File errors reported by the scanner
UDOIT, the checker built into Canvas. A missing title and missing tags on one file count as two.
Pharmacology
The one that remains is a 669-page book the scanner refuses by size, a limit of the scanner rather than a defect. Two licensed compilations were retired and replaced by a page of current public references.
Course score 94% → 98%Law & Ethics
Nothing had been done here before Saturday. The last five files needed a person's ruling on their layout; the instructor ruled all five in the review kit, and each is in the course under its original link.
Course score 99% · no file issuesWhat happened to the files
| Outcome | Pharmacology | Law & Ethics | What it means for a student |
|---|---|---|---|
| Replaced in place with a tagged, titled version | 59 | 115 | Same link, same file name; a screen reader now gets a title, headings, lists and real tables. |
| Accessible e-book filed and linked beside the original | 30 | 91 | A second door: reflowing text on a phone, in the student's own reader. |
| Rebuilt as a course page | 5 | 2 | Saved web articles become pages that read like the site did. |
| Textbook chapters cut from the publisher's own e-book | 17 | 59 | The chapter the syllabus assigns, in the publisher's accessible edition, not a scan. |
| Retired: duplicates, stale schedules, superseded outlines, licensed compilations | 31 | 11 | Fewer dead ends in the file list. Every one backed up first. |
| Whole book rebuilt by AI from the PDF | 1 · 307 pp | — | A book with no publisher e-book, read page by page and rebuilt. |
Across both courses: about 50 course-page edits, each verified by reading the page back, and a ledger of every file with a fingerprint, so a changed original is noticed and its accessible copy refreshed.
How it works, in plain terms
The scanner says what's wrong. It lists each file that lacks a title, headings, or readable text.
Find the best source. The original slide deck or Word file, the publisher's own e-book of the textbook, or the web page the reading was saved from. The instructor supplies what only they have.
AI rebuilds the document. When there is no source, AI reads the PDF page by page, recognizes headings, tables and figures, describes the images, and writes a clean, structured copy. Two independent readers check each other before a structure is trusted.
Swap it in, keep the doors open. The repaired PDF replaces the original under the same link. An e-book or a course page goes beside it. Originals are backed up, and nothing is deleted or replaced without the instructor's word.
A person reviews what the AI wasn't sure about. Flagged pages come back as a review kit the instructor can rule on themselves.
What the instructor actually did
- Found and handed over sources: 24 lecture decks, 7 handouts in Word, 7 publisher e-books, 4 saved web articles.
- Said yes to each batch of replacements and each retirement.
- Ran the scanner again and reported the numbers.
Why it is this much work
A course built on one textbook and its publisher bundle inherits the publisher's accessibility claim and is never really checked. A course built from chapters of five books, journal papers, the instructor's own handouts and decks, and articles worth reading gets a scanner score, and the score falls hardest on the instructor who put the most in. The point of Glass is to stop penalizing that instructor.
Counts come from the scanner's file reports of September 13 and 14, 2026 and the repair ledgers as of Monday evening, September 14. Every action is logged with a backup, and every original can be put back.