Changelog

October 2026 — independent-review corrections and additions

• Offline access: a learner route for when site downloads do not open, a source-only archive build for instructors, and a link check that requires the route on every page with downloads (F1).
• CI: explicit UTF-8 file I/O, independent platform results, and framework tests and notebooks as parallel jobs (F2).
• Practice checks: Windows path spellings, asymmetric error counts, missing lag values, named pipeline steps, call-notation feedback and non-numeric results, with regression tests (F3, F9, F10).
• RT-IoT transfer: validation compares the model with always-review and never-review over a wider fixed grid; prevalence and workload are reported (F4).
• Setup: one kernel name per profile, ordered notebook steps, and a WSL Ubuntu 24.04 route for framework modules on Windows (F6).
• Teaching pages: module-specific explanations, descriptive links and reasoning tasks for Modules 1–4, 7 and 9–14; a located worked research comparison; glossary, status and scope corrections (F5, F7, F8, F11–F14).
• Module 12 monitoring reports a zero-variance feature's standardized shift as unavailable, with its raw mean difference, instead of 0.
• Further improvements (stacked PRs #12–#14): tracked-file source archives; learner-first navigation; consolidated rubric; concept self-checks; core stability, tuning (with gradient boosting) and encoding exercises; complete-study exercises; evaluated retrieval, a JSON API and experiment tracking; learning paths, an IT case thread, what's new and community pages.
• Learning outcomes are unchanged. A learner pilot and human accessibility reviews are still to come; see the owner review guide.

Entries describe their historical editions. Use current verification scope and maintainer instructions for the present edition.

0.6.0 — repository audit and verification

• Add canonical module registry, architecture inventory, audit findings, and complete early-module instructor/foundation pages.
• Add instructor adoption/assessment/pacing/alignment guidance, AI collaboration, model-use summaries, citation, reporting guidance, and reusable authoring/accessibility templates.
• Exercise the earlier Module 3/4 public CLIs from another working directory, completing entry-point coverage.
• Validate and execute notebooks with real kernels; retain executed copies outside clean source.
• Fix installed-package cache placement; test isolated-wheel inference and source-archive completeness.
• Add captioned/scoped reading tables, project-prefix HTTP tests, and manual-review/release records.

0.5.0 development

• Complete Modules 6–14 using actual external records, commented notebooks/scripts, original practice/feedback, optional depth, and instructor notes.
• Add classification/clustering/fairness, TensorFlow training/reload, local LLM evaluation, actual adapter updates/replay, checked deployment/monitoring, and bounded read-only model routing.
• Preserve observed weak-model/schema failures as evidence; no synthetic dataset or mocked model is substituted.
• Separate core Windows/Linux and framework Linux CPU verification, both notebook groups, actual report assembly, archive checks, extracted CLI, and package build.
• Record rights, reviewed source/model versions, platform scopes, and pending manual accessibility checks.

0.4.0 development

• Add Module 5 with commented notebook, stage scripts, core/optional learning routes, original practice, worked feedback, and instructor alignment notes.
• Use real UCI Bike Sharing data with retained CC BY 4.0 attribution, original bytes, and a reviewed archive hash; retain a separate original synthetic smoke fixture.
• Separate chronological preparation, training-only pipelines, validation selection, frozen final evaluation, and target-free inference.
• Compare mean/persistence baselines with simple linear models; include a deliberately invalid leakage example outside selection.
• Generate text, CSV/HTML tables, optional described figures, and controlled source/data downloads.
• Execute the implementation remotely in Windows/Linux CI; local terminal, manual assistive-technology, and rendered-figure verification remain unavailable/pending.

0.3.0 development

• Add Module 4 with preprocessing rationale, chart selection, interpretation, original practice, and worked feedback.
• Add original synthetic service-day workload, provenance, contracts, and a shared exploration workflow.
• Generate four analytical figures and a misleading-axis demonstration, each paired with its actual CSV/HTML table and short/detailed descriptions.
• Add the optional visualization dependency extra, headless plotting, a guided notebook, and behavioral checks.
• Extend navigation, controlled downloads, and CI through Module 4. Manual assistive-technology testing remains pending.

0.2.0 development

• Add Modules 1 and 2 with outcomes, original practice, discussion/solo alternatives, worked feedback, and skippable extensions.
• Add standard-library-only shared functions and scripts, three curated notebook checks, and synthetic CSV/JSON fixtures with provenance.
• Order the beginner route and site navigation from Module 1 through Module 3, retaining direct module entry.
• Add boundary, file-validation, preservation, and isolated-script behavior checks.
• Continue documenting pending manual accessibility and notebook-interface testing.

0.1.0 development

• Add the first standalone data-processing module with original practice, discussion, feedback, and a rubric for adaptation.
• Add a synthetic network-event contract and reusable preparation function with traceable rejection reasons.
• Add a separated script and a guided notebook sharing implementation.
• Add a static reading site, controlled source downloads, and basic semantic/link checks.
• Add development checks and experimental execution configuration; document unverified accessibility/platform routes.

0.7.0 — mixed-background learning support

Added Python Foundation routing, readiness checks, a Python-to-ML file/table bridge, student acquisition/hosted/troubleshooting guidance, glossary and project/pilot support. All 14 modules now include guided construction, executable notebook steps, independent changes and misconception feedback. Added validation calibration tables, sample-size arithmetic, matched real-record lexical/LLM comparison and base/adapted generation checks. Added a real UCI phishing case with separate pipeline stages and a combined notebook. Private source documents remain excluded. Human learner/assistive-technology and live provider testing remain separately recorded limitations.