Instructor Guide and Review Record
This module is a candidate reference implementation for future curriculum upgrades. It includes authored explanations, explicit problems and solutions, runnable learner and instructor code, assessment criteria, and remediation. It has not been independently reviewed by a subject expert or piloted with learners. Author checks do not justify upgrading the whole semester's quality status.
Deliverables and review scope
The revision repairs stopped-retry/binomial confusion, averaging of percentiles, overclaims about CLT and confidence coverage, probability-axiom wording, unqualified density claims, sample-versus-population latency language, Poisson rate units, covariance assumptions, and a misnormalized triangular example. It replaces misleading book-section links with topic assignments into book indexes.
The original runnable package covers six functions and four labs; the required inference extension adds a separate four-function package. Public checks enumerate stopped sequences and finite samples, verify score-interval endpoints through the score equation, exercise zero/all-success boundaries, and test invalid numeric inputs. Simulation illustrates uncertainty but has no brittle "random output must match exactly" pass criterion.
The case studies and mistake clinic have also been rewritten around explicit models and counterexamples. A comprehensive external mathematical review should include all concept self-checks and interpretation claims; runnable checks do not validate every explanatory claim. The required assessed path is specified in the index and competency map.
Validation commands for maintainers
From the repository root:
python labs/probability-statistics/check.py --module reference
python scripts/package-probability-lab.py --check
npm run validate:content
npm run check:links
The reference runner uses its neighboring fixture regardless of working directory. Its output path is relative to the working directory; choose one deliberately. After changing lab source, rebuild the downloadable archive with python scripts/package-probability-lab.py and run --check again. The archive includes only the eight explicitly listed teaching files, never learner submissions or generated reports. Build the foundations site and check the actual download link after content changes. Refresh teaching-unit mappings with node -r ./scripts/register-ts.cjs ./scripts/generate-reference.ts --teaching-units-only.
Human review protocol
Have a probability/statistics reviewer solve assessment items without the key and compare assumptions, results, and scoring. Use a second reviewer on a small set of anonymized sample answers to identify rubric disagreements. Revise ambiguous criteria before treating scores as reliable.
Have a programming reviewer extract the package into a fresh directory and run setup without installing dependencies. Confirm unfinished starter code fails and the instructor solution passes. Ask the reviewer to introduce a fixed-trial retry model and a with-replacement PMF; the public checks should reject both.
Learner pilot
Pilot with at least one prerequisite-ready beginner and one learner with programming experience. This is a practical first sample, not evidence of general effectiveness. Observe setup unaided, request think-aloud explanations, and record actual time by phase, stuck points, hints, first-attempt errors, and completion evidence. Recheck two competencies after3–7 days. Do not collect sensitive personal information.
| Observation | Record | Revision trigger |
|---|---|---|
| Prerequisite diagnostic | Missing skill and successful repair | A hidden prerequisite blocks the required path |
| Setup | OS, Python version, failed command | Instructions require undocumented support |
| Lesson and practice | Misread claim, example, time, hint | Repeated misunderstanding or implausible workload |
| Assessment | Per-competency score and reviewer disagreement | Rubric accepts unsupported reasoning or rejects a sound model |
| Delayed transfer | Changed assumptions noticed, solution quality | Memorized results fail to transfer |
Keep an issue log with page, evidence, proposed change, reviewer, and recheck outcome. Publish the observed workload range only after collecting it; the current hours remain estimates.
Reusable standard
Apply the same sequence to the next module: prerequisite placement, precise lesson claims, worked derivations, faded practice, executable artifacts where useful, separate solutions, per-competency assessment, remediation, and human validation. See the authoring standard for the broader rollout checklist.
Continuous models and inference extension
Review PS9–PS12 and solve the key independently. Inspect transformation endpoints, triangular support, the all-zero likelihood case, prior sensitivity, assignment multiplicities and the sharp-null justification. Pilot the additional 25–35 hours separately; no independent review or learner calibration is claimed by this implementation.