Skip to main content
See the full BENCHMARKS.md for detailed numbers.

​
Summary

LayerOperationTypical Time
Glassboxvalidate_input (100 chars)~396 ns
Glassboxvalidate_command~248 ns
Glassboxvalidate_output (100 chars)~215 ns
Ironcladvalidate_binary~1.1 µs
PersonaVoice filter (clean, 100 chars)~3.9 µs
PersonaVoice filter (clean, 1000 chars)~11.8 µs
ShadowStatic analysis (10 lines)~7.4 ms
ShadowSecrets analysis (100 lines)~428 µs
Cortextrack_edit (single)~85 ns
Cortexdetect_patterns (100 edits)~426 µs
Glassbox containment adds less than 1µs per call — effectively zero overhead on any LLM pipeline where the LLM call itself takes 500ms+.

​
Running Benchmarks

# All crates
cargo bench --workspace

# Specific crate
cargo bench -p laminae-glassbox

# Specific benchmark
cargo bench -p laminae-shadow -- static_analyzer

# HTML reports
cargo bench --workspace
open target/criterion/report/index.html