoring. The instrument must be able to return “no dangerous rhetoric detected”; a coder who never returns null is miscalibrated.
2.1 DS — Dehumanization
- Question: does the text render an identifiable group or person as less than fully human?
- Gate — Ordinary-Insult vs. Dehumanization: harsh insult is not dehumanization. The marker is the attribution of a non-human / sub-human category — vermin/infestation, animals, disease/poison/contamination, invasion/flood, sub-human ranking, “not people.” Safe harbor: corrupt, stupid, evil, radical, un-American, the worst. Candidate DS: vermin, animals, poisoning the blood, infestation.
- Guardrail: code the rhetorical move in the quoted text, never the speaker’s psychology. Never publish “the speaker is a dehumanizer.”
2.2 TE — Targeting / Exclusion
- Question: is an identifiable group targeted, and what treatment of it is urged or normalized?
- Prerequisite: mandatory target-class specification (Section 4) — no TE code fires until the out-group is coded T1–T5.
- Gate — Policy-Advocacy vs. Targeting: advocating a policy about a group (immigration levels, enforcement, visas) is not, by itself, targeting. The marker is (a) an ascriptive target (T1) or a political target given ascriptive/biological properties (T5), and (b) treatment urged that goes to belonging or existence (expulsion-as-cleansing, denial of membership by descent, invasion/replacement structure, purity/contamination framing). Lawful restrictive policy pursued as policy is safe harbor.
- Public construct name: “group exclusion/targeting” (conduct) — never “ethno-nationalist” (ideology diagnosis). The academic literature is backing only.
2.3 IR — Incitement-Risk
- Question: given who is speaking, to whom, in what context, and how, do the recognized risk factors for speech-catalyzed harm cluster?
- Gate — Adversarial Politics vs. Incitement-Risk: ordinary adversarial/mobilizing campaign rhetoric is safe harbor. IR fires only when risk factors cluster: high-influence speaker + named/identified target + directive or “accusation-in-a-mirror” self-defense content + broad reach + a context of pre-existing tension or prior violence.
- Guardrail: IR reports which risk factors are present, never that a speech caused or will cause violence. Carry the Brandenburg disclaimer and the RAND stochastic-causation caveat wherever IR appears.
Section 3: Scoring Protocol (per construct, 1–5)
For each eligible, in-scope utterance:
- Apply the Eligibility Guard (Section 1) — if not a public figure in public capacity, stop.
- For each construct, apply its gate (Section 2). If the gate is not passed, record 0 / not-scored for that construct and note it in “What Is NOT Scored.”
- Specify the target class T1–T5 (Section 4) before any DS/TE code above ambiguous.
- Assign the 1–5 score using the scales below, with a verbatim span + locator, a rationale, and a mandatory good-faith alternative reading (if none can be articulated, cap the score at 1).
- Run the rival-hypothesis test (Section 6) for any score ≥ 2 and document the result.
- Write the transparency block — the disclosed quote plus the evaluative determination.
3.1 Dehumanization Score (DS)
| DS | Band | Standard |
|---|---|---|
| 1 | Ambiguous | Contested; ordinary-insult reading at least as plausible |
| 2 | Implicit | Metaphor class conveyed by entailment without the bare word |
| 3 | Explicit single-instance | A clear dehumanizing category applied once |
| 4 | Explicit + biologized | Tied to hereditary/biological defect or contamination (“poisoning the blood,” “bad genes”) |
| 5 | Explicit + eliminationist entailment | Sub-human framing joined to existence-negating logic (“root out,” “eradicate”) applied to persons/populations |
3.2 Targeting / Exclusion Score (TE)
| TE | Band | Standard |
|---|---|---|
| 1 | Delegitimization | The target’s belonging/standing framed as illegitimate |
| 2 | Exclusion from public life | Target should be removed from institutions, funding, platforms, the public square |
| 3 | Coercive-treatment framing | Mass removal/denaturalization/expulsion urged or celebrated as the remedy |
| 4 | Existential/purity framing | Target’s presence/existence framed as intolerable — invasion, replacement, cleanse, purity/contamination |
| 5 | Destruction framing | Rhetoric urging or valorizing physical destruction of the group/population |
Record the object of any eliminationist verb — acts / organizations / ideology / persons-populations. Verbs on conduct or organizations weigh far less than the same verb on persons or an ascriptive identity.
