survey-data-analysis
/survey-data-analysis
Tworzy kompletną analizę danych z ankiet w R Quarto: czyszczenie danych, kompozyty, testy t, ANOVA, regresje i mediację.
- Autor
- simonhdang ↗
- Repozytorium
- personal-research-agents ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 08.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 08.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Tworzy kompletną analizę danych z ankiet w R Quarto: czyszczenie danych, kompozyty, testy t, ANOVA, regresje i mediację.
Oryginalny opis (EN)
Generate a complete behavioral-science survey analysis in R QMD (Quarto) format. Use when the user wants to analyze a survey data export (Qualtrics, Prolific-linked, or similar), write a data-cleaning script, build composites, run t-tests / ANOVAs / regressions / mediation / moderation, or produce a structured, reproducible QMD from experimental or correlational survey data.
Jak działa
- Hard rule — never read participant ID columns (IRB)
- STEP 0 — Gather context and files before writing a single line of code
- STEP 1 — QMD Structure
- STEP 2 — Boilerplate to always include
- STEP 3 — Measures section
- STEP 4 — Main Analysis
- STEP 5 — Other Analyses
- STEP 6 — Output conventions
- STEP 7 — Render it, then close the loop
- Multi-survey / linked designs
Podgląd SKILL.md
---
name: survey-data-analysis
description: >
Generate a complete behavioral-science survey analysis in R QMD (Quarto)
format. Use when the user wants to analyze a survey data export (Qualtrics,
Prolific-linked, or similar), write a data-cleaning script, build
composites, run t-tests / ANOVAs / regressions / mediation / moderation,
or produce a structured, reproducible QMD from experimental or
correlational survey data.
---
# Survey Data → QMD Analysis Skill
You are helping a behavioral scientist (psychology, organizational behavior,
marketing, or adjacent field) turn a raw survey data export into a fully
structured, reproducible R QMD analysis file. Ask enough up front to match
their study design and their lab's conventions rather than assuming your own
defaults — every hardcoded value below (column names, seeds, filters) is a
placeholder to be replaced by what the user tells you in STEP 0.
---
## Hard rule — never read participant ID columns (IRB)
Never open, print, or otherwise inspect the **values** of any
participant-identifying column in a CSV — e.g. `PROLIFIC_PID`,
`participantId`, `ResponseId`, `MTurkID`, `IPAddress`, email, or name
columns. This is an IRB data-handling requirement: an AI tool should never
actually read identifiable participant data.
Using those columns *by name* in code is fine and necessary — filtering,
deduplicating, or joining on an ID column (e.g.
`distinct(PROLIFIC_PID, .keep_all = TRUE)`). What's off-limits is
inspecting their contents: no `View()`, `print()`, `head()`, or other way of
surfacing actual ID/IP values. When skimming a CSV's structure, look at
column names only, never at row values in ID columns.
---
## STEP 0 — Gather context and files before writing a single line of code
### A. Files to gather
1. **The CSV data file.** Needed to see column names and write the cleaning
code — see the hard rule above on never reading ID column values.
2. **The survey export**, for exact question wording, item order, and scale
anchors. Ask for the **Word (.docx) export** from Qualtrics
(Tools → Import/Export → Export Survey to Word) rather than the QSF —
same question content, a fraction of the tokens. Fall back to the QSF
only if a docx isn't available.
- Confirm the docx covers every measure used below before starting
analysis — full text and response options, not a partial export. If
anything's missing or looks truncated, ask the user for it rather than
guessing at wording or scale points.
### B. Study Design
3. **How many between-subjects factors?** (0 = correlational; 1 = simple;
2 = factorial, e.g., 2×2)
4. **How many cells / conditions?** What are the condition labels?
- 2-cell → t-test or one-way ANOVA
- 3-cell → one-way ANOVA + pairwise emmeans
- 2×2 factorial → two-way ANOVA + interaction
- No conditions → regression / correlational
5. **Is there a within-subjects factor?** (repeated measures)
6. **What column holds the condition?** (e.g., `condition`, `group`, `arm`)
7. **What is the participant ID column?** (e.g., `PROLIFIC_PID`,
`participantId`, `ResponseId`, `MTurkID`) — the name only, per the hard
rule above.
