We developed two approaches:
Browser version: Uses a Web Worker to process PGN locally, with a 25 MiB input limit to keep browser resource usage manageable.
Windows portable: Uses disk-backed aggregation for larger PGN collections, generating standard 16-byte Polyglot BIN records offline.
Benchmark results:
Successfully processed an 820 MiB real-world PGN collection.
A separate high-cardinality test generated a 250.9 MiB BIN containing 16,441,386 entries.
Output records were validated for sorting and duplicates.
The interesting challenge was handling large collections without letting memory consumption grow uncontrollably.
The project is free and open source:
https://github.com/elcriollito/CAISSA-Book-Creator
Sharing the implementation and benchmark results for anyone interested in Polyglot book generation and chess engine tooling.