Curriculum Audit and Expansion — September 2026

The plan has a strong systems-engineering spine and a much richer assessment structure than a typical reading roadmap. Its most urgent weakness was uneven teaching depth: completed-looking navigation and readiness claims coexisted with empty worked-example slots and generic lab instructions. Increasing the number of topics without repairing that gap would make the guide larger without making it more teachable.

Review scope and confidence

The initial inventory contained 2,198 curriculum Markdown files, including 990 concept pages, across four phase apps, 56 core modules, and the two five-module specializations. The book manifest directory contained 50 entries. Review covered curriculum and source inventories, readiness and quality matrices, validation/generation scripts, roadmap and full-book sequencing, and targeted teaching-page samples across foundations, systems, data, architecture, production, and specializations.

All curriculum Markdown was scanned for structural/template indicators and size; selected pages were inspected for teaching quality. This was not a sentence-by-sentence technical review of all 990 concepts, a reading of every local book, or a learner pilot. The source map identifies which local PDFs and notes were actually inspected. Existing local edits were preserved; no claim is made that all working-tree changes originated in this review.

Findings and repairs

Priority	Finding and evidence	Change made	Remaining limitation
High	30 worked-examples.md files contained Example Slot 1: Foundation and empty answer tables	replaced all 30 with two topic-specific worked examples, transfer checks, and evidence instructions	examples still need independent technical and learner review
High	30 guided-labs.md files repeated Lab 1: Trace or Model without concrete module inputs	replaced all 30 with bounded scenarios, inputs/failures, acceptance checks, and remediation	learner implementations and elapsed times have not been piloted
Medium	55 reading-guide.md files ended with the same generic Resource Triage advice and no topic assignment	replaced all 55 with source/topic selection, active work, and a stopping contract compatible with both reading routes	each selected edition and operational tool version still needs execution-time checking
High	npm run validate:content passed before the empty teaching slots were repaired	added errors for missing or recognized scaffolded worked-example, lab, and reading-guide pages in learner-ready modules, including specialization modules through their existing validator path	signature checks cannot establish factual correctness or catch every possible template
Medium	advanced concepts existed in isolated references, build-your-own projects, and broad guides, without one bounded assessed extension	added ten linked studios, prerequisite diagnostics, 380-580 hour planning envelope, assessment contract, and maintenance residency	new extension is authored and awaiting independent review; proposed concentrations are not complete courses
High	core quality-matrix dimensions are not-started, despite learner-ready structural status	made review limits explicit and preserved readiness/quality distinctions	external review, learner pilots, and assessment calibration remain open

The 30 repaired module groups span launchpad, orientation, mathematics, algorithms except its already-authored graph worked examples, software design, systems programming, and OS/networking. The additional 25 reading-guide repairs cover Semesters 6-10. The work replaces 115 existing teaching pages and adds the advanced lesson and program documents; it does not replace the existing concept sequence or full-book path.

Two sampled navigation defects were also repaired: the module authoring template pointed to the removed docs/ location, and the simple-design concept labeled a meaningful-names chapter as a refactoring source. The latter now explicitly uses the book index because the generated library has no reliable Chapter 12 destination. This repair does not establish that every generated chapter grouping is semantically correct.

Curriculum judgment

Retain the proof-to-algorithms-to-systems-to-data progression, project throughlines, explicit security/AI specializations, and quality matrix. There was no need to add a second generic algorithms semester or another list of architecture patterns. The useful expansion connects existing foundations to harder evidence:

• computational limits and reductions, with an exact baseline and justified pruning;
• typing, compiler transformations, and semantic preservation;
• calculus/numerical experimentation beyond matrix vocabulary;
• bounded protocol models with actual counterexample traces;
• equal-workload performance experiments and overload accounting;
• storage recovery and consistency-history checks;
• accessible client state, stale responses, and offline conflict handling;
• test-oracle design, fuzzing, and minimized failures;
• research reproduction with a narrower claim than the original paper when scale differs;
• maintenance, migration, user feedback, recovery, and handover over time.

