Benchmarks
Looking for the indicator library's numbers?
This page is about Wickra Verify. Wickra's own indicator benchmarks — the comparison against TA-Lib, talipp, pandas-ta and the other Rust TA crates — live at wickra.org.
wickra-verify ships a Criterion bench crate (crates/verify-bench) measuring the cost of verify (recompute the backtest with the wickra-backtest engine + compare the report field by field), and of compare and canonicalize in isolation.
verify is dominated by the engine recomputation and scales roughly linearly with the number of bars; the field comparison and canonical hashing are a small, near-constant tail. The figures below are indicative, from a local run on Windows (x86_64, release profile); the authoritative numbers come from the nightly bench.yml run on the CI runners.
| Operation | Input | Time (indicative) |
|---|---|---|
verify | 200 bars, 2 indicators | ~2.8 ms |
verify | 1 000 bars, 2 indicators | ~12 ms |
verify | 5 000 bars, 2 indicators | ~79 ms |
verify | 5 000 bars, 10 indicators | ~89 ms |
compare | small vs large report | sub-millisecond |
canonicalize | small vs large report | sub-millisecond |
verify from 200 → 5 000 bars grows about linearly (≈ 28× the time for 25× the bars), as expected from the engine recomputation; adding indicators (2 → 10) is a modest per-bar cost. compare and canonicalize are negligible next to the recomputation — which is the point: verification's cost is re-running the backtest, not checking the numbers.
Run locally:
cargo bench -p verify-benchThe numbers above are the ones in the repository's BENCHMARKS.md, measured with the commands it names.