Archived 2026-10-01. Superseded by the three-course series — SWE 101 → SWE 102 → SWE 103. Kept for the reasoning it records; it is no longer the plan.
Status: adopted in modified form, 2026-10-01. Rather than one re-ordered 26-week scheme, the two tracks were merged into weekly lanes and split into three short courses. The reasoning below still holds.
Why change it
The scheme was re-ordered once already, around “interview relevance”. Reading it against Track A week by week shows it still works against the flagship, and it no longer fits its own time budget.
- The flagship needs things months before the scheme teaches them.
| The flagship needs | By (Track A) | The scheme teaches it in |
|---|---|---|
| Tests running in CI | week 3–4 | week 24 |
| An AI feature working | week 7–8 | weeks 20–22 |
| Evals harness | week 9–10 | week 23 |
| Hardening: errors, rate limits, cost and latency numbers | week 11 | weeks 19, 23 |
| Observability | week 12 | week 24 |
| System design reps, one per fortnight | from week 12 | weeks 5–8 (then nothing for 4 months) |
-
The time budget doesn’t fit. Track B has 2–4 hours a week. Weeks 2–4 now link about 33,000 words of lessons, each with a practice task of 1–3 hours, and the notebook method asks for a full page per topic, across about 148 topics.
-
Depth goes where the target role doesn’t ask. Four weeks of networking (link layer, subnetting, congestion control, QUIC), four of distributed systems (Paxos, Raft), and database internals down to LSM trees. The target is a TypeScript full-stack engineer who ships AI product features.
-
Weeks 2 and 4 overlap. Both send you to the same five structuring-a-backend lessons.
-
The two tracks disagree on the DSA order. The scheme follows the vault’s 15 pattern lessons. The hire track lists a different order, including greedy, tries and bit manipulation, none of which has a pattern lesson here.
The principles
- Learn it the week before you build it. For weeks 1–12, Track B follows the flagship’s milestones.
- Two core lessons a week, plus one practice task. Everything else in a week is optional, and the audit decides: K → skip, H → core only, D → core plus optional.
- Practice targets the flagship wherever a lesson allows it, so studying and building are the same hours.
- Notebook pages for core topics only. Optional topics get a line in the index, not a page.
- Depth the role rarely asks for moves to the CS spine (week 27+), not out of the vault.
- Finish in 26 weeks, not 30, leaving two buffer weeks before Track A ends at week 28. School will eat some weeks; the plan should expect that.
The proposed order
Core is what the week requires. Optional is read only if the audit says D, or if there’s time.
Weeks 1–12 · Build the flagship, learning each piece the week before
| Wk | Block | Core | Optional | Flagship (Track A) | By Sunday |
|---|---|---|---|---|---|
| 1 | Introduction & the audit | what SWE is · SDLC | roles | Pick the flagship | The audit, done honestly |
| 2 | Design principles | coupling & cohesion · SOLID | clean code | — | Run the coupling tool on the flagship; fix one finding |
| 3 | Structuring a codebase | layers · layer vs feature | hexagonal · modular monolith | w3–4: skeleton deployed, CI green | Flagship’s folder structure decided, with a boundary lint rule |
| 4 | Wiring & testing | DI & wiring · testing fundamentals | testing a backend | w3–4: CI green | Flagship has a composition root and its first real tests in CI |
| 5 | Patterns | behavioural · structural | creational | w5–6: core feature | One pattern applied in the flagship’s core feature, with a sentence on why |
| 6 | LLMs: the ground floor | how LLMs work · calling models | the role · model landscape | w5–6: core feature | Closed-book: why does a model hallucinate, mechanically? |
| 7 | Prompting, structure, retrieval | structured output · prompt engineering | RAG (core if the flagship retrieves) | w7–8: AI feature · applications start | The flagship’s AI feature returns validated, structured output |
| 8 | Evals ⭐ | evals · AI practice exercises | — | w9–10: evals next week | A golden set of 20+ cases for the flagship, written before building the harness |
| 9 | LLMs in production | reliability · cost & latency | safety · resilience patterns | w9–10: evals in CI | Evals harness running in CI, with numbers you can quote |
| 10 | Tools and agents | tools & MCP · agents | multimodal | w9–10 | Closed-book: when is an agent the wrong architecture? |
| 11 | Hardening | backend practices (validation, errors, rate limits, idempotency) · observability | the stack | w11: hardening | Flagship answers “is it up, is it fast, what does it cost per request” |
| 12 | Securing what you shipped | input & output · headers & SOP (XSS, CSRF) | prompt injection · secrets | w12: ship + write-up | Threat model: the flagship’s top five risks, and what you did about each |
Weeks 13–16 · System design, as the interview reps begin
Track A starts one system-design rep per fortnight from week 12. This block feeds them.
| Wk | Block | Core | Optional | By Sunday |
|---|---|---|---|---|
| 13 | Approach & scale | the approach · scalability | the round | URL shortener, 45 minutes, out loud |
| 14 | Availability & caching | caching · load balancing | availability · CAP | — (rep on alternate weeks) |
| 15 | Data, messaging, communication | messaging · communication | databases at scale | Notification system |
| 16 | Architectural patterns | monolith vs microservices · transactional outbox | data & integration · microservices patterns | Payment system — from your nextvibe ledger experience |
Weeks 17–25 · Depth the interviews probe
Week 26 · Review and mock loops
Unchanged from the current week 30: two system-design mocks, two timed coding mocks, the project story at three lengths, six STAR stories, and the re-audit.
Weeks 27–28 · Buffer
Nothing scheduled. This is where the weeks school took get made up. If none were lost, start the spine early.
Week 29+ · The CS spine
The current week 31+ list, plus everything marked “spine” above.
What this costs
Honestly stated, so you can disagree with it:
- Less networking and distributed-systems depth before the job hunt. Link layer, subnetting, congestion control, QUIC, Paxos and Raft come after. A role that asks about them in depth is probably not a junior TypeScript or AI role.
- System design reps start in week 13, not week 5. But under the current scheme, the design exercises finish in week 8, while Track A’s design reps start in week 12, so they’re four months stale by then.
- Fewer notebook pages. Optional topics stop getting a page.
What it means for lesson conversion
The converted Software design block (13 lessons) is still used, in weeks 2–5. The next block to convert becomes weeks 4–12: testing fundamentals, testing a backend, the AI-engineering lessons 01–14, backend practices, observability, and the two web-security lessons — about 22 lessons, not the architecture block that was next.
Open questions — your call
- Which project is the flagship, and does it already have an LLM feature? Weeks 6–10 assume the AI feature is built in weeks 7–8. If it’s my-applicant or nextvibe with AI already working, weeks 6–7 shrink and evals can come a week earlier.
- Is moving the networking and distributed-systems depth to the spine acceptable? It’s the biggest cut. If you want one of them kept, say which, and something else moves.
- Which DSA order do you want to follow? The vault’s 15 pattern lessons (the scheme’s order), or the hire track’s list? The hire track’s list includes greedy, tries and bit manipulation, which have no lessons here — choosing it means writing three.
- Is “two core lessons plus one practice task” the right weekly size? It’s sized for 2–4 hours. If you have more in practice, it can be three.
Related
- The current scheme of work
- Track A — Hire · Track B — Foundation
- Lesson quality — conversion progress
Software engineering
Infrastructure & security
Computer science
Hardware & engineering