Weekly operating model

A practical operating contract for the 108-week apprenticeship. Exact week titles, modes, gates, and recovery windows live in the schema and interactive curriculum; this document explains how to work them without burning out or turning optional depth into a hidden prerequisite.

Core first, depth second

Every week has two lanes:

Core deliverable: the one artifact that keeps the learner moving. It is sized for the stated weekly effort and is the requirement for progress.
Optional depth work: a proof, from-scratch implementation, benchmark, failure analysis, or production trade-off note. It raises the ceiling but does not decide whether the learner may continue.

Every core artifact records four kinds of evidence:

Code or implementation (or the equivalent design artifact for a writing week)
Benchmark or result
Design document or technical explanation
Short retrospective / learning note

The curriculum tracks evidence shipped, not attendance.

Modes: match the work to the deliverable

Mode	What you ship	Example
`implement`	Working code with tests and a runnable entry point	Cache, evaluator, limiter, or router
`read_diagram`	A design doc, trade-off note, comparison, or methodology document	Training map or disaggregated-serving plan
`deploy_benchmark`	A live deployment or benchmark run	Index, vLLM, dashboard, or SGLang comparison

The mode is a scope guard. If a read_diagram week turns into 15+ hours of coding, stop and write the document. If an implement week has no running artifact after roughly 8 hours, trim scope until it runs.

Across the 108-week spine there are 69 implement, 35 read_diagram, and 4 deploy_benchmark weeks. The four GPU-dependent weeks are 75, 84, 87, and 89; they are declared in ../schema/weeks.yaml and summarized on the interactive curriculum.

Time budget

The course is sized at 10–15 hours per week. A 15-hour week is two buckets:

15 hr/week total
├── 6–8.5 hr   fixed weekly overhead
│              ├── coding drill
│              ├── system-design question or pattern
│              ├── behavioral story refinement
│              ├── technical explanation out loud
│              ├── Algorithmic Forge drill
│              └── evidence record + Sailboat retrospective
└── 7–9 hr     this week's core deliverable

Do not spend the entire budget on optional depth. If a week slips, protect the core artifact, record the constraint, and use a recovery window rather than silently expanding the next week.

Monday–Friday rhythm

Monday     theory / mathematics
Tuesday    implementation
Wednesday  systems, papers, architecture
Thursday   evaluation + interview practice
Friday     Algorithmic Forge drill + Sailboat retrospective
Weekend    recovery, revision, or deliberate depth

Interview preparation is continuous. The weekly minimum is:

one coding drill;
one system-design question or pattern;
one behavioral story refinement; and
one technical explanation out loud.

Gates, remediation, and recovery

Milestone gates are real dependencies. A failed gate pauses new gate-dependent content until a focused remediation cycle supplies the missing evidence. It does not erase prior work and does not require restarting the phase.

Recovery windows are deliberate buffer windows for catching up, rerunning a benchmark, closing an evidence gap, or taking a planned rest. The exact windows are maintained in ../schema/weeks.yaml; the current criteria are rendered at the gates page.

Generated learner-repo commands:

make week=14        # prepare one journal and its evidence scaffold
make recovery=14   # prepare a recovery plan after a week
make remediate=18  # prepare targeted remediation for a gate
make gate           # check the current blocking checkpoint
make progress       # inspect evidence/progress status

The evidence review ritual

End each week by asking:

What capability was the destination?
What helped progress (wind)?
What slowed progress (anchor)?
What could derail next week (rocks)?
Am I ahead, on track, or drifting?
What is the single next heading?

Use the Sailboat template and keep the answer short. The goal is navigation, not a performance review.

Full curriculum and modes
Milestone gates
schema/weeks.yaml — structured week source
cookiecutter/README.md — generated learner-repo commands

Back to the syllabus.