Benchmark results

What we measured

QueryBolt maintains a benchmark suite of curated query patterns — each one exercises a specific optimization rule and runs end-to-end through the same pipeline your queries use. Every pattern executes on a live engine against a named synthetic scenario, capturing the original and optimized query’s execution time, rows examined (MySQL) or logical reads (SQL Server), and engine-reported cost. These measurements describe the tested scenario rather than every production distribution or parameter value.

Headline numbers

In the most recent suite run, 260 patterns produced a qualifying optimization — 161 on MySQL 8 and 99 on SQL Server 2022. Across those optimizations the median run scanned 98% fewer rows, and the algorithmic rewrites cut execution time by 99%+ — over 100× faster.

Representative measured results

A spread of individual patterns from the suite. Each figure is a measured median across repeated runs on a live engine:

• MySQL 8 · DATE(col) equality rewritten to a half-open range: 99.7% fewer rows scanned
• MySQL 8 · correlated aggregate subquery rewritten to a window function: 99.8% less execution time
• MySQL 8 · covering-index recommendation: 98.8% fewer rows scanned
• SQL Server 2022 · filtered-index recommendation: 99.96% fewer logical reads
• SQL Server 2022 · covering-index recommendation: 99.1% fewer logical reads
• SQL Server 2022 · LEFT() / SUBSTRING() prefix rewritten to be sargable: 97.9% fewer logical reads

How to read these numbers

Each pattern is seeded with data where its optimization applies, so these figures show what a qualifying optimization delivers — not an average across every possible query. Your own results depend on your schema, data distribution, and query shape, which is exactly why every QueryBolt run benchmarks your query in isolation and reports its own measured before/after instead of a generic estimate.

Kept honest

The suite is re-run as the optimizer changes, so these numbers track the product rather than a one-off snapshot. See Benchmarking methodology for how each measurement is taken and the minimum-improvement bar a change must clear before it counts as an optimization.