The ongoing maintenance and technical optimization of data quality are at the heart of this project. In this changelog, I document all major updates, structural changes, and bug fixes in detail. Simply expand the relevant version to read all the information about the changes.
Version dated October 20, 2026
Minor issues in Event and Site tags has been fixed, Hoster changed
A bug that caused the name memory to overflow in the *.si4 database format has been fixed.
The download host has been changed.
Version dated October 11, 2026
Improved and More Reliable Database Exports
Since the last release on 8 July 2026, the export process for the complete databases and monthly updates has been substantially improved. The main focus has been on consistent files, stricter date validation, and a more reliable separation of online and over-the-board chess.
Mandatory Date Validation
All regular export files now undergo the same date validation:
- Complete dates must be valid calendar dates.
- Known years must be between 1400 and 2099.
- Dates must not be later than the cut-off date of the respective export.
- Unknown dates and plausible partial dates such as
2026.??.??remain permitted. - Obviously corrupted or ambiguous dates are no longer exported as unknown dates. Instead, the affected games are excluded from the export.
Date errors are corrected automatically only if the correct date can be derived unambiguously and consistently from available tournament information or the EventDate tag. Years are no longer completed based on assumptions alone.
A separate audit report with the suffix .date-rejections.jsonl is generated for every export. It lists games excluded because of date problems and documents the reason for each rejection.
Standardized PGN Dates
Plausible but inconsistently formatted dates are converted to the standard PGN format. This includes year-only dates and dates with an unknown month or day. Dates that cannot be interpreted unambiguously are not automatically “repaired”; instead, they are handled by the date validation process.
Reproducible Databases and Monthly Updates
The order of exported games is now independent of the order in which source files were imported or updated.
Games are sorted by:
- Date, with the newest games first
- Unknown dates last
- Event
- Site
- Round
- Complete game content as a unique sorting criterion
As a result, the same dataset and settings always produce the same export file. Complete rebuilds and incrementally updated datasets no longer differ in order solely because of how they were processed.
More Reliable Separation of Online and Over-the-Board Chess
Web addresses in the Site tag are standardized during export:
- Known playing platforms are exported as
Chess.com,Lichess,GameKnot, orChess24. - ChessBomb is identified as a broadcast platform.
- For other web addresses, only the actual hostname is retained; paths and technical URL components are removed.
Domains that clearly belong to playing platforms are now taken directly into account when assigning games to the online export. However, an arbitrary web address alone is not considered proof that a game was played online. In particular, Lichess broadcasts continue to be classified as online games only when there is specific evidence that the event itself was played online.
The existing rules for correspondence chess and historical games remain unchanged.
Consistent Elo Rating Floor
The same Elo rating floor can now be applied to complete databases and monthly updates. A game is excluded because of this threshold only if both players have ratings and both ratings are below the configured floor.
Games for which one or both ratings are missing remain included, ensuring that historical or incompletely documented material is not lost solely because Elo information is unavailable.
Navigation labels
- Monthly OTB Update
- Monthly Online Update
- Version History