3.3 Incitement-Risk Score (IR)
Score each Rabat factor 0–3 (context / speaker / intent / content-and-form / extent / likelihood-including-imminence), record all six, then map the cluster:
| IR | Band | Standard |
|---|---|---|
| 1 | Low | Isolated factor(s); low-influence speaker, or no identifiable target, or negligible reach |
| 2 | Guarded | Some clustering; identifiable target but weak directive content or limited reach |
| 3 | Elevated | High-influence speaker + identified target + broad reach; content adversarial but not directive |
| 4 | High | Above + directive or “accusation-in-a-mirror” content, in a context of pre-existing tension |
| 5 | Severe | All factors cluster: high-influence speaker, named target, directive/self-defense content, mass reach, prior-violence context |
3.4 Evidence-tier gating (mirrors DMA/SMA)
- Documented (verbatim quote from primary transcript/video/official account): eligible for scores 1–5.
- Credibly reported (reputable secondary reporting of the quote, not yet tied to primary): caps at 2.
- Paraphrase / no verbatim span: not eligible — do not score.
Coded implicit/coded-appeal claims at ≥2 require at least two of: prior in-subculture usage, contemporaneous reception evidence, structural entailment, or the speaker’s own prior explicit usage. “It sounds like historical fascist vocabulary” is never admissible on its own.
Section 4: Target-Class Specification (mandatory before any DS/TE code ≥ 2)
Specify who the out-group in the text actually is:
| Code | Class |
|---|---|
| T1 | Ascriptive — race, ethnicity, national origin, religion, descent, immigration status |
| T2 | Political/ideological — parties, movements, protest formations |
| T3 | Institutional — courts, press, universities, civil service, NGOs |
| T4 | Behavioral/criminal — persons defined by alleged conduct |
| T5 | Composite/sliding — T1 (ascriptive/biological) properties attached to a T2/T3/T4 group |
T5 is decisive. The move from political to biological framing (attributing to a political opponent the properties of a diseased/degenerate/foreign/defective population) is the mechanism by which political-opponent rhetoric acquires the structure of group-directed dangerous speech. If the text stays cleanly T2/T3, most DS/TE codes should return low, and the honest finding is “authoritarian/eliminationist toward political opponents, not group-directed dangerous speech” — a real, reportable finding.
Section 5: Comparative Baseline & Base Rates (the symmetry test — required before publishing)
Before any DRA block is published:
- Run the identical instrument against the matched, bipartisan comparison corpus in
docs/dangerous-rhetoric-comparison-baseline.md(same-genre officials across administrations of both parties; the same administration on different threats; opposed officials on their own targets). - Report indicator frequency per 1,000 words, never raw counts, with the comparison-corpus rate alongside.
- Score the subject as an ordinal band —
within-norm/elevated/outlier— never a percentile. - If the comparison corpus scores similarly, the finding is about the genre, not the speaker — drop or reframe.
The bipartisan reference class is the anti-bias and defamation defense. Deviation is a signal for scrutiny, not proof.
Section 6: Rival Hypotheses (document before any score ≥ 2)
State why the passage is not better explained by: R1 ordinary counterterrorism/law-enforcement rhetoric (severe language attaches to conduct, T4); R2 lawful policy advocacy (argues a policy without ascriptive/biological targeting or existence-framing); R3 ordinary partisan attack / banal patriotism (harsh but standard politics about a T2/T3 target); R4 religious/moral-tradition language (doctrinally sourced); R5 hyperbole without factual/eliminationist content. A documented, failing rival-hypothesis test is a prerequisite for any published score ≥ 2.
Section 7: Two-Tier Rigor & the Claim Ladder
| Tier | Requires | Licenses (max claim rung) |
|---|---|---|
| Standard (default) | LLM first pass + one human reviewer; gates + target-class + §5 symmetry + §6 rival check | Rung 1 “contains [marker] at [locator]”; Rung 2 “employs the rhetorical structure of [dehumanization / group-targeting / incitement-risk]” |
| Publication | Two blind coders (≥1 adversarial, ideologically diverse via patriot-constitutional-law-advisors); Krippendorff’s α ≥ 0.80; pre-registration; §5–§6 documented |
Rung 3 “consistent with scholarly definitions of [X]”; the Section 8 fascism gate |
**Rung 4 — “the speaker is a [fascist/racist/…]” — is not supported at any tier. If PU publishes such a claim, it must be attributed to named scholars in their own voice, with named dissenters**, never a house-voice assertion derived from DRA.
