What problem does it solve? Lecture slides contain claims about model names, prices, context windows, and benchmarks that were accurate when written but quietly go stale between semesters. Manually auditing every deck for outdated statements is tedious and error-prone. ## Core Features & Use Cases - Pattern-based sweep: Flags seven categories of decay-prone claims (model names, prices, context limits, benchmarks, dated phrases, year references, links) across all .qmd decks in a course directory. - Noise reduction by design: Blanks fenced code blocks before matching (pinned versions there are correct) and skips decks the course tracker marks as taught, since those are frozen by course rule. - Use Case: Before teaching a new semester, run the sweep over your course folder to get a categorized report of every claim like "GPT-4 costs $30/M tokens" or "the latest model" that needs a human check against a primary source. ## Quick Start Run the currency sweep over my course directory and list every outdated model name, price, and dated claim that needs re-verification.
