LoreLint

Inspiration

I maintain a large personal worldbuilding document that has grown to roughly 793 pages. Updating a document of that size can introduce small mechanical mistakes that are hard to notice by reading: a character may disappear, an ID may be duplicated or changed, a JSON block may break, or an important setting value may silently vanish.

I wanted a tool that checks these predictable accidents without asking an AI to decide what is canon.

What it does

LoreLint is a local, read-only web app for comparing an older DOCX file with an updated DOCX file.

It detects:

Missing and newly added characters
Duplicate or changed IDs
Name and ID mismatches
Invalid JSON blocks
Missing required fields
Values that disappeared during an update
Broken heading hierarchy
Possible faction moves
Exact duplicate character blocks

Each finding includes useful context such as the character name, ID, faction, source file, and paragraph number. Results can be exported as CSV or a human-readable HTML audit report.

How I built it

I used Codex with GPT-5.6 to plan the product, implement the Python and Streamlit application, generate fictional DOCX test fixtures, write pytest tests, diagnose document-structure variations locally, and verify the Windows workflow.

The app deliberately uses deterministic checks for core validation. It does not use an AI model to decide creative truth or modify source material.