Benchmark

Setup

These numbers come from a simple local synthetic run against the bundled OMOP 5.4 DDL using valid PERSON.csv rows and --json.

• Linux 5.4
• 12 CPUs on the host
• single validator process
• no intra-file parallelism
• success-path validation only
• helper script in bench/

Results

Rows	File size	Default --json	--json --turbo
50K	4.8 MB	13.36 s	2.47 s
100K	9.7 MB	26.82 s	4.75 s
250K	24.5 MB	67.07 s	11.61 s
500K	49.3 MB	134.32 s	23.29 s

Takeaway

The main result is straightforward: on this workload, --turbo is consistently much faster than the default engine.

For these local runs, the speedup was roughly:

• 5.4x faster at 50K rows
• 5.6x faster at 100K rows
• 5.8x faster at 250K rows
• 5.8x faster at 500K rows

So in practice:

• default engine: safer baseline, slower
• turbo engine: faster, but maintained as a second engine

Recommendation

Use the default engine when:

• the CSV is not especially large
• you want the most conservative path

Use --turbo when:

• you are validating large files
• runtime is the main reason to switch
• you are staying within the current tested schema model

Caveat

Treat these numbers as a local synthetic reference, not a guarantee. Real throughput will move with disk speed, CPU, row width, error rate, and report mode.