Measurement log
One entry a week from week 1. Every entry records a number, the sample size it was computed at, and the implementation that produced it.
Due · Week 12
Course weight
10%
What to submit
Submit one Markdown file. Tables with explanatory notes are fine; an essay is not required.
A record of what you measured, written as you went.
Due end of week 12, and assessed as a whole rather than entry by entry. Deadlines, extensions and the use of generative tools are course-wide and live on the policies page.
What to keep
- Twelve entries, one per teaching week, dated.
- Every number carries its N, its reference set and its implementation. From week 5, record FID∞ beside finite FID as specified below; label any population baseline separately.
- Submit one Markdown file. Tables with explanatory notes are fine; an essay is not required.
When FID∞ enters the log
| Teaching week | What the entry retains |
|---|---|
| 1–4 | The session’s measurements and protocol. FID∞ is not yet required; do not invent or retrospectively add an estimate. |
| 5 | First recorded FID∞ practice: run the lecture ladder three times at fixed settings. Keep the finite N = 60 estimate, fitted intercept and population baseline separately for each run. Label the reduced-dimensional experiment. The session also assigns the sampling and inversion work. |
| 6 onward | Apply the comparison when the session calls for it, beginning with A and your own D in Week 6. Record the shared N ladder, reference protocol and repeated intercepts alongside the finite scores. |
FID∞ is a fitted estimate here, not another single-N measurement. Record all sample sizes used, draws per size, dimensions, seeds and implementation, and whether the reference is fixed or redrawn. Keep the fitted result’s variation and any checks of fit residuals or N-range sensitivity. For an analytical population baseline, write N: not applicable (population calculation).
If an activity does not compute FID∞, enter not computed and explain why. The Python workspace starter and Week 7’s three-score panel do not fit it; their population baselines are not substitutes. A failed required fit also needs its attempted settings, reason and next check. Honest reporting does not turn missing required work into completed work. Preserve earlier entries and date later corrections rather than overwriting them.
How it is read
Completeness is half the mark.
The other half is for predictions checked against measurements, with discrepancies written down rather than tidied away. “5.31, expected 5.12, cause unknown” preserves a discrepancy worth investigating. An expected result also needs an explanation; agreement alone does not validate the implementation.
At least one entry must state what you predicted before measuring and why, report what you observed, and explain the agreement or discrepancy. Expected and unexpected results are equally eligible. If the cause of a discrepancy is unknown, say so and identify a check that could help distinguish possible causes. Keep the original prediction visible rather than rewriting it after seeing the result.
A correction belongs in the week you found it rather than in the week you were wrong. A log rewritten to be right stops being a record of what you knew when, which is the only thing it is for.
Supporting experiment / Workspace
Practise the prediction before writing the log.
01 Predict
If the distributions stay fixed, what should change when you change the seed, then increase repetitions? Distinguish variation from a persistent gap.
02 Open the lab
Python workspace / sample-size experimentRun all starter cells. Change only the seed, run all again, then increase repetitions in a separate run.
03 Record
Save the source cells and record seed, sample sizes, repetitions, moment estimator and reference-drawing protocol. Keep the population values as the baseline.
04 Result
Copy the mean estimates and their discrepancies from the population values into a small comparison table. Check your prediction across runs rather than selecting one favourable seed.
05 Explain the limitation
Label this as supplementary one-dimensional practice. It does not replace the twelve weekly entries or their required bench measurements and FID∞ comparisons. Record FID∞ as not computed for this starter, not as zero or as its population distance.
Keep this supporting experiment distinct from the task above. Workspace use and reporting guidance →
Submission review
Spec
The brief poses the problem and leaves room for a response. The spec is the fixed part: what a marker checks. Some lines can be checked mechanically; the rest call for judgement.
Tick each requirement after reviewing your evidence. These ticks record your review, not a grade or confirmation that the requirement is met.
Without JavaScript, ticks last until you leave this page.
How it is marked
| Criterion | Weight |
|---|---|
| Completeness across the twelve weeks | 50% |
| Predictions checked against measurements | 50% |