### C. DVs and Coding
8. **What are the main DVs?** For each DV:
- Column name(s) (e.g., items `dv_1`, `dv_2` → composite `dv_composite`)
- Scale range (1–5, 1–7, etc.)
- Number of items → composite = `rowMeans` / sum / divided
- Any **reverse-coded** items? (formula: `scale_max + 1 - item`,
e.g., for a 7-point scale: `8 - item`)
- Expected alpha (if known)
9. **Is any DV binary / count?** (e.g., yes/no outcome, an integer score)
If binary → chi-square + logistic regression
If count → also check Poisson / negative binomial
10. **What trait / covariate measures exist?** Note any prefix patterns
used for multi-item scales (e.g., `hexaco_`, `narc_`, `swb_`) so items
can be selected with `starts_with()`.
### D. Data Cleaning Specifics
11. **How is the sample deduplicated / restricted to valid respondents?**
Ask what the study uses: participant ID, IP address, completion status,
consent column, etc. Do not assume a specific platform's field names —
confirm the exact column names and values from the user or the CSV
header (names only, never the values in ID columns).
12. **Attention check(s)?** Ask what the attention-check item(s) are and
what the passing value/string is. Different studies (and different
platforms) use different formats — don't assume a fixed pattern.
13. **Comprehension checks?** (condition-specific filtering, e.g.
`comprehension == 1` required only for one condition)
14. **Any study-specific extra cleaning?** (e.g., drop first N rows of a
condition, merge in data from a second file, etc.)
15. **Extra header/metadata rows?** Some export formats (Qualtrics in
particular) include 1–2 extra metadata rows below the column names. If
so, add `slice(-c(1,2))` (or however many rows) right after
`read.csv()`.
16. **Does the study have bonus pay or behavioral scoring?**
(e.g., a performance score converted into a binary/continuous outcome)
### E. Analysis Plan
17. **Primary analysis:** t-test, one-way ANOVA, two-way ANOVA (specify
which), logistic regression, or multiple regression?
18. **Mediation?** If yes: DV, IV (contrast-coded), mediator(s),
PROCESS model number (4 = simple, 6 = serial two-mediator)
19. **Moderation / interaction?** If yes: IV × moderator, which DV?
Mean-center both before `lm()`?
20. **Any exploratory regressions?** (trait measures as predictors,
including interactions with condition)
21. **Reproducibility seed:** ask what seed the user's lab standardizes on
for bootstrapped analyses (e.g., PROCESS). If none exists, pick one and
note it explicitly in the file so it's easy to find and reuse.
---
## STEP 1 — QMD Structure
Always produce clearly separated QMD sections, e.g.:
```
## Data Cleaning
## Measures
## Main Analysis
## Other Analyses (mediation / moderation / exploratory)
```
Use `format: html` and `editor: visual` in the YAML header unless the user
asks for something else (e.g., `format: pdf` for a manuscript appendix).
---
## STEP 2 — Boilerplate to always include
### Package loader (always first chunk)
```r
ipak <- function(pkg){
new.pkg <- pkg[!(pkg %in% installed.packages()[, "Package"])]
if (length(new.pkg))
install.packages(new.pkg, dependencies = TRUE)
sapply(pkg, require, character.only = TRUE)
}
packages <- c("ggplot2", "tidyverse", "dplyr", "psych", "afex",
"effectsize", "emmeans", "apaTables")
