The dataset, covering 4.68 million half-hour measurements across 210 municipal parking lots in Moscow, was collected from March 2025. The data is available on GitHub, Kaggle, and Hugging Face under the CC BY 4.0 license.
An investigation into negative occupancy rates revealed that 19,792 rows contained values as low as -1000. These anomalies occurred between May and December 2025, before a clamping function was implemented in the collector script in January 2026.
Attempts to recalculate the data using historical capacity values failed due to drift in reference data. Analysis showed that 74 out of 190 parking lots changed their capacity over the 19-month period, with one lot recording 21 different capacity values.
The study also identified inconsistencies in the free_spaces counter, which was incorrectly documented as a total sum. Data shows that 2.5% of rows have free_spaces exceeding common_spaces, and 2.0% of rows show free handicapped spaces exceeding total capacity.
The author opted to update the documentation rather than modify the raw data to preserve the integrity of the observational archive. Updated README files and Kaggle descriptions now include details on known data quirks and specific filters for users.
