Courses / Data literacy and statistics / Data: from records to defensible conclusions / Audit a service dashboard and propose a fair trial Build and defend your work Audit a service dashboard and propose a fair trial A fictional service claims it became faster because its dashboard mean fell. Build an evidence brief using a cleaned record audit, a distribution summary and a comparison design. Preserve the raw data and distinguish a changed record from a changed service.
Choose an audience, investigate alternatives, copy evidence from your experiments, and produce an artifact. You can open this project at any time. Reports, numerical transfer checks and human review are separate outcomes.
Your project workspace Choose the kind of work you want to produce School investigation College investigation Workplace application Investigate an invented school-club queue. Create a before/after table, explain missing waits and propose a fair scheduling trial.
1. Collect and compare evidence Required experiment sources:
Include every source above and at least one pair with different inputs from the same experiment. Explain confounding changes yourself; the presence check does not judge your interpretation.
Open an experiment and change an input or capture a trial to make evidence available here.
2. Write and revise your report Decision and measurement contract State who the dashboard describes, its unit of observation, desired decision and a metric that actually measures the outcome.
Raw-to-clean audit Record every rule, retained/excluded counts, missing values and why your rule is justified. Include two captures that change only one repair.
Distribution and honest chart Compare mean and median with units. Describe tail behavior, chart baseline and how missingness could change the conclusion.
Fair experiment and uncertainty Specify allocation, primary outcome, comparison groups and stopping rule. Explain sampling assumptions separately from bias.
Recommendation and unresolved evidence Give a supported recommendation and one counterexample or sensitivity result that could reverse it.
Your artifact: code, analysis, prototype or decision brief This field stores up to 12,000 characters of inert text. Code runs only through the explicit bounded editor in the relevant lessons; project attachments never execute.
Load a text artifact Review your work against the rubric Raw observations remain distinguishable from repairs; missing values are not silently converted to observed zeros. Summaries and plots use named denominators and units; selective omission is discussed. Comparison design holds relevant conditions constant or explains remaining confounding. Uncertainty assumptions and limits are explicit; a numerical interval is not treated as protection against bias. Recommendation cites experiment captures and states what further evidence is required. 11 evidence requirements remain Capture evidence from clean an event log. Capture evidence from averages and missingness. Capture evidence from design a fair experiment. Capture evidence from quantify sampling uncertainty. Include two different input configurations from the same experiment to compare. Write your decision and measurement contract. Write your raw-to-clean audit. Write your distribution and honest chart. Write your fair experiment and uncertainty. Write your recommendation and unresolved evidence. Add your project artifact: code, analysis, prototype or decision brief. Record my report Download report and evidence Editing the report or evidence returns it to draft. Recording verifies field presence and input comparisons, not correctness of your reasoning or code.
3. Check transfer to unfamiliar cases These authored cases use different numbers or conditions from the lesson practice. They are public local checks, not secure certification. Each case permits three submissions. Opening, first responses and attempt counts survive imports; no reset creates fresh evidence.
Case data-missing-independent Open case data-missing-independent Case data-composition-independent Open case data-composition-independent Case data-precision-independent Open case data-precision-independent
Return to the course · My Work and backups
