Document Consistency Checker

The Problem of Document Inconsistency

In technical reports, project documents, or academic papers, a subtle yet critical challenge often arises: internal inconsistencies. These go beyond mere typos, encompassing deep contradictions in data, logic, and standards. For instance, the same engineering quantity might appear with different values in various sections, or the sum of itemized investments might not match the grand total. Such errors directly undermine a report's professionalism and credibility. Manually proofreading for these issues is time-consuming and error-prone, especially in lengthy documents or those referencing multiple interrelated standards.

How the Consistency Checker Works

This Document Consistency Checker automates this process through a structured workflow, primarily aimed at engineers and technical writers.

1. Intelligent Content Extraction: The skill parses multiple file formats (e.g., .docx, .pdf, .xlsx) using libraries like python-docx and pymupdf to extract raw text, paragraphs, and tables for analysis.
2. Autonomous Deep Analysis: The core intelligence lies in the model's ability to read the extracted content and autonomously detect five categories of inconsistencies:
   • Category A: Data Contradictions. Identifies mismatched values for the same metric across sections and errors in summation checks (e.g., verifying if itemized funds sum to the total investment).
   • Category B: Logical Anomalies. Catches timeline errors (e.g., a completion date exceeding the project duration) and mismatches between planning goals and construction content.
   • Category C: Numbering & Structural Issues. Checks for discrepancies between table of contents and body headings, and skipped section numbers.
   • Category D: Terminology & Formatting Issues. Flags mixed usage of terms (e.g., "small reservoir" vs. "small-scale reservoir") and inconsistent units (e.g., "ten thousand yuan" vs. "yuan").
   • Category E: Standard Citation Issues. Utilizes a built-in static library (standards_library.json, containing 141 water conservancy standards) to verify if cited standards are current, if their names match the numbers, and if formatting (like book title marks 《》) is correct. For standards not in the library, it attempts web verification.
3. Intelligent Grading & Annotation: All issues are strictly graded:
   • P0 (Critical): Must be fixed, such as direct data contradictions or citing an obsolete standard with a clear successor.
   • P1 (To Verify): Should be verified, like numbering errors or unlisted standards where the latest version can be inferred.
   • P2 (Suggested): Can be optionally fixed, including minor terminology mix-ups or formatting inconsistencies.
4. Actionable Output Generation: The skill produces two key files: a structured Proofreading Report (.md) summarizing all issues and suggestions, and an Annotated Source File (_annotated.docx). The latter uses yellow highlighting on problems in the original text, paired with orange-highlighted explanatory notes at the end of paragraphs, facilitating direct location and correction within Microsoft Word.

Usage Boundaries & Key Considerations

For optimal results, please note:
- Context Dependency: The skill requires reading the entire document to assess consistency. It is strongly recommended to save specific sections as new documents before submission if only partial checking is needed; otherwise, the effectiveness may be reduced.
- Standard Library Updates: The built-in standards_library.json needs manual updates (recommended quarterly) to include newly published standards. For unlisted standards, the skill will attempt web-based verification.
- Output Environment: The generated annotated file (_annotated.docx) should be opened with Microsoft Office Word, as rendering may differ in WPS.
- Automation Limits: The skill relies on model-based reading and logical rules; it does not replace expert review for domain-specific accuracy. Issues marked "To Verify" still require author confirmation based on professional judgment.