| name | ads2-exam-workflow |
| description | Use this skill whenever the user is answering an ADS/ADS2 coding challenge, online problem-solving exam, or ICA-style RMarkdown report and needs the overall answer structure. Trigger on ADS exam, ADS2 exam, coding challenge, RMarkdown exam answer, ICA report, "how should I write the answer", or when multiple statistical tasks must be presented coherently. |
ADS2 Exam Workflow
Use this as the top-level workflow skill. ADS2 marking rewards a readable report, not only correct code.
Default Answer Shape
For each question, produce:
- Data import and inspection: show data loaded, structure, missing values, duplicates.
- Cleaning or reshaping: only what is needed; explain it.
- Useful plot or summary: boxplot, scatterplot, barplot, histogram, trend plot, paired plot.
- Method choice: name the method and justify using variable types/design.
- Hypotheses: state H0 and H1 in context.
- Assumptions: check or explain why a non-parametric/simulation alternative is used.
- R code and output: include code and key result.
- Conclusion: contextual interpretation, limitations, and next step.
Fast Report Template
### Data checking
I imported the data and checked its structure, missing values, and duplicate rows. [State findings.]
### Plot and summary
[Describe the plot and what it suggests.]
### Method
The response variable is [response]. The explanatory variable(s) are [predictors]. Because [design/variable types], I used [method].
H0: [null in context].
H1: [alternative in context].
### Assumptions
[State checks and results.]
### Result
[Report test statistic/model coefficient/p-value/CI/effect size.]
### Conclusion
At alpha = 0.05, I [reject/fail to reject] H0. This suggests [contextual conclusion]. A useful next step is [specific recommendation].
Time Priority
If time is short, do these first:
- correct data import and plot;
- correct method and hypotheses;
- p-value/result and contextual conclusion;
- one sentence on assumptions or limitations.
Next-Step Recommendations
Good ADS2 recommendations are specific:
- Use power analysis before recommending a larger sample size.
- Add a control group if the design lacks one.
- Collect key covariates such as age, sex, baseline severity, batch, site, or genotype.
- Use a paired design if biologically appropriate.
- Run Tukey HSD or another post-hoc comparison after a significant ANOVA.
- Use Fisher exact test or collect more data if chi-square expected counts are too small.
Avoid generic "more research is needed" without saying what and why.
Required Language
- Use "fail to reject H0", not "accept H0".
- Use "evidence suggests", not "prove".
- Say "association" unless the design supports causation.
- Mention units where possible.