These additions support advanced ability. They do not demonstrate a population ranking such as top 0.1%, professional seniority, or accreditation. Those claims would require definitions and independent evidence outside a document collection.

What is still incomplete

1. Independent technical review. Start with concurrency, persistence, isolation, networking, security, and numerical claims. Require a named reviewer, exact scope, reproduced example, finding, and resolution. Do not promote the quality matrix simply because this audit exists.
2. Learner-pilot evidence. Run a small pilot across beginner, intermediate, and advanced material. Record prerequisite failures, time spent, help needed, abandonment, and fresh-problem retention after a delay. The schedule estimates are not calibrated learning-time data.
3. Assessment calibration. Have two reviewers independently score the same weak, adequate, and strong artifacts. Resolve rubric disagreement before treating a percentage or distinction as objective. The new rubric is an authored proposal.
4. Full specialty courses. The new client, numerical, compiler, and research studios are bounded introductions with depth paths. Complete frontend, graphics, robotics, scientific-computing, and advanced-language concentrations remain proposals. They need their own diagnostics, lesson sequences, implementations, assessment evidence, and review before readiness claims.
5. Source and route quality. The existing check:links command detects raw-book links; it does not comprehensively prove that relative targets, anchors, external sites, or the chapter behind a descriptive title are correct. Docusaurus currently warns on broken links. A semantic source-link audit remains necessary, especially for generated book chapter groupings.
6. Operational currency. Books are good foundations for stable mechanisms; cloud permissions, CI syntax, framework APIs, and service limits need version-specific official documentation. No blanket claim is made that all old snippets are current.

Next improvements, in order

Next work	Concrete deliverable	Completion evidence
Calibrate one vertical slice before more expansion	launchpad-to-one-project pilot plus one advanced studio	observed time/friction, delayed transfer, revised lessons
Review mechanisms with the largest consequence of error	concurrency, recovery, distributed-state, security review packet	reproduced failures and reviewer-signed corrections
Strengthen link checking	local route/anchor checks, generated-source mapping checks, separately scheduled external checks	seeded bad-link cases detected without confusing intentional external/reference links
Maintain a small executable example corpus	original minimal programs for high-risk lessons	pinned environment, actual expected outputs, reproducible failure/fix pairs
Add specialty courses only after choosing a goal	one concentration at a time with a finite reading commitment	complete lessons and external review rather than another topic index

Validation record

The initial content validator passed despite the documented placeholders. After repair, content validation and the raw-book-link check pass. The protocol model's code was executed: the naive handler exposes a duplicate-effect trace and the atomic model has no counterexample within its declared finite state space.

Final verification on 2026-09-16:

Check	Result	Scope
npm run validate:content	passed after the final source-link correction and mapping refresh	structural, source-backbone, readiness, and manifest rules
npm run check:links	passed	raw-book-link policy only; not a comprehensive web/link audit
Scaffold regression check	passed	temporarily injected each of the three known scaffold forms; the validator rejected all three; original files restored
Protocol model execution	passed	naive counterexample found; no counterexample in the repaired bounded model
New-page relative-file links and code fences	passed	16 new Markdown documents; all checked relative file targets exist and fences balance
Portal, foundations, systems, architecture, production, reference builds	all six passed	Docusaurus compilation and static output in each app's build/review-20260916 directory
Teaching-unit mapping refresh	completed	1,315 mappings processed using the new --teaching-units-only mode; book pages were not cleared/rebuilt
git diff --check	passed	working-tree whitespace check

Normal sandboxed builds encountered Windows EPERM errors while writing generated files. Authorized build execution with separate .docusaurus/review-20260916 cache directories resolved that environment issue. Builds emitted a dependency warning about dynamic require in vscode-languageserver-types, but completed successfully. No package upgrades or deployment were performed. The final sampled source-link correction was additionally checked through regenerated mappings and content validation.

A successful build checks rendering and compilation, not the educational claims above. The six-app checks do not certify every external URL, book grouping, command in the curriculum, or assessment outcome. The audit and new studios remain self-reviewed authoring work until the independent review and pilot gates are completed.