System
Every dataset, what it is, where it came from, and how much of it there is. Read live from the database rather than stored, because a stored row count goes stale the moment a pipeline runs.
Footfall and point-of-sale figures are generated for a fictional brand and have no observed counterpart. Every row carries its label.
ERA5 reanalysis on roughly an 11 km grid, so the four sites land in nearby cells about a degree apart. The zones differ in how they RESPOND to weather, not in what weather they get.
Fetched from the source named below and not modified.
| Source | Open-Meteo historical reanalysis and forecast |
|---|---|
| Licence | CC BY 4.0 — attribution to Open-Meteo required |
| Verified | 2026-09-07 |
| Built by | pipeline/weather.py |
Islamic holidays are determined by moon sighting, so those rows are calculated and carry one day of uncertainty. School breaks are approximate windows: KHDA's calendar is not machine-readable.
Assembled by this project from public listings because no redistributable feed exists. Dates sit inside each series' established window and are not individually confirmed.
| Source | u.ae public holidays, converted with the Umm al-Qura calendar |
|---|---|
| Licence | Public information; Hijri conversion via hijridate (MIT) |
| Verified | 2026-09-07 |
| Built by | pipeline/calendar.py |
visitdubai.com returns 403 to non-browser clients and no redistributable feed exists. Series, venues and windows are real; the dates of a given edition sit inside the established window and are not individually confirmed.
Assembled by this project from public listings because no redistributable feed exists. Dates sit inside each series' established window and are not individually confirmed.
| Source | Public event listings, enriched from Wikipedia |
|---|---|
| Licence | Curated by this project from public listings; each row links the organiser's own page. Wikipedia summaries are CC BY-SA 4.0. |
| Verified | 2026-09-07 |
| Built by | pipeline/events.py |
No public hourly footfall series exists for a Dubai cafe. This is generated from the brand kit's seat counts and trading hours, coupled to the weather, calendar and event tables. The evaluation against a seasonal-naive baseline is what keeps it honest.
Generated. There is no observed counterpart — a real pilot would replace this with an actual footfall or point-of-sale export.
| Source | pipeline/footfall.py, seeded |
|---|---|
| Licence | Generated by this project; no third-party rights |
| Verified | 2026-09-07 |
| Built by | pipeline/footfall.py |
A basket-level extract for RFM segmentation, sampled from the generated transactions with a heterogeneous customer base so the segmentation has real structure to find.
A generated extract standing in for the export a pilot would provide.
| Source | pipeline/footfall.py, seeded |
|---|---|
| Licence | Generated by this project; no third-party rights |
| Verified | 2026-09-07 |
| Built by | pipeline/footfall.py |
Generated 2026-10-13T00:07:42+00:00. Full provenance, including the sources that refused, is in docs/datasets.md.