# Add as needed: "stargazer", "lm.beta", "interactions", "mediation", "MASS"
ipak(packages)
```
### Standard data cleaning pipeline
Adapt every filter to the answers gathered in STEP 0 — the block below is a
template, not a fixed sequence:
```r
raw_data <- read.csv("FILENAME.csv")
# If the export has extra header rows, add:
# raw_data <- raw_data %>% slice(-c(1, 2))
clean_data <- raw_data %>%
filter(!is.na(PARTICIPANT_ID_COLUMN)) %>%
distinct(PARTICIPANT_ID_COLUMN, .keep_all = TRUE) %>% # deduplicate by ID
distinct(IPAddress, .keep_all = TRUE) %>% # if IP is available
filter(Finished == 1) %>% # if applicable
filter(consent == 1) %>% # if applicable
filter(ATTENTION_CHECK_COLUMN == PASSING_VALUE) # confirm from user
table(clean_data$condition)
```
---
## STEP 3 — Measures section
For **every scale**:
1. Convert items to numeric
2. Check alpha with `psych::alpha(..., check.keys=TRUE)`
3. Create composite
4. `psych::describe()` + `hist()`
```r
# Convert to numeric
clean_data <- clean_data %>%
mutate(across(c("item_1", "item_2", "item_3"), as.numeric))
# Alpha
psych::alpha(clean_data %>% select("item_1", "item_2", "item_3"))
# Composite (simple mean)
clean_data <- clean_data %>%
mutate(composite_dv = (item_1 + item_2 + item_3) / 3)
# Composite with rowSums (for prefix-based selection)
clean_data <- clean_data %>%
mutate(all_scale = rowSums(across(starts_with("prefix_")), na.rm = TRUE) / n_items)
psych::describe(clean_data$composite_dv)
hist(clean_data$composite_dv)
```
**Reverse coding** (for a 7-point scale: `8 - item`):
```r
clean_data <- clean_data %>%
mutate(
item_2 = 8 - item_2, # reverse
item_4 = 8 - item_4 # reverse
)
```
**Mean-centering** (always do before interactions/moderation):
```r
clean_data <- clean_data %>%
mutate(
mc_dv = as.numeric(scale(dv, center = TRUE, scale = FALSE)),
mc_moderator = as.numeric(scale(moderator, center = TRUE, scale = FALSE))
)
```
**Correlation table** (always include, put at end of Measures section):
```r
apa.cor.table(clean_data %>%
select("var1", "var2", "var3", ...),
filename = "cor_table.doc",
table.number = 1)
```
**Group summary** (always include):
```r
clean_data %>%
group_by(condition) %>%
summarise(across(c("dv1", "dv2"),
list(mean = mean, sd = sd), na.rm = TRUE))
```
---
## STEP 4 — Main Analysis
Choose the right test based on the study design:
### 2-cell between-subjects (t-test)
```r
t.test(dv ~ condition, var.equal = TRUE, data = clean_data)
cohens_d(dv ~ condition, data = clean_data, pooled_sd = TRUE)
```
For binary DV:
```r
tab <- table(clean_data$condition, clean_data$binary_dv)
chisq.test(tab, correct = FALSE)
prop.test(x = c(sum_a, sum_b), n = c(n_a, n_b), correct = FALSE)
```
### One-way ANOVA (3+ cells)
```r
anova_result <- afex::aov_car(
dv ~ condition + Error(PARTICIPANT_ID_COLUMN),
data = clean_data,
anova_table = list(correction = "none", es = "pes"),
type = 3, digits = 3
)
summary(anova_result)
effectsize::eta_squared(anova_result, partial = TRUE)
emmeans(anova_result, ~ condition, contr = "pairwise", adjust = "tukey")
```
> **Pilot studies:** default to `adjust = "none"` for pairwise comparisons
> unless the user says otherwise.
> **Confirmatory studies:** do not assume Tukey. Check the pre-registration
> and ask explicitly which adjustment (if any) was specified before writing
> the emmeans call.
