# Roulette research dataset guide

Research cut-off: 4 October 2026. This package contains 926 observations, 36 country/scope rows covering 31 jurisdictions, 128 catalogue rows and a source/query audit trail.

## Files

- `global-roulette-market-intelligence.pplx.md`: 26-section report and complete requested master table.
- `roulette-master-data.csv`: primary analysis file, with evidence status, definition, source type and retrieval date beyond the requested columns.
- `roulette-country-dataset.csv`: national/subnational availability register. Blank means not established, not zero; see Scope before using a country row.
- `roulette-search-demand.csv`: Ahrefs overview, country and monthly-history observations; modelled search estimates, not player counts.
- `roulette-catalogue-snapshots.csv`: visible BetMGM Netherlands and LeoVegas Ireland-facing title/provider observations; not wheel counts or complete regulated catalogues.
- `roulette-source-register.csv`: read, failed and browser-extracted source URLs.
- `roulette-search-log.csv`: query/discovered-URL pairs; search results are not automatically verified evidence.

## Units and aggregation

Money in master data uses full currency units unless Unit explicitly says otherwise. Most original reports used millions or thousands; those were converted in the data builder. Original currencies are retained and no FX-normalised global sum is calculated.

Do not sum all rows. Annual/quarterly windows, state/subregion figures, comparator games and duplicated primary/secondary historical observations intentionally overlap. Filter by Metric, Country, Region, Year, channel, variant and evidence status first.

GGR/GGY/casino win, turnover/stakes, physical-table drop, deposits, supplier revenue and player outcomes are not interchangeable. Denmark's monetary values retain its report's GDP-adjusted2026level; GB uses fiscal years and excludes Northern Ireland; Macau includes tourism-driven casino activity.

## Evidence labels

A = government/regulator/audited report; B = established research/modelled dataset or clearly identified primary operational disclosure; C = identifiable industry reproduction; D = unclear secondary evidence, not used as an established statistic. Source reliability is not the same as representativeness or confidence interval.

CALCULATED ESTIMATE labels include simple exact arithmetic on reported inputs as well as calculations from rounded inputs. The notes identify formulas; they are not independent observations. SURVEY RESULT and OBSERVED SAMPLE DATA retain their sample restrictions. Provider specifications are not audited throughput measurements.

## Reproducibility

The project stores build scripts and extracted evidence separately in `reproducibility-evidence.zip`. Ahrefs connected-tool results and browser-observed DGOJ/DICJ figures were transcribed into the builder with their retrieval scope; public documents are linked in each row. Query logs include failed or unproductive searches so missing data is not silently represented as zero.

The research is a bounded public-evidence audit, not a worldwide census. A future update should replace source vintages coherently, retain revisions, capture full original regulator downloads where possible, and append new periods instead of silently overwriting comparable history.