raw-record-vs-report-consistency

name	raw-record-vs-report-consistency
display_name	原始记录与报告一致性比对
description	比对原始记录与鉴定报告的一致性，检查关键字段和数据是否匹配。

Skill: raw-record-vs-report-consistency

Description

This skill performs a cross-consistency check between Raw Records (original detection/inspection data) and the final Inspection/Appraisal Report. It verifies critical information consistency, specifically focusing on dates, instrument IDs, and record IDs.

It provides a structured JSON output highlighting discrepancies, exact matches, and conditional matches (e.g., date ranges).

Input Contract

Input Files

• Raw Records Collection: One or more files representing the original data (Images, PDFs, or OCR Text).
• Inspection/Appraisal Report: A single document representing the final report (Word, PDF, or Text).

Prerequisites

• The Raw Records and the Report must belong to the same project.
• The Report must contain at least:
  • Inspection/Appraisal Date (single date or date range).
  • Instrument/Equipment ID (often found in "Methods" or "Equipment" sections).
  • Raw Record ID or Project ID reference.

Output Structure

The output will be a JSON object containing:

• overall_status: Summary status (pass, fail, needs_review).
• project_match_summary: High-level status for specific checks.
• checks: Detailed array of individual rule validations.
• summary: A human-readable text summary of the findings.

See output_schema.json for the exact schema definition.

Execution Steps

1. Identify and Load Documents:

   • Ingest raw_records list.
   • Ingest report_document.

2. Extract Key Fields:

   • Dates: Extract inspection dates from raw records and inspection/appraisal dates (ranges supported) from the report.
   • Instrument IDs: Extract equipment identifiers from both sources.
   • Record IDs: Extract record unique identifiers from raw records and citations in the report.

3. Data Normalization:

   • Normalize dates to ISO 8601 (YYYY-MM-DD).
   • Normalize IDs (remove whitespace, standardize case).

4. Execute Validation Rules:

   • Apply R1 (Date Consistency).
   • Apply R2 (Instrument ID Consistency).
   • Apply R3 (Raw Record ID Consistency).
   • See rules.md for detailed logic.

5. Calculate Overall Status:

   • Determine overall_status based on the severity of rule results (see Logic below).

6. Generate Output:

   • Construct the JSON response with findings, evidence locations, and severity levels.

Status Logic

Condition	Overall Status
All checks are pass	pass
Only pass and conditional_pass exist	pass
Any fail with severity major	fail
Any needs_review (and no fail)	needs_review