Every coded instance carries ai_assisted, ai_model, human_reviewer_id, its rigor tier, and its max supported claim rung. AI never adjudicates coder disagreements.
Section 8: The Fascist-Rhetorical-Structure Gate (publication tier only; corpus-level; never a per-speech score)
DRA carries no routine fascism score. Fascism is a family-resemblance minimum, invalid to apply to a single utterance (Eco’s own caution) and the most defamation-exposed label available. It is handled only as a gated synthesis finding over a corpus, at the publication tier.
Sources — Tier 1 required: Griffin (palingenetic ultranationalism: decadence-diagnosis and rebirth-promise) + Paxton (mobilizing passions). Tier 2 supporting: Stanley, Finchelstein, Schmitt, Klemperer. Excluded as primary: Eco Ur-Fascism (scaffolding only); “Britt’s 14 characteristics” (no standing — do not use or cite).
A “fascist rhetorical structure” finding may be reported only if all four hold: (1) both Griffin components present (a decadence indicator ≥2 and a rebirth indicator ≥2); (2) ≥4 Paxton mobilizing indicators ≥2; (3) the §5 symmetry test and §6 rivals (esp. R2 populist-radical-right, R5 Schmittian-without-fascism) documented as failing to account for the pattern; (4) α ≥ 0.80 on gate-relevant indicators.
**If the gate fails, report what the corpus is — authoritarian, illiberal, nativist, eliminationist-toward-political-opponents, populist-radical-right, Schmittian. A failed gate is a more accurate result, not a null.** Any positive finding is rendered as the named scholars’ assessment with dissenters (Rung 4 discipline), never house voice.
Section 9: Defamation-Law Protection Framework
Reuses the DMA/SMA architecture: Truth (every score rests on a verbatim, sourced quote — the literally-true factual predicate); Fair-report privilege (public statements); Opinion privilege (Milkovich — the scores are evaluations of the text, cited separately from the quote); Public-figure doctrine (Sullivan — the Eligibility Guard); good-faith uniform method (published framework, bipartisan baseline, no animus).
Absolute guardrails: never assert the speech is illegal/unprotected/”incitement” in the legal sense (Brandenburg disclaimer wherever IR appears); never assert a speech caused/will cause a specific act of violence (RAND); code the utterance, not the mind; publish the coded spans/locators alongside every characterization; route any Rung 3 / Section 8 finding through counsel + the claims-integrity-audit skill.
Required legal disclaimer (rendered in every DRA block):
> **Framework disclosure:** The Dangerous Rhetoric Assessment applies a published analytical
> framework to documented, public statements by a public figure. The Dehumanization, Targeting,
> and Incitement-Risk scores are evaluative judgments about the quoted text, not statements of
> criminal guilt, not findings that the speech is legally unprotected, and not claims that any
> speech caused a specific act of violence. Each score cites its verbatim source quote. This
> assessment is subject to update as new evidence emerges or prior evidence is corrected.
Section 10: Output Format (Profile Integration)
The DRA block is placed after the Basis for Inclusion block and after any DMA/SMA block, and before Documented Actions. It never replaces DMA (actions) or SMA (veracity); it sits alongside them as the rhetoric axis.
