Government Source Authority Tiering Prompt

This prompt is for verification engineers who need to classify government sources into authority tiers before using them as evidence in automated fact-checking pipelines. It produces a structured tier assignment based on jurisdiction, statutory mandate, reporting chain, and publication type. Use this prompt when your verification system must distinguish a federal court ruling from a municipal press release, or a national statistical agency dataset from a legislative committee draft. The output feeds downstream evidence weighting, source selection, and audit trail generation.

The ideal user is an engineer or pipeline architect building a system that automatically ingests, classifies, and weights government documents. Required context includes the source's URL, title, publishing body, and any available metadata about its legal or administrative authority. The prompt is designed to work with incomplete metadata, but it will explicitly flag missing information that prevents a confident tier assignment. It does not evaluate the factual accuracy of the source content itself—only its institutional authority.

Do not use this prompt for non-government sources, for evaluating the factual accuracy of the source content itself, or when the source metadata is completely absent. It is not a general credibility scorer for news articles, academic papers, or corporate reports. If you need to assess a think tank's policy paper or a university study, use the Domain Authority Assessment Prompt instead. For sources where the government affiliation is ambiguous—such as state-owned enterprises or quasi-governmental organizations—the prompt will flag the ambiguity rather than force a classification.

This prompt template is designed to be pasted directly into your verification pipeline. It instructs the model to act as a source authority classifier, producing a structured tier assignment for a government source. The prompt uses square-bracket placeholders for all dynamic inputs—source metadata, the claim domain, and any specific output constraints—so you can inject fresh data on each call without modifying the core instructions. The output is a strict JSON object that your application can parse, validate, and log for audit trails.

textCopy
You are a source authority classifier for automated fact-checking pipelines. Your task is to evaluate a government source and assign it to an authority tier based on its jurisdiction, official mandate, and reporting chain. You must reason strictly from the provided metadata. Do not infer authority from the source's name alone.

## INPUT
- Source Metadata: [SOURCE_METADATA]
- Claim Domain: [CLAIM_DOMAIN]
- Output Schema: [OUTPUT_SCHEMA]
- Constraints: [CONSTRAINTS]

## TIER DEFINITIONS
- **Tier 1 (Primary Authority):** The source is the official, legally designated body with direct jurisdiction and a statutory mandate over the claim domain. It produces original data, rulings, or official records.
- **Tier 2 (Secondary Authority):** The source is an official government body with related but not direct jurisdiction, or it republishes Tier 1 data with official commentary. Includes government statistical agencies reporting on another department's domain.
- **Tier 3 (Tertiary Reference):** The source is a government-affiliated entity without a direct regulatory or data-collection mandate for the claim domain (e.g., a government-funded research institute, a legislative committee report, or a government press release summarizing external data).
- **Tier 4 (Non-Authoritative Government Source):** The source is a government website, official social media account, or public affairs office that communicates policy but does not produce or officially republish verifiable evidence for the claim domain.
- **Tier 0 (Unclassifiable):** There is insufficient metadata to determine the source's authority tier.

## CLASSIFICATION RULES
1. Always check the source's statutory mandate against the claim domain. A federal statistical agency is Tier 1 for its own statistics but may be Tier 2 or 3 for another department's regulatory domain.
2. A court is Tier 1 for its own rulings and case records within its jurisdiction.
3. A press release from an otherwise Tier 1 body is Tier 3 for factual claims unless it directly publishes new official data.
4. If the source metadata does not clearly establish the reporting chain or mandate, default to Tier 0 and explain what information is missing.
5. For claims involving multiple jurisdictions, evaluate the source against the most specific applicable jurisdiction.

## OUTPUT INSTRUCTIONS
- Return a single JSON object conforming exactly to the provided [OUTPUT_SCHEMA].
- Include a `classification_notes` field that explains your reasoning with specific reference to the source's mandate, jurisdiction, and reporting chain.
- If the tier is 0, the `classification_notes` field must list the specific missing metadata fields that prevented classification.
- Do not include any text outside the JSON object.

## CONSTRAINTS
[CONSTRAINTS]

Adaptation guidance: Replace [SOURCE_METADATA] with a structured object containing fields like name, agency_type, jurisdiction_level, statutory_mandate, parent_department, and reporting_chain. The [CLAIM_DOMAIN] should be a short string like environmental_regulation or labor_statistics. Define [OUTPUT_SCHEMA] as a JSON Schema object your application expects—at minimum, include tier (integer 0-4), confidence (float 0-1), and classification_notes (string). Use [CONSTRAINTS] to inject domain-specific rules, such as “Prefer the most recent annual report from a statistical agency over a monthly press release.” Before deploying, run this prompt against a golden dataset of 20–30 government sources with known tier assignments to calibrate your confidence thresholds and catch jurisdiction-misclassification failures.

This prompt is designed to be a deterministic classification step within a larger automated fact-checking or evidence-weighting pipeline. It should be called after a source has been identified as a government entity but before its output is used to support or refute a claim. The prompt's tiered output (e.g., Tier 1: Primary Mandated Authority, Tier 2: Official Statistical Agency, Tier 3: Sub-agency or Regional Office) is a structured signal that downstream systems can use to adjust evidence confidence scores or to route claims for human review. Do not treat this as a standalone chatbot interaction; it is a function with a defined input contract and a strict output schema.

To integrate this into an application, wrap the prompt in a service function that accepts a source_name, jurisdiction, and optional document_context. The function should first perform a rapid metadata lookup against a known registry (like a cached copy of the U.S. Government Manual or an international equivalent) to pre-populate the [KNOWN_AGENCY_METADATA] field. This grounds the model's reasoning in verified facts and prevents hallucination of agency mandates. The model call should be configured with temperature=0 and a strict JSON mode or structured output API to enforce the [OUTPUT_SCHEMA]. On the application side, implement a post-processing validator that checks for schema compliance, verifies that the tier field is one of the allowed enum values, and confirms that the justification field contains a citation to the provided metadata. If validation fails, retry once with the error message injected into the [CONSTRAINTS] field; if it fails again, log the raw output and flag the source for manual tiering.

For high-stakes verification workflows, such as those used in legal or public safety contexts, this prompt's output should never be the final word. The application harness must log the full input, output, model version, and validator results to an immutable audit trail. Use a human-in-the-loop queue for any source that cannot be confidently tiered (e.g., a confidence_score below 0.9) or that is classified as Tier 4 (Unrecognized/Unverifiable). This ensures that ambiguous government entities, such as quasi-governmental organizations or newly formed task forces, receive expert review before their authority is weighted in a fact-checking decision. The primary failure mode to monitor is the model confidently misclassifying a state-level agency as a federal one due to name similarity; your validator should cross-reference the jurisdiction field in the output against the input to catch this.