MaskOps 2.0 shipped this week. Before I told anyone, I looked at my own numbers. Two of them were lying to me, in opposite directions.
MaskOps is a Rust plugin for Polars that does PII masking inside the dataframe: RUT, CPF, credit cards, IBANs, and twenty-odd more families, air-gapped, with no network call, ever. If you have reached for Microsoft Presidio and found it carries no Latin American identifiers, that is the gap MaskOps fills. It does check-digit-validated RUT, CPF, and CURP detection alongside the EU, US, and APAC families, as a native Polars expression. Version 2.0 is the enterprise line: configurable patterns, structured extraction, an audit pass that counts what it masked, and format-preserving encryption (GDPR Art. 4(5) pseudonymization) for the reversible cases. That part I was sure of. The numbers around it, less so.
The first number lied against me: the benchmark
The last thing I checked was the benchmark table in my own README. It said MaskOps ran at 0.4× to 0.7× the speed of plain Python re. Slower than the language I wrote it to replace.
I almost opened the profiler. Instead I read the benchmark harness. I should have read it first.