Format A: Profile With a DRA Finding
---
## Dangerous Rhetoric Assessment
**Rigor tier:** [Standard / Publication] · **Max supported claim rung:** [1 / 2 / 3]
**Comparative baseline:** [within-norm / elevated / outlier] versus the bipartisan same-genre
reference class (`dangerous-rhetoric-comparison-baseline`). [One sentence with the per-1k-word
basis. Carries the "signal, not proof" and selection disclosures.]
| Construct | Score | Target class | Highest-scored span |
|-----------|-------|--------------|---------------------|
| Dehumanization (DS) | [0-5] | [T1-T5] | [short label] |
| Targeting/Exclusion (TE) | [0-5] | [T1-T5] | [short label] |
| Incitement-Risk (IR) | [0-5] | [T1-T5] | [short label] |
### Scored Passage 1: [short title]
**Construct(s):** [DS / TE / IR] · **Score:** [n] · **Target class:** [T1-T5]
**Statement:** "[exact verbatim span]" ([date], [venue — rally / briefing / floor / post])
**Source:** [primary transcript/video/official account; retrieval date]
**Determination:** [Why this passes the construct gate, citing the specific marker. State the
ordinary-politics / policy-advocacy alternative that was considered and why it was rejected
(the rival-hypothesis result). For IR, list which Rabat factors cluster.]
---
### What Is NOT Scored
[Passages excluded and why: private-citizen speech (Eligibility Guard), ordinary insult /
policy advocacy / adversarial politics (safe harbor), T2/T3-only targets, paraphrase-only
material, and any construct whose gate was not passed.]
### Assessment Basis
Standard tier evaluates public, attributable rhetoric only; scores are evaluative judgments
about the quoted text. The bipartisan comparison band is ordinal, not a rate. [Notes.]
> **Framework disclosure:** The Dangerous Rhetoric Assessment applies a published analytical
> framework to documented, public statements by a public figure. The Dehumanization, Targeting,
> and Incitement-Risk scores are evaluative judgments about the quoted text, not statements of
> criminal guilt, not findings that the speech is legally unprotected, and not claims that any
> speech caused a specific act of violence. Each score cites its verbatim source quote. This
> assessment is subject to update as new evidence emerges or prior evidence is corrected.
Format B: Profile With No DRA Finding
---
## Dangerous Rhetoric Assessment
**Finding:** None
**Determination:** This public figure's documented public statements represent [vigorous
adversarial politics / lawful policy advocacy / harsh but ordinary partisan rhetoric]. Where
statements are pointed or offensive, they do not [attribute a sub-human category / target an
ascriptive group with existence-framing / cluster the incitement-risk factors] required for a
Dangerous Rhetoric Assessment score.
**Framework note:** The DRA evaluates the transition from protected adversarial rhetoric to
group-directed dangerous speech, not hard-edged or offensive politics. Vigorous, even harsh,
political speech — including sharp attacks on political opponents and institutions — is
protected and is not scored.
Section 11: Anti-Patterns and Guardrails
What This Skill Is NOT
- NOT a citizen-speech scorer — private citizens are categorically excluded.
- NOT an insult meter — harsh, offensive, or contemptuous politics about a T2/T3 target is protected and unscored.
- NOT an inclusion gate — it runs downstream of
patriot-private-citizen-inclusion-gate, never as a substitute. - NOT an action scorer — rhetoric is never treated as a democratic-mechanism action (that is DMA).
- NOT a veracity scorer (that is SMA) nor an election-propaganda scorer (that is the ombudsman).
- NOT a legal incitement finding, and NOT a causal claim that a speech produced violence.
- NOT a routine fascism scorer — fascism is a gated, corpus-level, publication-tier, scholar-attributed synthesis only.
Hard Guardrails
- Eligibility Guard runs FIRST — public figures in public capacity only.
- Gate before score — no DS/TE/IR without passing that construct’s gate.
- Target-class specification before any DS/TE ≥ 2.
- Verbatim span required — paraphrase is never scored; credibly-reported caps at 2.
- Mandatory alternative reading — none articulable → cap at 1.
- Symmetry test + per-1k-word base rate before any published characterization.
- Rival-hypothesis test documented for any score ≥ 2.
- Code the utterance, not the mind; never assert the speech is illegal or caused violence.
- Rung caps by tier — Rung 3 and the fascism gate require the publication tier (α ≥ 0.80).
patriot-sanity-checkruns last and may flag disproportionate scores.
