Historical baseline review: edition 0.6.0
Archived maintainer evidence. Statements about branches, available tools and test counts describe the historical audit; use current status and commit-specific CI for the present edition.
Review date: October 25, 2026. Scope: all 159 tracked source files at baseline commit 0b3dbdee92dff0977c68d069cf291cbf87a66c43, plus the changes in this audit.
Evidence and scope
All baseline text/source files were read through GitHub. The local terminal environment was unavailable; execution is performed by the commit-specific GitHub Actions workflow. Earlier private planning attachments could not be independently reread. This is a comprehensive repository and visible-requirements review, not a claim that an inaccessible original design was fully reconciled.
See the completed baseline integration review for the current commit and CI evidence. The baseline complete run was 40154906162. New checks require the audit commit's own successful run; earlier results do not prove new behavior.
Findings and changes
| Requirement or gap | Baseline finding | Audit action and evidence |
|---|---|---|
| Transferable modular architecture | No published architecture inventory | REPOSITORY_STRUCTURE.md documents source, generated artifacts, profiles, and boundaries |
| Stable module inventory | Paths repeated in independent scripts | curriculum/modules.json feeds navigation, kernel selection, downloads, and integrity checks |
| Consistent teaching packages | M1 missing foundation; M1–4 missing instructor pages | Added five substantive pages; all 14 now require the same five-page package |
| Independently adaptable teaching | Global assessment/adoption/pacing guidance incomplete | Added assessment criteria, Canvas linking, teaching plan, alignment status, and AI collaboration |
| Course/ABET claims | Private original crosswalk not available in repo | Preserved existing references; made unverifiable authority and advisory labels explicit |
| Reusable authoring | Policy required a template that was absent | Added lesson, practice, case, instructor, model/data-card, AI-log, accessibility-review templates |
| Notebook verification | Python cells ran through exec, not a Jupyter kernel | nbformat validation and fresh nbclient kernels retain executed copies for every module |
| Installed-package use | Wheel built but not installed and run | Added isolated target-install inference/schema checks using real cached rows |
| Cache portability | Cache path assumed a source checkout | Added explicit runtime override and user-cache fallback outside installed package files |
| Reading accessibility | Stale policy; Markdown tables lacked caption/scope/overflow support | Corrected status, added named table regions/captions/scoped headers and structural checks |
| Manual accessibility | No completed task-based evidence | Added honest NOT RUN log and reproducible task record; no false pass or conformance claim |
| Source distribution | Archive check proved included bytes, not completeness | Compare archive membership against all tracked files, then byte-match and run extracted CLIs |
| Hosting architecture | Relative paths checked only on disk | Added actual HTTP HEAD tests under /ml_ai_it/ without deploying |
| Responsible reuse | Citation, model-use summary, reporting templates absent | Added citation metadata, model limitations, PR/accessibility/bug reporting guidance |
| Release readiness | No consolidated release decision record | Added checklist separating automated gates from owner/manual decisions |
Completed first audit run
Run 40157447112 passed all four jobs for commit a208c3506c4a2711a6802fcd4da3824ba9bc8b27. Core Windows/Linux each discovered 64 behavior tests, passed 55, and intentionally skipped nine framework tests. Linux CPU separately passed all nine framework tests. All 14 notebooks/90 teaching cells executed in real kernels.
The build produced 105 reading pages and 11 worked reports; all 116 HTML pages passed the defined structural/link checks. All 186 tested HTTP routes under /ml_ai_it/ responded successfully. The 193-file source archive exactly matched tracked source and bytes; external cache digests/attribution, extracted CLIs, wheel build, and isolated installed-wheel inference passed.
A follow-up adds two direct CLI integration tests for the earlier Module 3/4 scripts, updates the setup/status instructions, and makes notebook validation code easier to maintain. The final commit must pass its own workflow with 194 tracked files and 66 discovered behavior tests; consult the draft review's current commit checks for that evidence. These counts do not turn pending human checks into passes.
Verification expectations
The workflow runs 66 discovered behavioral tests in core environments: 57 core checks and nine intentionally skipped framework checks. A separate Linux CPU job runs those nine framework checks, including actual training, generation, adapters, replay, app forms, and server health. Report each scope separately; skipped tests are not passes.
All 14 notebooks contain 90 curriculum code cells. Eight core notebooks contain 48; six framework notebooks contain 42. The temporary validation cell is not counted or retained as tutorial content. Each notebook runs in a fresh kernel from the checkout root. Executed copies preserve outputs; source copies remain clean.
Structure verification requires every module's five pages, notebook, and script; parses all tracked Python; and checks source-publication coverage. Build verification checks generated reading/report HTML, local fragments and links, table semantics, project-prefix HTTP routes, all source archive members, four external cache archives, extracted source workflows, a wheel build, and actual installed-wheel inference.
Items still needing human or owner evidence
- Screen-reader, keyboard, zoom/reflow, contrast, rendered chart/notebook/app presentation, and the complete learner journey: NOT RUN in this audit. Use the manual review log; successful code execution cannot substitute.
- At the baseline audit, a dedicated accessibility contact was not recorded. The current reporting guide supplies the maintained accessibility route; conduct concerns follow the current conduct policy.
- Exact original outcome/course crosswalk and adopted program criteria: owner review required. No official ABET attainment is certified.
- Learner piloting and pacing validation: pending.
- Public release, visibility, semester edition, and Pages deployment: separate owner actions. At that audit, the material was on a development branch; subsequent integrations are recorded in the repository history.
- Complete transitive dependency locks and independent macOS/hosted/framework-Windows support: not supplied or claimed. Tested constraints and pinned model/dataset identities define the exercised scope.
No additional Canvas assessments were invented. Existing companion practice, solo discussion, public feedback, and optional extensions support independent use without altering master assignments. Empty production, slide, and video folders were not added to suggest uncreated assets are complete.
How to reproduce
Use development instructions, the workflow at the audited commit, and the release checklist. The local execution outage means the final CI job logs, rather than an unavailable local terminal, are the authoritative runtime evidence for this review.
Subsequent mixed-background enhancement
The supplied private syllabus/module/project documents were reread locally on October 25, 2026. That resolves the earlier inability to access those sources for this enhancement's scope; institutional approval and unseen current Canvas policies remain separate. Edition 0.7.0 adds the student routes, construction and engineering improvements. Earlier test/page/file counts above belong to their historical commits and are not the new release inventory. Private source documents remain excluded from Git and downloads.