Source Selection Justification Prompt

This prompt is for verification engineers and pipeline builders who need to explain why one source was preferred over alternatives in an automated fact-checking workflow. Use it when your system has already scored or ranked candidate sources and must produce a traceable, human-readable justification for the selection decision. The ideal user is building a production verification pipeline where source selection decisions must be auditable—either for internal quality assurance, compliance reporting, or reviewer handoff packets. The prompt takes structured assessments of credibility, recency, and domain authority as input and produces a comparative justification that can be logged, reviewed, and included in verification reports.

This is not a source scoring prompt itself. It assumes you have already evaluated each candidate source along dimensions like credibility, recency, domain authority, and relevance to the claim. The prompt's job is to synthesize those structured assessments into a clear explanation of why one source was chosen over others. You should wire this prompt into your pipeline after source scoring and before evidence matching or report generation. The output should be stored alongside the verification decision so that downstream reviewers—whether human analysts, compliance auditors, or automated QA systems—can understand the selection logic without replaying the entire scoring process.

Do not use this prompt when you need to perform the initial source scoring, when you lack structured assessments for each candidate, or when the selection decision is trivial (e.g., only one source available). It is also inappropriate for real-time, low-latency applications where a full comparative justification adds unacceptable overhead; in those cases, log the scores and defer justification to an asynchronous post-processing step. For high-risk domains such as healthcare, legal, or financial verification, always route the generated justification through human review before it becomes part of a final report. The justification should never be presented as the sole evidence for a claim—it is metadata about your process, not a substitute for the source content itself.

This prompt template is designed to be pasted directly into your application's model call. It instructs the model to act as a verification analyst and produce a structured, comparative justification for a source selection decision. The core task is to explain why a specific source was chosen as the primary evidence for a claim over other available sources, based on authority, recency, and relevance. The output is intended for audit trails, reviewer context, or end-user transparency, not for making the selection itself.

textCopy
You are a meticulous verification analyst. Your task is to generate a clear, concise justification for why a specific source was selected as the primary evidence for a claim over other available alternatives.

# INPUTS
- **Claim:** [CLAIM_TEXT]
- **Selected Source:** [SELECTED_SOURCE_CITATION]
- **Alternative Sources Considered:** [ALTERNATIVE_SOURCES_LIST]
- **Domain Context:** [DOMAIN_OR_TOPIC]

# JUSTIFICATION CRITERIA
Base your justification on a comparative analysis of the sources across these dimensions:
1.  **Authority:** Compare the expertise, official mandate, and reputational credibility of each source for the specific domain.
2.  **Recency:** Evaluate the publication dates relative to the claim's time-sensitivity. Explain why the selected source's timeliness is sufficient or superior.
3.  **Relevance & Directness:** Assess how directly each source addresses the specific claim. Prefer sources providing primary evidence over those offering commentary or inference.

# OUTPUT FORMAT
Produce a single paragraph of 3-5 sentences. Do not use bullet points or markdown formatting in the final output. The justification must:
- Start by stating which source was selected.
- Explicitly compare the selected source against the most credible alternative(s) on at least two of the criteria above.
- Conclude with a definitive statement on why the selected source is the most reliable evidence for this specific claim.
- If no alternative sources were provided, justify the selection on its own merits against an implicit standard of high-quality evidence for the domain.

# CONSTRAINTS
- Do not hallucinate any details about the sources that are not provided in the input.
- If the provided information is insufficient to make a clear determination, state that explicitly and explain what information is missing.
- Maintain a neutral, analytical tone.

To adapt this template, replace the square-bracket placeholders with data from your verification pipeline. The [ALTERNATIVE_SOURCES_LIST] should be a formatted list of citations that were retrieved and scored but not selected. For high-stakes domains like healthcare or finance, always route the generated justification for human review before it is published or logged in an official audit trail. A common failure mode is post-hoc rationalization, where the model invents a plausible-sounding reason that doesn't reflect the actual scoring logic. To mitigate this, test the prompt by providing inputs where the 'selected' source is intentionally weaker on one criterion to see if the justification acknowledges the trade-off or ignores it.

The Source Selection Justification Prompt is not a standalone artifact; it is a decision node in a larger verification pipeline. In production, this prompt typically fires after a retrieval step has returned multiple candidate sources for a given claim. The application layer must supply the claim text, a ranked list of candidate sources with their metadata (authority scores, recency, relevance), and the output schema. The model's job is to produce a structured justification explaining which source was selected and why others were rejected. This output is then consumed downstream by evidence matching prompts or stored in an audit trail.

To integrate this prompt reliably, wrap it in a harness that validates the output against a strict JSON schema before accepting it. The schema should require fields such as selected_source_id, justification_summary, comparative_reasoning (an array of objects with source_id, rejection_reason, and comparative_factor), and confidence_level. If validation fails, implement a single retry with the validation error message injected into the prompt context. If the retry also fails, log the failure and route the decision to a human review queue with the full candidate set and raw model output attached. This prevents silent acceptance of malformed justifications in high-stakes verification workflows.

Model choice matters here. Use a model with strong reasoning capabilities and strict JSON mode support. Temperature should be set to 0 or near-zero to maximize deterministic, repeatable justifications. Log every invocation with a unique decision_id, the full prompt payload, the raw and validated outputs, and the latency. This audit trail is critical for downstream governance and for debugging post-hoc rationalization failures, where the model invents plausible-sounding but factually incorrect reasons for preferring a source. Pair this prompt with an independent evaluation step that spot-checks whether the cited rejection reasons (e.g., 'source is outdated') are factually true given the source metadata. If the eval detects a mismatch, flag the decision for review and feed the correction back as a counterexample for future prompt refinement.