The datasets inside packages are read from their data/ folders when the code is analysed. Each object is described by its class, its shape and its columns, and fingerprinted so the same data can be found in other packages.
Loadable only
Only objects that load by name with data() are listed, 76,161 on CRAN and 8,045 on Bioconductor, so the list is not every file in data/.
Columns
Tables open to their columns, and each kind of column has figures of its own.
Fingerprint
Each object's values are fingerprinted, so the same data in another package is found whatever its name. On CRAN 8,314 datasets hold data that is also in another package, 4,094 distinct pieces of data in all.