Independent Source Agreement Scoring Prompt

This prompt is for verification engineers and AI pipeline builders who need to determine whether multiple sources independently confirm a factual claim or merely echo a single origin. The core job-to-be-done is detecting echo chambers, circular citations, and syndication chains that inflate perceived corroboration. Use this when your fact-checking pipeline has already extracted a claim and retrieved multiple sources that appear to support it—before you assign a high confidence score based on 'multiple sources' that are actually one source republished. The ideal user operates an automated verification system and needs a programmatic independence score to weight evidence correctly, route claims for human review, or prevent false corroboration from propagating into final reports.

Do not use this prompt for initial source retrieval, claim extraction, or single-source credibility assessment. It is not designed to evaluate source authority, recency, or factual accuracy in isolation—those are separate upstream prompts. This prompt assumes you already have a target claim and a list of candidate sources with extracted text. It is also not a replacement for full provenance chain reconstruction when investigating deliberate disinformation; for that, pair it with the Source Provenance Chain Reconstruction Prompt. The prompt works best when sources include publication dates, author bylines, and explicit attribution statements. Performance degrades when sources are stripped of metadata, lack timestamps, or come from platforms that systematically obscure origin (e.g., aggregator sites without clear syndication labels).

Before wiring this into production, define your independence threshold: what score triggers 'independent confirmation' versus 'echo chamber' versus 'uncertain—human review'? The prompt produces a structured score and explanation, but your application must decide the action. For high-stakes domains (legal, medical, financial), always route uncertain and echo-chamber classifications to human review with the full source set and the prompt's reasoning attached. The next section provides the copy-ready template with placeholders for your claim, sources, and output schema.

This prompt template is the core instruction set for the Independent Source Agreement Scoring workflow. It is designed to be copied directly into your prompt management system, IDE, or orchestration layer. The template uses square-bracket placeholders for all dynamic inputs, constraints, and configuration parameters. Before using it in production, you must replace each placeholder with the actual values for your verification run. The prompt instructs the model to analyze a set of sources that appear to corroborate a claim, determine whether they represent independent confirmation or echo-chamber repetition, and produce a structured independence score with a detailed rationale.

textCopy
You are an expert verification analyst. Your task is to evaluate whether multiple sources independently confirm a claim or merely echo each other.

## CLAIM TO VERIFY
[CLAIM_TEXT]

## SOURCES TO EVALUATE
[SOURCE_SET]

## EVALUATION INSTRUCTIONS
1. For each source, extract the core evidence supporting or contradicting the claim.
2. Analyze the relationship between sources. Determine if they represent:
   - **Independent Confirmation**: Sources gathered evidence separately, using different methods, data, or primary observations.
   - **Echo Chamber**: Sources reference, cite, or republish each other without adding new evidence.
   - **Syndication/Republication**: One source is a verbatim or near-verbatim copy of another (e.g., wire service, press release reprint).
   - **Common Ancestor**: Sources share a single upstream origin (e.g., same study, same press release, same witness) but do not cite each other directly.
3. Assign an **Independence Score** from 0.0 to 1.0:
   - 1.0: Fully independent, no shared origin or cross-citation.
   - 0.0: Complete echo chamber, all sources trace to a single origin.
4. Flag any **Circular Citation** patterns where Source A cites Source B, which cites Source A.
5. Flag any **Syndication False Positives** where sources appear independent but are actually republished from a common wire service or PR feed.

## CONSTRAINTS
[CONSTRAINTS]

## OUTPUT SCHEMA
You must respond with a valid JSON object conforming to this schema:
[OUTPUT_SCHEMA]

## EXAMPLES
[EXAMPLES]

## RISK LEVEL
[RISK_LEVEL]

To adapt this template, start by replacing [CLAIM_TEXT] with the exact claim under investigation. The [SOURCE_SET] placeholder should receive a structured list of sources, each with an identifier, full text or excerpt, publication metadata, and any known attribution chain. The [CONSTRAINTS] placeholder allows you to inject domain-specific rules, such as 'Treat all sources from the same parent company as potentially non-independent' or 'Flag any source published within 24 hours of the original as likely syndication.' The [OUTPUT_SCHEMA] must be a strict JSON Schema definition that includes fields for the independence score, source relationship graph, circular citation flags, syndication warnings, and a human-readable justification. The [EXAMPLES] placeholder should contain at least two few-shot examples: one showing clear independent confirmation and one showing an echo chamber with a common ancestor. The [RISK_LEVEL] parameter controls the model's caution: set to high for regulated domains where false independence could cause material harm, which will trigger more conservative scoring and explicit uncertainty language.

After adapting the template, validate it against your golden test set before deployment. Pay special attention to syndication false positives—sources from different domains that publish identical wire copy often fool naive independence checks. Also test for circular citation detection by constructing adversarial examples where sources form a citation loop. If the output schema validation fails in more than 5% of test runs, add explicit formatting examples to the [EXAMPLES] block or tighten the schema constraints in [CONSTRAINTS]. For high-stakes verification pipelines, always route independence scores below 0.3 to human review and log the full source relationship graph for auditability.

This prompt is designed to sit inside a verification pipeline after claim extraction and evidence retrieval but before final confidence scoring. It expects a claim and a list of source objects that have already been matched as potentially relevant. The harness should call this prompt once per claim, passing all candidate sources together so the model can assess cross-source relationships. Do not call it per-source-pair; the independence analysis requires the full source set to detect echo chambers and circular citation patterns.

Wrap the prompt in a function that enforces the output schema before returning results. Use a JSON Schema validator to confirm the response contains independence_score (0-1 float), echo_chamber_detected (boolean), independent_sources_count (integer), echo_chamber_sources (array of source IDs), and rationale (string). If validation fails, retry once with the error message appended to the prompt as a correction hint. If the second attempt also fails, log the failure and route the claim to a human review queue with the raw model output attached. For model choice, use a model with strong reasoning capabilities and a context window large enough to hold all source excerpts plus the prompt. GPT-4o or Claude 3.5 Sonnet are appropriate defaults. Set temperature to 0 or near-zero to maximize reproducibility of independence judgments.

Log every invocation with the claim ID, source IDs, independence score, echo chamber flag, and model version. This audit trail is essential for downstream verification reports and for debugging false positives. The most common production failure mode is syndication false positives: when two sources appear independent but are actually republishing the same wire service story. Mitigate this by including publication timestamps and syndication metadata in each source object passed to the prompt. If your source ingestion pipeline cannot reliably extract syndication indicators, add a pre-check that compares publication timestamps and author bylines before calling this prompt, and flag near-simultaneous publication with identical ledes for human review regardless of the model's independence score.