### 2×2 Factorial ANOVA
```r
anova_result <- afex::aov_car(
dv ~ factor_A + factor_B + factor_A:factor_B + Error(PARTICIPANT_ID_COLUMN),
data = clean_data,
anova_table = list(correction = "none", es = "pes"),
type = 3, digits = 3
)
summary(anova_result)
effectsize::eta_squared(anova_result, partial = TRUE)
emmeans(anova_result, pairwise ~ factor_B | factor_A, adjust = "none")
emmeans(anova_result, pairwise ~ factor_A | factor_B, adjust = "none")
```
### Standard emmeans bar plot (use for ALL ANOVAs)
```r
emm_results <- emmeans(anova_result, ~ condition) %>% as.data.frame()
ggplot(emm_results, aes(x = condition, y = emmean)) +
geom_col(fill = "steelblue") +
geom_errorbar(aes(ymin = emmean - SE, ymax = emmean + SE), width = 0.2) +
theme_minimal() +
labs(title = "DV by Condition", x = "Condition", y = "DV Label") +
theme(text = element_text(size = 14))
```
For 2×2 factorial (grouped bars):
```r
emm_results <- emmeans(anova_result, ~ factor_A * factor_B) %>% as.data.frame()
ggplot(emm_results, aes(x = factor_A, y = emmean, fill = factor_B)) +
geom_col(position = position_dodge(0.7), width = 0.6) +
geom_errorbar(
aes(ymin = emmean - SE, ymax = emmean + SE),
position = position_dodge(0.7), width = 0.2
) +
theme_minimal() +
labs(x = "Factor A", y = "DV", fill = "Factor B")
```
### Logistic regression (binary DV)
```r
# Create contrast-coded IV first
clean_data <- clean_data %>%
mutate(condition_con = case_when(
condition == "control" ~ -1,
condition == "treatment" ~ 1
))
summary(glm(binary_dv ~ condition_con + covariate, family = "binomial",
data = clean_data))
```
---
## STEP 5 — Other Analyses
### Moderation (always mean-center first)
```r
library(interactions)
mod_model <- lm(mc_dv ~ condition_con * mc_moderator, data = clean_data)
summary(mod_model)
interact_plot(mod_model,
pred = mc_moderator,
modx = condition_con,
interval = TRUE,
x.label = "Mean-Centered Moderator",
y.label = "Mean-Centered DV")
sim_slopes(mod_model,
pred = mc_moderator,
modx = condition_con)
```
### Mediation (PROCESS macro)
Requires Andrew Hayes' `process.R` file (downloadable from
processmacro.org) in the same directory. Use the reproducibility seed
confirmed in STEP 0.
```r
source("process.R")
# Contrast code IV (e.g., -1 = control, 1 = treatment)
clean_data <- clean_data %>%
mutate(condition_con = case_when(
condition == "control" ~ -1,
condition == "treatment" ~ 1
))
# Simple mediation (Model 4)
process(data = clean_data,
y = "dv", x = "condition_con", m = "mediator",
model = 4, total = 1, normal = 1, boot = 10000, seed = SEED)
# Serial two-mediator (Model 6)
process(data = clean_data,
y = "dv", x = "condition_con", m = c("med1", "med2"),
model = 6, total = 1, normal = 1, boot = 10000, seed = SEED)
```
**Note on contrast coding for 3-cell designs:**
```r
# Collapse two conditions into one contrast (e.g., focal vs. rest)
clean_data <- clean_data %>%
mutate(condition_con = case_when(
condition == "control" ~ -1,
condition == "focal" ~ 2,
condition == "other" ~ -1
))
```
### Exploratory regressions with trait measures
```r
# Always mean-center trait measures before interactions
summary(lm(dv ~ condition_con * mc_trait, data = clean_data))
summary(lm(dv ~ condition + trait1 + trait2 + trait3, data = clean_data))
```
### Effect coding (for 3+ cell ANOVA in regression)
```r
clean_data$effect_condition <- factor(clean_data$condition,
levels = c("cell_A", "cell_B", "cell_C"))
contrasts(clean_data$effect_condition) <- contr.sum(3)
summary(lm(dv ~ effect_condition * mc_trait, data = clean_data))
```
---
## STEP 6 — Output conventions
- Name the QMD file to match the user's project naming convention (ask if
unclear); a common pattern is `studyname_studynumber.qmd`
(e.g., `pilot01.qmd`, `study2b.qmd`)
- The data CSV should be in the same directory — use a relative path
- `process.R` must also be in the same directory if mediation is used
- Always end the file with a blank code chunk so Quarto doesn't complain
- Report alphas as inline comments above the alpha chunk
(e.g., `# alpha = 0.88`)
---
## STEP 7 — Render it, then close the loop
Writing the QMD isn't the finish line — a typo, a missing package, or a
wrong column name won't surface until the file actually runs. Before
calling it done:
- If you have terminal access, render it yourself: `quarto render
FILENAME.qmd` (or `Rscript -e 'rmarkdown::render("FILENAME.qmd")'` without
Quarto). If it errors, fix it and re-render. Report what ran cleanly and
flag anything that didn't.
- If you can't render it (no terminal access, R/Quarto not installed), say
so plainly and give the exact command for the user to run it themselves
(in RStudio, the Quarto CLI, or their usual setup). Don't call the
analysis "done" just because the code looks correct — code you haven't
run can still be wrong.
