Advanced Engineering Path
The ambition is exceptional engineering judgment: explain unfamiliar systems, build correct components, investigate failures, and improve a system over time. No curriculum can establish a "top 0.1%" ranking without a defined population and independent outcome evidence. This path makes the ambition assessable through work that other engineers can inspect.
Complete the existing core first, or pass equivalent diagnostics. Use the full-book degree when sequential book study is your preferred route. This extension adds advanced coursework and a maintenance residency; it does not compress them into the existing 96-week schedule.
Status: newly authored extension, awaiting independent technical and learner review. It is not a new Reviewed semester and is not included in the core readiness matrix. The ten lesson studios contain teaching examples, assignments, expected results, and gates; proposed concentrations below are planning options, not completed courses.
The extended sequence
| Block | Prerequisites | Lessons | Planning allowance | Required evidence |
|---|---|---|---|---|
| A: Deeper theory and language tools | S1 proofs and linear algebra, S2 algorithms, S4 interpreter | 1-3 | 80-120 hours | reduction argument, typed interpreter and optimizer, numerical experiment |
| B: Correctness under failure | S5 concurrency, S6 transactions and distribution | 4, 6, 8 | 100-150 hours | finite model, crash-recovery report, minimized fuzz failure |
| C: Measurement and user-facing systems | S1 statistics, S5 networking, S7 contracts; elementary HTML/JS for lesson 7 | 5, 7 | 60-90 hours | reproducible performance experiment, accessible offline workflow |
| D: Research and maintenance residency | S8-S10 plus A-C | 9-10 | 140-220 hours, including 12-16 calendar weeks of maintenance | replication report, reviewed changes, releases, incident drill, handover |
Total: 380-580 additional focused hours. At ten hours per week, budget 38-58 weeks, then add personal breaks and remediation. These are authoring estimates, not measured completion times. Record actual hours in a pilot before treating the schedule as calibrated. The numerical lesson is an introduction to its experiment, not a substitute for a full calculus course.
Start with the lesson sequence and use its assessment contract.
What counts as an advanced result
- Correctness: state invariants, reproduce a failure, and justify the repair. Passing familiar tests alone is insufficient.
- Transfer: handle a changed requirement or workload that was not in the worked example.
- Measurement: preserve raw observations, environment, workload, uncertainty, and a result that contradicts your initial guess when one occurs.
- Judgment: compare a simple baseline and at least one rejected alternative against real constraints.
- Stewardship: maintain a system through upgrades, migrations, recovery, and another person's use.
- Communication: another person can reproduce the work and challenge the reasoning without reconstructing missing assumptions.
AI-assisted work must identify what was assisted and how it was checked. Complete the diagnostic, a transfer attempt, and a short defense without generated answers. This measures independent understanding while allowing useful tools during implementation.
Choose one depth concentration
After the shared extension, choose one direction for a further 120-240 hours. These are proposed project lanes, not promises of complete lesson coverage.
| Direction | Existing resources to reuse | Finishing project and external check |
|---|---|---|
| Languages and developer tools | SICP, Types and Programming Languages, compiler readings | interpreter plus typed IR and two justified optimizations; differential tests and code review |
| Storage and distributed infrastructure | Database Internals, DDIA, OSTEP | recoverable storage engine or replicated service; crash schedule, history checking, documented consistency |
| Performance and runtime systems | CSAPP, Computer Organization and Design, Systems Performance | optimize one real bottleneck; equal-workload comparisons and retained regression benchmark |
| Human-facing product systems | frontend library, Eloquent JavaScript, W3C tutorials | keyboard-usable offline workflow; user-task observation and conflict-resolution tests |
| AI or security | existing specializations | reuse their assessed projects; do not double-count the same unchanged artifact |
| Graphics, scientific computing, or robotics | local graphics, mathematics, and robotics collections | choose only after the numerical prerequisites; simulation with a reproducible accuracy or timing comparison |
Before starting a proposed lane, write its bounded specification, prerequisites, reading allocation, reviewer, and pass conditions. If a suitable reviewer is unavailable, mark the result self-assessed. Do not equate a published repository or merged patch with an independent assessment of every competency.
Reading without an endless queue
Use the advanced resource map. Keep one primary technical book active at a time. A new book enters the plan only when it supports a named assignment; full-book learners schedule its entire commitment explicitly. The local resource collection is a library, not a requirement to read every item.
The curriculum audit explains what was repaired and what still needs independent review.