Stage 4 of 6 · Method · the same evidence as Week 7, one rung higher
Review like an engineer
Week 7 made the case from the learning research: retrieval beats rereading, and spacing beats massing. This page assumes that case and adds the final-exam upgrade: a ladder of review activities ordered by how much they make you generate, a practice circuit assembled from material the course already wrote for you, and the one honest schedule for a week that also owes the project a rung.
The retrieval ladder
Review activities are not equal, and the ordering principle is effortful generation: the more an activity makes you produce the answer rather than recognize it, the more it strengthens recall. The classic result is Roediger and Karpicke's testing effect (2006): students who practised retrieving a text outperformed students who restudied it, by more as the delay grew, despite feeling less confident along the way. That feeling is the trap to plan around: rereading feels smooth because recognition is easy, and the smoothness is mistaken for knowing. Climb instead:
Spend your hours high on the ladder. Rung 1 is for patching gaps the higher rungs expose, nothing more. Rung 5 is the one this course can offer that most courses cannot: your study material runs, so "do I understand QoS?" has an experiment, not just a paragraph.
Why the bench is a legitimate study tool
Rung 5 is prediction first, always: write what the firehose will show before unplugging the node, what curl will return before stopping the bridge, which way the loop will respond before the step test. Running commands without a prediction is rung 1 wearing a lab coat. A wrong prediction is the most valuable artifact a review week can produce; it is a misconception caught before the exam could catch it, and the clinic is the list of the usual ones.
The practice circuit
The course has already written your question bank; the circuit just sequences it. One lap is about ninety minutes; run laps across days rather than back to back, because the spacing is doing half the work:
Narrate the journey (rung 3): blank page, photon to pixel, then check against the story page. Ten minutes.
One equation pass (rung 3): one of the three passes on the card, rotating each lap. Fifteen minutes.
Two hubs' Check-yourself sets (rung 2, honestly): Weeks 8 to 12, two per lap, answers covered until yours is committed aloud. Thirty minutes.
One clinic pass (rung 2): read each wrong belief in Clinic 2 and catch the pull of any that still feel plausible; those get rung-5 experiments. Ten minutes.
The journal line: what the lap exposed; tomorrow's lap starts there. Five minutes.
Your own paper trail is circuit material too: the Week 7 reflection-log method means you already own a written record of what confused you once, and old confusions are the exam's favourite wells. Foundations feeling soft at any point routes to Week 7's hub, whose practice exam and clinic still stand.
The half-width week, laid out
Front-load the rung, space the laps. The shape was chosen in the Week 12 studio; this strip is just that plan with dates on it. Adjust the days to your life, keep the two invariants: project early, review spaced, and Sunday evening consolidates rather than acquires.
Checklist for this stage
Check yourself
Rereading Week 10 feels productive and the pages feel familiar. Using the ladder's logic, what is actually being measured, and what five-minute test replaces the feeling?
Familiarity measures recognition fluency, how easily the text processes on sight, which rises with every exposure whether or not you could produce any of it. The five-minute replacement is a rung-3 probe: blank page, write the PID law with every symbol and unit, the ess formula, and the two reasons the D-term is dangerous on a noisy PV; then open the page and diff. The diff is your actual state. If it is clean, move to rung 4 or 5; if not, you just found tonight's work, which the rereading was papering over.
A classmate plans one six-hour review block on Sunday, arguing total hours are what count. Make the evidence-based counter-argument, and name the extra risk specific to this particular Sunday.
The spacing effect: distributed practice reliably beats massed practice at equal total time, because each spaced retrieval happens after partial forgetting, and effortful retrieval is what strengthens the memory; six massed hours give the later hours nothing to retrieve effortfully and fatigue degrades them anyway. The testing effect compounds it: spaced laps are retrieval-heavy, a cram block drifts toward rereading. The Sunday-specific risk is the collision this course designed around: if anything slips, Sunday's buffer is already spent, and Monday carries both the exam and D3 with no slack; the strip's front-loaded rung exists precisely so Sunday has nothing load-bearing on it.
Design a rung-5 experiment that tests your model of retained messages, prediction included, and say which clinic belief it would expose.
Prediction first, written: "I will publish a retained 55 on a scratch topic, subscribe afterward from a second terminal, and I predict the late subscriber receives exactly one message, the 55, immediately; I further predict that after publishing a retained empty message, a fresh subscriber receives nothing." Run: mosquitto_pub -r -t bench/test -m 55, then subscribe; then mosquitto_pub -r -n -t bench/test, subscribe again. If your prediction expected the broker to replay several past values, the experiment just exposed the clinic's "the broker saves all messages" belief, cheaply, with the ghost-clearing move rehearsed as a bonus.