Skip to main content

Workload and progression contract

Plan 468–676 focused hours. These are author planning estimates awaiting independent review and learner pilots, not measured completion times. Track reading, implementation, debugging and assessment separately. Extra prerequisite remediation and optional depth are additional; passive waits are not study time.

Allocation and scope​

  • Processes and scheduling: 70–100 hours. Original process APIs, scheduling policies, Linux inspection and practice 50–72; scheduling-contract extension 20–28.
  • Memory management: 75–110 hours. Original paging, allocators, mappings, CoW, replacement and NUMA work 55–82; translation-contract extension 20–28.
  • Concurrency: 98–141 hours. Original locks, semaphores, deadlock, atomics and debugging 58–83; queue-close extension 22–32; native parallel-work clinic 18–26.
  • Filesystems and I/O: 80–115 hours. Original layouts, journaling, actual filesystem experiments and I/O APIs 58–83; crash-recovery extension 22–32.
  • Networking: 85–125 hours. Original layers, addressing, TCP/UDP, HTTP/TLS, real sockets and packet analysis 63–93; framing extension 22–32.
  • Required security strand: 30–40 hours. Threat model 6–8; packet evidence 6–8; authentication failures 8–10; least privilege 4–6; recovery proof 6–8.
  • Semester integration: 30–45 hours. Adapt/package the existing service 12–18; checkpoint and cumulative review 6–9; three-hour exam and two-hour defense 5; repairs and delayed transfer 7–13.

Modules total 408–591 hours; strand and integration add 60–85. Extensions replace the former short guided labs. Original conceptual breadth, quizzes and real-system work remain in the module allocations. Reuse existing implementations and measurements: a trace can support several requirements, but its creation time counts once. Strand time covers distinct security work; integration time covers adaptation and synthesis, not rebuilding five labs.

At 10 hours/week allow 47–68 weeks; at 15, 32–46; at 20, 24–34. Breaks and further repairs add calendar time. The ten-week arc is a sequence of topics, not a completion deadline.

Entry check and order​

Before Module 1, trace a partial C write, explain descriptor ownership, identify stack/heap lifetime, and run a small program with a reproducible build. Repair weak Semester 04 gates first. Use a Linux environment for Linux-specific observations; record kernel, tool and platform versions. A Python model does not replace observations of actual processes, mappings or sockets.

Complete modules in order. Start the service skeleton after scheduling; inspect its memory behavior after virtual memory; add bounded shared state after concurrency; state the persistence contract after filesystems; connect framed requests after networking. Start threat modelling with the skeleton and revise it after each new boundary. Integrate the security strand into that same service.

Advancement to Semester 06​

All PS1–PS6, VM1–VM6, CQ1–CQ6, PC1–PC4, FS1–FS6 and NW1–NW6 must score at least 2. All five security-strand categories must also score at least 2. Keep the existing checkpoint minimum of 80%, exam minimum of 70% overall and 60% in each major section, complete project and integrity requirements. These requirements all apply together; a high average cannot compensate for a missing gate.

Use a ledger: requirement, artifact path, command or trace, prediction, result, limitation, independent explanation, score, repair, fresh retest. Score 0 for missing/incorrect, 1 for partial or prompted, 2 for independently correct with evidence, and 3 for justified transfer and explicit limitations. Below 2 means repair required — incomplete. A plan to fix a gap is not permission to advance.

Integrated defense and transfer​

In two hours, analyze a service that accepts a framed write, queues it, persists it and replies. Mark the boundaries between bytes received, request decoded, permission checked, operation queued, data durable and response observed. Predict what happens if the connection closes or the process fails at each boundary. Explain why a successful socket write alone proves neither client receipt nor durable application state.

Then change one assumption: two workers process the same request identifier concurrently. Identify what the bounded queue, redo model and TCP stream each establish, and what none establishes about deduplication or transaction isolation. Carry that unanswered database question into Semester 06 rather than inventing an exactly-once guarantee.

After at least 48 hours, repeat the weakest case with different timing and state, without solutions. Keep first-attempt errors, corrected reasoning and new evidence. Review one Semester 03 interface contract and one Semester 04 partial-I/O invariant alongside the service. Waiting time is outside the budget. A learner-ready label describes authored material; external review and timing pilots remain pending.

Independent practice route and evidence reuse​

The following entry checks and changed-case practice make the existing module work inspectable. Use them within the current original-practice, lab and retention allowances; no full-book reading or new parallel project is added. An existing artifact can satisfy both an original exercise and this route when it demonstrates the same claim. Keep one evidence index linking each required gate to the exact artifact, actual run, assistance and reviewer status. Do not waive a topic simply because a different lab passes.

Keep authored material, author execution, learner attempts, independent review and delayed transfer as distinct statuses. Retain existing construction, security, HCI, AI-use, world-class and capstone strands and all per-gate minima. Real-system and human requirements remain pending until observed; planned observations are not completed evidence. Record actual study time before revising estimates.