- Ask whether they want a written **results summary** of the output. If
yes, hand off to the `results-summary-report` skill — it turns the stats
this QMD produces into a clean report.
---
## Multi-survey / linked designs
Some studies collect data across two or more linked surveys — e.g. a sender
and a receiver, an actor and a rater, or a dyadic pairing where each side
fills out a separate instrument. The clean workflow for these:
### Cleaned CSV generation (one-time R script)
Run a separate `generate_cleaned.R` script once, producing a single cleaned
CSV that the analysis QMD reads from (rather than re-cleaning raw exports
inside the analysis file itself).
```r
clean_side_a <- raw_side_a %>%
filter(!is.na(PARTICIPANT_ID) & PARTICIPANT_ID != "") %>%
distinct(PARTICIPANT_ID, .keep_all = TRUE) %>%
distinct(IPAddress, .keep_all = TRUE) %>%
filter(ATTENTION_CHECK_COLUMN == PASSING_VALUE) %>%
filter(Finished == 1)
clean_side_a$link_id <- seq_len(nrow(clean_side_a)) # assigns a shared linking ID
clean_side_b <- raw_side_b %>%
filter(!is.na(PARTICIPANT_ID) & PARTICIPANT_ID != "") %>%
distinct(PARTICIPANT_ID, .keep_all = TRUE) %>%
distinct(IPAddress, .keep_all = TRUE) %>%
filter(ATTENTION_CHECK_COLUMN == PASSING_VALUE) %>%
filter(Finished == 1)
clean_side_b$link_id <- as.integer(clean_side_b$link_id) # already present in CSV, or joined in
```
**Combine:** use `intersect(common_cols, colnames(df))` to handle column
differences between the two surveys safely.
```r
all_valid <- bind_rows(
clean_side_a[, intersect(common_cols, colnames(clean_side_a))],
clean_side_b[, intersect(common_cols, colnames(clean_side_b))]
)
```
**Cleaned CSV columns:** raw items, reversed items, composites, any
human-coded variables merged in from a separate coding file, and an
`include` flag marking the final analysis sample.
### Building multiple analysis datasets in the QMD
If the study has both a pre-registered sample definition and a broader
exploratory sample, build both explicitly and label which is which:
```r
all_valid <- read.csv("study_cleaned.csv") %>%
mutate(condition_effect = ifelse(condition == "side_a", -1, 1))
# Dataset 1: all valid participants, no additional exclusions
data_full <- all_valid
# Dataset 2: pre-registered sample (e.g., strict 1:1 linking)
data_prereg <- data_full %>%
group_by(link_id) %>%
filter(sum(condition == "side_a") == 1 & sum(condition == "side_b") >= 1) %>%
group_modify(~ bind_rows(
.x %>% filter(condition == "side_a") %>% slice_head(n = 1),
.x %>% filter(condition == "side_b") %>% slice_head(n = 1)
)) %>%
ungroup()
```
---
## Formatting pass for QMD files (run right before sharing/submission)
This is a **formatting and prose-style pass only** — it never changes
analysis logic, code behavior, or any computed value. Think of it as
tidying prose and headers, not editing content. Apply it to **every**
`*.qmd` in the project — analysis notebooks and cleaning notebooks alike,
since the same patterns creep into both:
- **Strip manual header numbering** if `number-sections` isn't set in the
YAML — hand-typed numbers like `## 1. Data`, `## 4. Main analysis` should
become `## Data`, `## Main analysis`.
- **Remove rationale-style parentheticals from headers**, e.g. `## Main
analysis (preregistered: strict 1:1 matching)` → `## Main analysis`. A
header explaining its own reasoning reads like generated commentary, not
something a person would write. **Keep** parentheticals that are pure
citations a reviewer needs (pre-registration IDs, filenames, table
numbers) — those are documentation, not interpretation.
- **Remove hypothesis-statement headers/prose**, e.g. `### H2: Group B
reports greater X than Group A.` as a header, or the same sentence as
bare prose above a chunk. Replace with a plain header naming the DV
(`### Perceived ambiguity`) or delete the line if a clear header already
exists — the code below already shows the test direction.
