The Wandering Fork Dataset

Data Description | Reference

This page documents the dataset used throughout the course. All data are entirely invented — no real personal or business data are included.

Downloads

File	Format	Description
wandering-fork.csv	CSV	Full long-format table (240 rows × 16 columns)
wandering-fork.xlsx	Excel	Same data as an Excel Table (tblFork) plus subset sheets
payroll-skew.csv	CSV	Small skewed example (11 rows) for Modules 4 & 7
lakeside-bakery.csv	CSV	Fresh 2-outlet dataset (120 rows) for Module 10 walkthrough

The Excel workbook contains four sheets:

• data — the full 240-row table as an Excel Table named tblFork
• harbour_30d — the first 30 days of the Harbour franchise (hand-computation subset)
• campus_30d — the first 30 days of the Campus franchise (hand-computation subset)
• dictionary — the data dictionary below

Data dictionary

Name	Statistical type	Unit	Meaning
franchise	Qualitative nominal	—	Franchise name: Harbour or Campus
day	Quantitative discrete	day number	Day index within franchise (1–120)
date	Date	YYYY-MM-DD	Calendar date
day_of_week	Qualitative nominal	—	Monday … Sunday
month	Qualitative ordinal	—	April … August (calendar order)
weather	Qualitative nominal	—	Sunny / Cloudy / Rainy
temperature_c	Quantitative continuous	°C	Daily temperature
event_nearby	Qualitative nominal (binary)	—	Yes / No — event near the truck
promo_active	Qualitative nominal (binary)	—	Yes / No — promotion running
staff_on_duty	Quantitative discrete	people	Staff working that day
foot_traffic	Quantitative discrete	people	Passers-by count
customer_count	Quantitative discrete	customers	Number of customers served
items_sold	Quantitative discrete	items	Total items sold
daily_revenue_eur	Quantitative continuous	€	Total revenue for the day
waste_kg	Quantitative continuous	kg	Food waste (right-skewed)
satisfaction_rating	Qualitative ordinal	1–5	Customer satisfaction rating

Designed properties (verified by verify_dataset.py)

1. Harbour days 1–30 match the legacy CSV exactly on the eight original columns.
2. No missing values; no duplicate (franchise, day) pairs.
3. items_sold ≥ customer_count; customer_count ≤ foot_traffic; revenue > 0 everywhere.
4. Campus revenue is right-skewed (skewness > +0.5); Harbour revenue is near-symmetric (|skew| ≤ 0.5).
5. Strong customer–revenue correlation in both franchises (r ≥ 0.80); weak temperature–revenue correlation (|r| ≤ 0.35).
6. Weather × promo dependence visible in conditional profiles (≥ 15 pp row-profile gap in at least one franchise).
7. At least one 1.5 × IQR outlier per franchise.
8. Every category appears at least 5 times (no empty contingency cells).

Secondary datasets

payroll-skew.csv (Modules 4 & 7)

11 rows: one “CEO” salary (€180,000) plus ten staff salaries (€25k–€45k). Strong right skew for illustrating mean vs median vs mode and Pearson’s skewness coefficient.

lakeside-bakery.csv (Module 10)

Two outlets (Riverside, Hilltop) × 60 days = 120 rows. Variables: outlet, day, date, day_of_week, month, weather, temperature_c, foot_traffic, customer_count, items_sold, daily_revenue_eur, waste_kg, satisfaction_rating. Used for the Module 10 synthesis walkthrough on a fresh dataset.

How the data were invented

All values were generated by a seeded Python script (generate_dataset.py) using numpy/pandas. The script:

• Copies the original 30-day Harbour data verbatim for the eight legacy columns.
• Generates the remaining 90 days for Harbour and all 120 days for Campus with realistic seasonal, weekday, weather, and promotional effects.
• Adds the new columns (month, event_nearby, promo_active, staff_on_duty, foot_traffic, waste_kg, satisfaction_rating) for all 240 rows, including the first 30 Harbour days.
• Enforces logical constraints (e.g., customer_count ≤ foot_traffic).
• Calibrates two Harbour outliers (days 45 and 80) and Campus event-driven skew to satisfy the invariants above.

The seed is fixed (42) so the dataset is fully reproducible.