Remediation workflow: intake → audit → fix → verify → sign-off

The repeatable pipeline used for the pilot module, ready to run per course at catalog scale. Web version of program/remediation-workflow.md. Markdown source.

The repeatable pipeline used for the pilot module in this repository, ready to run per course at catalog scale.

Diagram

Diagram source (Mermaid). If it does not render, the text description below is the same process.

flowchart TD
    A[Intake: course selected by prioritization model] --> B[Snapshot: preserve originals verbatim]
    B --> C[Audit: automated tools + scripted keyboard/reflow/contrast]
    C --> D[Issue log: WCAG criterion, severity, users affected, fix, effort]
    D --> E[Triage: severity order; faculty consulted on content meaning]
    E --> F[Fix: rebuild content accessibly - HTML over PDFs, captions from transcripts, labeled forms]
    F --> G{Verify: re-run all automated tools\n+ scripted keyboard tasks}
    G -->|failures found| E
    G -->|clean| H[Human pass: VoiceOver protocol\n+ caption accuracy check]
    H -->|issues found| E
    H -->|pass| I[Faculty sign-off: content accuracy preserved]
    I --> J[Publish + log remediation time per content type]
    J --> K[Monitor: new-term content re-check; feedback channel]
    K -->|new issues| D

(If your Markdown viewer does not render Mermaid, the text below is the same process.)

Text description

1. Intake. The course arrives from the prioritization model with its score and effort estimate. The disability-services office flags any active accommodation requests, which jump the queue.
2. Snapshot. Originals are preserved verbatim (in this repo: source/ and a renderable "before" copy). Nothing is fixed in place before evidence is captured: automated tool exports, screenshots, keyboard logs.
3. Audit. axe-core + Lighthouse + a second CLI engine (pa11y) on every page; scripted keyboard tab-through, focus-visibility check, 320 px reflow, and measured color contrast; PDF tag-tree inspection; media availability check (captions? transcript? audio description?). Output: an issue log with WCAG criterion, severity, users affected, proposed fix, and effort (S/M/L).
4. Triage. Order by severity (task-blocking first), batch by content type, and loop in the faculty author for anything that touches meaning (alt text for disciplinary images, quiz wording, plain-language edits).
5. Fix. Prefer rebuilding to patching: HTML over PDF, real text over images of text, captions derived from official transcripts, semantic tables, native form controls. Each fix is recorded against its issue ID.
6. Verify. Re-run every automated tool and scripted check; re-run the keyboard task tests; confirm zero critical/serious issues (or documented, accepted exceptions).
7. Human pass. A human runs the screen-reader protocol (evaluation/human-protocol/voiceover-protocol.md) and the caption-accuracy check (evaluation/human-protocol/caption-accuracy-check.md); ideally a paid reviewer who uses assistive technology daily reviews the result.
8. Faculty sign-off. The author confirms the remediated version still teaches the same content — accessibility work must never silently change meaning.
9. Publish + measure. Deploy, log time-per-content-type (it feeds the next prioritization round), and archive the before/after evidence.
10. Monitor. Each new term, re-check new content; keep a feedback channel for students and instructors; re-audit on any template change.

Roles at scale