Common Mistakes to Avoid
| Mistake | Correction |
|---|---|
| Scoring a private citizen’s speech | Categorically excluded by the Eligibility Guard |
| Scoring harsh insult as dehumanization | Only a sub-human category passes the DS gate |
| Scoring policy advocacy as targeting | Lawful policy about a group is safe harbor unless T1/T5 + existence-framing |
| Treating a T2 political attack as group-directed danger | Most T2/T3-only rhetoric returns low; report it as authoritarian-toward-opponents, not group-directed |
| Publishing “the speaker is a racist/fascist” | Rung 4 — unsupported; attribute to named scholars with dissenters if published at all |
| Asserting the speech is “incitement” | Never — carry the Brandenburg disclaimer; scores are analytic opinion below the legal bar |
| Claiming a speech caused an attack | Never — IR is risk factors present; causation is unfalsifiable (RAND) |
| Reporting raw hit counts | Use per-1k-word rates against the bipartisan baseline |
Section 12: Composition with Other Skills
DRA sits in the public-figure profile skill chain alongside DMA and SMA as the rhetoric axis:
1. patriot-private-citizen-inclusion-gate (eligibility — runs first, non-speech anchor)
2. accountability-profile-verification (evidence and sourcing)
3. { malice-evaluator (actions) | spokesperson-misinformation-assessment (statement veracity)
| dangerous-rhetoric-assessment (rhetoric — THIS SKILL) } as applicable
4. patriot-sanity-check (proportionality validation — runs last)
Interactions
- Receives from
patriot-private-citizen-inclusion-gate: subject classification (public official / public figure / private citizen) — determines eligibility and the Sullivan framing. - Receives from
accountability-profile-verification: evidence-tier classifications — determine scoring eligibility and the score cap. - Escalated from
patriot-speech-analyzer: passages flagged as group-directed / dehumanizing / violence-adjacent (a class the analyzer’s institutional-erosion taxonomy does not score). - Reads:
docs/dangerous-rhetoric-comparison-baseline.md(bipartisan reference class);docs/dangerous-rhetoric-methodology.md(canonical rules). - Composes with:
patriot-constitutional-law-advisors(ideologically-diverse adversarial-coder selection at the publication tier),source-credibility-scoring(FRS/DCS for the context packet). - Hands off: election-doubt rhetoric →
election-propaganda-ombudsman; statement veracity →spokesperson-misinformation-assessment; institutional-action markers →malice-evaluator. - Passes to
patriot-sanity-check: the DRA block for proportionality review.
Section 13: Examples
Example A — Passage scored (DS 4 / TE 4, target class T5→T1)
Subject: a senior administration official (public official, public rally — eligible).
Statement: “They’re poisoning the blood of our country — they’re not people, they’re animals, and we’re going to root them out.” ([date], campaign rally.)
Eligibility: public official, public capacity — eligible. Target class: T5 — an immigration/national-origin population (T1 ascriptive) given hereditary-contamination properties. DS gate: passes — “poisoning the blood,” “animals,” “not people” are sub-human/biological categories, not ordinary insult. DS 4 (explicit + biologized). TE gate: passes — “root them out” attaches an eliminationist verb to persons/populations with purity/contamination framing. TE 4 (existential/purity). Rival test: R1 (conduct/T4)? No — the frame is descent/biology, not alleged acts. R2 (policy)? No — this is not a claim about immigration levels. Rejected. Determination: disclosed quote + “the passage applies a hereditary-contamination and sub-human metaphor class to a national-origin population and urges its removal in purity terms; it is not a policy claim or ordinary insult.” Comparative baseline: run against the bipartisan corpus per 1k words before publishing.
Example B — Passage NOT scored (safe harbor, target class T2)
Statement: “These radical, corrupt politicians are destroying our country and we will fight like hell to take it back.”
DS gate: fails — “radical/corrupt” is ordinary political insult; no sub-human category. TE gate: fails — target is a T2 political group; no ascriptive/existence framing. IR gate: “fight like hell” is standard mobilizing campaign rhetoric absent a named target and directive/self-defense content — fails the gate. Output: No score. Noted in “What Is NOT Scored” as protected adversarial politics (report as authoritarian-toward-opponents elsewhere if warranted, not as group-directed danger).
Example C — Excluded by the Eligibility Guard
Subject: a private citizen who posted a violent meme. Eligibility: private_citizen → not eligible. Not scored; noted once. (Accountability inclusion, if any, must rest on a non-speech anchor via patriot-private-citizen-inclusion-gate, not on DRA.)