Regulatory Submission Evidence Extraction Prompt Template

This prompt is designed for regulatory affairs and submission management platforms that need to extract evidence modules from source documents and rank them against specific agency submission requirements. The job-to-be-done is generating a structured submission-readiness gap report, not a free-text summary or a simple relevance score. The ideal user is a developer building a product for regulatory operations teams, where the upstream retrieval or document processing pipeline has already identified a set of candidate documents. The prompt assumes those documents are provided as input; it does not perform document retrieval itself.

Use this prompt when you have a defined submission checklist—such as the eCTD structure or an agency-specific publishing requirement—and you need to map provided documents to required modules. The prompt is built for domain-specific evidence ranking where generic relevance scoring fails because regulatory submissions require format compliance, data completeness, and alignment with strict agency checklists. The output is a gap report that identifies which evidence modules are complete, which are partial, and which are missing, along with specific reasons tied to the checklist criteria.

Do not use this prompt for open-ended document summarization, initial document retrieval, or workflows where the submission requirements are not explicitly defined. It is not a substitute for a regulatory information management system, nor does it make submission decisions. The prompt is a structured analysis tool that must be paired with human review before any regulatory submission decision. If your use case involves interpreting ambiguous guidance or making legal judgments about sufficiency, you need a human-in-the-loop review gate, not just this prompt.

This prompt template is designed to be pasted directly into your system instructions or sent as a user message. It instructs the model to act as a regulatory submission analyst, extracting evidence modules from provided source material and ranking them by submission requirement, data completeness, and format compliance. The output is a structured submission-readiness evidence gap report that maps each requirement to available evidence, flags gaps, and prioritizes remediation actions.

textCopy
You are a regulatory submission evidence analyst. Your task is to extract, map, and rank evidence modules from the provided source documents against the specified submission requirements.

## INPUT
Source Documents: [SOURCE_DOCUMENTS]
Submission Requirements Checklist: [REQUIREMENTS_CHECKLIST]
Agency: [AGENCY_NAME]
Submission Type: [SUBMISSION_TYPE]
Output Format: [OUTPUT_FORMAT]

## INSTRUCTIONS
1. Parse the source documents and identify all evidence modules (studies, reports, datasets, certificates, declarations, summaries).
2. Map each evidence module to the specific submission requirement it addresses from the checklist.
3. For each requirement, assess:
   - Evidence Completeness: Does the available evidence fully, partially, or not address the requirement?
   - Data Quality: Are there missing data fields, incomplete results, or unresolved discrepancies?
   - Format Compliance: Does the evidence conform to [AGENCY_NAME] formatting standards for [SUBMISSION_TYPE]?
   - eCTD Placement: If applicable, identify the correct eCTD section for each module.
4. Rank evidence modules by submission criticality:
   - Critical: Required for submission acceptance; gap blocks filing.
   - Major: Required for review; gap may trigger refusal-to-file or major deficiency.
   - Minor: Supporting or administrative; gap causes delay but not rejection.
5. For each gap identified, provide:
   - Gap description
   - Impact on submission readiness
   - Recommended remediation action
   - Estimated effort to resolve (Low/Medium/High)

## OUTPUT SCHEMA
Return a JSON object with this structure:
{
  "submission_id": "string",
  "agency": "string",
  "submission_type": "string",
  "readiness_summary": {
    "total_requirements": number,
    "fully_addressed": number,
    "partially_addressed": number,
    "not_addressed": number,
    "overall_readiness_score": "Ready / Conditionally Ready / Not Ready"
  },
  "evidence_modules": [
    {
      "module_id": "string",
      "requirement_ref": "string",
      "ectd_section": "string or null",
      "evidence_description": "string",
      "completeness": "Full / Partial / None",
      "data_quality_issues": ["string"],
      "format_compliant": true or false,
      "criticality": "Critical / Major / Minor",
      "source_document_ref": "string"
    }
  ],
  "gaps": [
    {
      "gap_id": "string",
      "requirement_ref": "string",
      "gap_description": "string",
      "submission_impact": "string",
      "remediation_action": "string",
      "estimated_effort": "Low / Medium / High"
    }
  ],
  "unmapped_evidence": ["string"],
  "review_notes": ["string"]
}

## CONSTRAINTS
- Only use evidence explicitly present in the source documents. Do not infer or fabricate data.
- Flag any requirement where evidence is ambiguous or requires interpretation as needing human review.
- If a source document references but does not include required data, mark it as a gap.
- Preserve exact section references and page numbers where available.
- For [AGENCY_NAME] submissions, apply the specific formatting rules for [SUBMISSION_TYPE].

