Data and downloads
Publication cutoff: 2026-10-26. Window starts 2021-10-27.
| Format | Competition | Available matches | Used | Players | Download |
|---|---|---|---|---|---|
| Test | Men | 197 | 182 | 399 | CSV |
| Test | Women | 11 | 11 | 72 | CSV |
| ODI | Men | 642 | 600 | 840 | CSV |
| ODI | Women | 344 | 317 | 463 | CSV |
| T20I | Men | 2539 | 2443 | 3671 | CSV |
| T20I | Women | 1646 | 1600 | 2315 | CSV |
Publication files
| Download | Contents |
|---|---|
| All player rankings | Format, gender, stable player ID, rank, coefficient, appearances, last appearance, and teammate-overlap diagnostics |
| Fitted models | Feature names, outcome classes, coefficients, intercepts, and regularization settings |
| Publication manifest | Match IDs, tuning splits and scores, exclusions, package versions, and checksums |
| Test provenance | Source-file hashes and acquisition receipt |
| ODI provenance | Source-file hashes and acquisition receipt |
| T20I provenance | Source-file hashes and acquisition receipt |
| Historical ODI experiment | Earlier fit cutoffs and alternative model specifications |
Coverage and exclusions
The source is Cricsheet’s international match archives. Available counts above cover the five-year window; used counts additionally apply the fixed-XI and outcome rules. Tests retain draws and ties. ODIs and T20Is require a decisive result. No-results and matches with ambiguous XIs or recorded match replacements do not enter the models. All players in eligible matches are ranked; low appearance counts are flagged, not removed.
This is not a census of international cricket. Source coverage differs by format and gender, and Cricsheet withholds some Afghanistan-related matches. The source remains subject to Cricsheet’s licensing terms. The preparation guide explains identity checks and date exclusions.
Using the downloads
Join players by stable Cricsheet person ID, not printed name. Keep format, gender, cutoff, and comparison group in the key when comparing ratings. Do not add scores across separately fitted models. A rating is a log win/loss odds coefficient; it is not a count of wins or a replacement-level estimate.
The saved publication is verified before every website build. Rendering regenerates format tables from those files without downloading match data or fitting models. See reproduction commands and the publication history.