Patriot Speech Analyzer
Instructions
Analyze political speech and text to detect markers of authoritarianism using Claude, producing structured severity assessments with evidence and confidence levels.
Marker Taxonomy
Evaluate each content item against these marker categories:
| Category | Markers |
|---|---|
| Executive overreach | Unilateral action, bypassing legislature, emergency power expansion, norm violation |
| Judicial undermining | Court-packing threats, ruling defiance, judicial independence attacks |
| Press hostility | Media delegitimization, journalist targeting, information control |
| Electoral interference | Voter suppression, election result denial, process manipulation |
| Rule of law erosion | Selective prosecution, DOJ politicization, pardon abuse |
| Democratic norm violation | Political violence tolerance, opposition criminalization, loyalty demands |
| Institutional capture | Inspector general removal, civil service politicization, oversight evasion |
Severity Scoring
Score each detected marker on a three-point scale:
| Score | Meaning | Criteria |
|---|---|---|
| +1 | Active threat | Concrete action taken or specific policy announced |
| 0 | Rhetorical signal | Language patterns consistent with the marker but no concrete action |
| -1 | Counter-signal | Actions or statements that strengthen democratic norms |
Evidence Extraction
For each detected marker, extract:
- Quote: The exact text passage triggering the detection (50–200 characters)
- Context: Surrounding context explaining why this passage is significant (1–2 sentences)
- Category: Which marker category from the taxonomy
- Severity: +1, 0, or -1 with brief justification
- Figure: The political figure associated with the statement or action
Confidence Calibration
Assign confidence: High (direct quote, unambiguous match), Medium (paraphrased/secondhand, requires interpretation), or Low (indirect inference, ambiguous). Discard low-confidence detections unless corroborated by a second source in the same scan cycle.
Analysis Protocol
When sending content to Claude: provide the normalized body and metadata, reference the marker taxonomy, request structured JSON output (markers, severity, evidence, confidence), instruct “no markers detected” for neutral content, and prohibit inference beyond what the text supports.
Output includes: content_hash, figure, markers array (each with category, severity, quote, context, confidence), overall_severity (weighted average), and analysis_timestamp.
Escalation to the Dangerous Rhetoric Assessment
This analyzer’s taxonomy scores authoritarian-institutional markers (executive overreach, judicial/press/electoral/rule-of-law/norm/institutional-capture). It does not score group-directed rhetoric — dehumanization, group targeting/exclusion, or incitement-risk.
When a passage by a public figure surfaces group-directed / dehumanizing / violence-adjacent content (sub-human metaphors, ascriptive-group targeting with existence-framing, or a cluster of incitement-risk factors), escalate to the dangerous-rhetoric-assessment skill for the deep, defamation-hardened DS/TE/IR scoring. Keep institutional-erosion markers here; hand group-harm rhetoric to the DRA. Do not escalate ordinary citizens’ speech — the DRA excludes it.
Examples
Input: Presidential statement announcing executive order bypassing congressional authorization on spending. Output: Marker: executive overreach (severity +1, confidence high), quote extracted, context: “Bypasses congressional appropriations authority.” Overall severity: +1.
Input: Senator’s floor speech defending judicial independence after court ruling. Output: Marker: judicial undermining (severity -1, confidence high), quote extracted, context: “Affirms separation of powers and court authority.” Overall severity: -1.