## RISK LEVEL
High. Regulatory submissions have legal and commercial consequences. All outputs must be reviewed by qualified regulatory affairs personnel before filing decisions are made.

To adapt this template, replace each square-bracket placeholder with your actual data. [SOURCE_DOCUMENTS] should contain the full text or structured extracts from your submission dossier. [REQUIREMENTS_CHECKLIST] should be the specific agency checklist or guidance document for your submission type. [AGENCY_NAME] and [SUBMISSION_TYPE] drive format compliance checks—for example, 'FDA' and 'NDA' or 'EMA' and 'MAA'. [OUTPUT_FORMAT] can be set to 'eCTD' or another agency-specified structure. For production use, consider splitting long source documents into manageable chunks and running this prompt iteratively per module, then aggregating results in your application layer.

Before deploying this prompt, build validation checks that verify the output JSON conforms to the schema, that all requirement references in the checklist appear in the output, and that criticality assignments are consistent. Implement a human-review gate for any submission where the readiness score is 'Conditionally Ready' or 'Not Ready'. Log every extraction run with the prompt version, source document hashes, and reviewer sign-off for audit trail purposes. Common failure modes include the model hallucinating evidence for gaps it cannot fill, misclassifying criticality when agency guidance is ambiguous, and failing to detect format non-compliance in complex tables or figures—plan your eval suite to catch these specifically.

This prompt is designed to be a deterministic module within a larger regulatory submission management platform, not a standalone chat interface. The primary integration point is a backend service that orchestrates the retrieval of submission dossiers, the extraction of evidence against a specific agency checklist (e.g., FDA, EMA), and the generation of a structured gap report. The application layer must own the checklist schema, the eCTD structure validation, and the final submission readiness score; the LLM's role is strictly to map unstructured document text to the structured evidence requirements defined by the application.

The implementation harness should follow a strict Retrieve-Then-Extract pattern. First, use a document parser to extract text from source PDFs, Word files, and XML eCTD backbones, preserving section hierarchy and table structures. Second, chunk the extracted text by eCTD module (M1-M5) and section. Third, for each submission requirement in the agency checklist, retrieve the top-k most relevant chunks using a hybrid search (keyword + dense vector) over the chunked dossier. Fourth, assemble the prompt with the [REQUIREMENT], [RETRIEVED_EVIDENCE_CHUNKS], and the [AGENCY_CHECKLIST_SECTION]. The model's output must be a strict JSON object conforming to the [OUTPUT_SCHEMA], which includes fields for requirement_id, evidence_status (FOUND, PARTIAL, MISSING), extracted_evidence_summary, source_locations (document name, section, page), and a compliance_gap boolean. The application must validate this JSON against the schema immediately. If validation fails, implement a retry loop (max 2 retries) by feeding the validation error back into the prompt's [PREVIOUS_ERRORS] placeholder.

For high-risk submissions, a human-in-the-loop (HITL) review gate is mandatory. After the LLM generates the evidence gap report, the application must not automatically finalize it. Instead, route all items flagged as compliance_gap: true or evidence_status: PARTIAL to a review queue. The UI should display the extracted evidence summary side-by-side with the source document snippet, allowing a regulatory affairs specialist to confirm, reject, or amend the LLM's finding. The final, human-verified report is the system of record. Log every prompt version, the full payload, the raw model output, the validation result, and the human review action to an immutable audit trail for internal governance and potential agency inspection. Avoid using the LLM to make final compliance judgments; its output is a structured draft for expert review.