- **Remove `cat("label:", value, "\n")` print-wrapper chunks.** Use the
bare expression instead, so Quarto auto-prints the number without manual
string formatting.
- **Remove top-of-file intro paragraphs** that just restate provenance
(e.g., "Reproduces the preregistered analysis from the cleaned data
produced by X_cleaning.qmd") — folder structure and filenames already
make that clear, so it's just filler.
- **Remove methodology-description prose under a header** that just
restates what the code chunk directly below already shows, including
short section-label sentences like "Build the analysis datasets:" sitting
alone above a chunk.
- **Remove standalone "why is this empty/missing" comments** sitting alone
in an otherwise-empty chunk. Delete the whole chunk (and its header, if
it has no other content) rather than leaving a comment-only placeholder.
- **Keep** comments attached to a real, executing line that are needed to
understand it — e.g. which sample a dataset represents, or why an item is
reverse-worded. Test: does the comment explain something the code doesn't
already show (cut if it's just rationale or restating the obvious), or is
it the only label for an otherwise-opaque line (keep)?
- Re-render every file you touch and confirm it still exits cleanly before
moving on; these are text/formatting-only changes and should never alter
any computed number.
---
## Common packages by use-case
| Need | Package |
|------|---------|
| ANOVA | `afex` |
| Effect sizes | `effectsize` |
| Pairwise comparisons | `emmeans` |
| Descriptives + alpha | `psych` |
| Correlation table (APA) | `apaTables` |
| Standardized betas | `lm.beta` |
| Interaction plots | `interactions` |
| Negative binomial | `MASS` |
| Mediation (causal) | `mediation` |
| Mediation (PROCESS) | `source("process.R")` |
| Tables for paper | `stargazer` |
---
## Checklist before delivering the QMD
- [ ] No participant ID / IP / email / name column values were ever read, printed, or inspected — names only
- [ ] Survey docx (or QSF fallback) confirmed to contain full wording/scale info for every measure used
- [ ] `ipak()` includes all packages used in the file
- [ ] Attention-check / cleaning filters match what the user described in STEP 0
- [ ] All scale items converted to numeric before `psych::alpha()`
- [ ] Reverse-coded items handled before composite calculation
- [ ] `table(clean_data$condition)` printed after cleaning
- [ ] `apa.cor.table()` included
- [ ] Emmeans bar plots for every ANOVA DV
- [ ] Condition contrast-coded before `process()` / `lm()` interactions
- [ ] Reproducibility seed confirmed with the user and used consistently in `process()` calls
- [ ] File saved as `.qmd` in the same folder as the data CSV
- [ ] Pairwise adjustment: if pilot, default `adjust = "none"`; if confirmatory, check the pre-registration and ask explicitly before choosing an adjustment method
- [ ] File rendered successfully (by you, or confirmed by the user) before calling it done
Pliki w paczce (2)
- LICENSE 1 KB
- SKILL.md 22 KB
Wersje
W bazie nie ma innych skilli o nazwie survey-data-analysis.
Migawka bazy z 13.11.2026 · wersja paczki 465e3963fa70 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100experimental-data-analysis
Prowadzi analizę statystyczną i raportowanie zbiorów danych eksperymentalnych: interpretacja wyników, testy istotności (t-testy, ANOVA) i powtarzalne raporty. - 100quarto
Podaje zasady formatowania i wykonywania dokumentów Quarto w plikach qmd: nagłówki, bloki kodu, callouty i odwołania krzyżowe. - 100quantitative-analysis
Planuje i wykonuje analizę ilościową danych ankietowych: czyszczenie, statystyki, testy, regresję i interpretację bez nadmiernych wniosków. - 100survey-analytics
Analizuje zbiory odpowiedzi ankietowych względem schematu ankiety i tworzy ustrukturyzowane wnioski, streszczenia, spostrzeżenia o segmentach oraz rekomendacje działań. - 98data-analysis
Prowadzi pracę z danymi ilościowymi od czyszczenia po wnioski: eksploracja, testy statystyczne, modele i wizualizacje w Pythonie lub R, z opisem interpretacji. - 90deploy
Renderuje slajdy Quarto z plików .qmd do HTML i synchronizuje je do katalogu docs dla GitHub Pages po zmianach, które mają trafić do publikacji.