Performance Test Evidence

This directory contains evidence files generated by the performance benchmark test suite.

Purpose

Performance tests generate JSON evidence files that provide:

• Detailed timing metrics (mean, p95, p99, min, max)
• System information (CPU, OS, Python version, NumPy version)
• Test parameters (number of trials, data sizes)
• Timestamp of test execution

Usage

Evidence files are automatically generated when running performance tests:

# Run all performance tests
pytest tests/industry_standard/test_performance_benchmarks.py -v -m perf

# Or set environment variable
SCBE_RUN_PERF=1 pytest tests/industry_standard/test_performance_benchmarks.py -v

Evidence Files

Each test generates a JSON file with the following structure:

{
  "test_name": "layer1_complex_state",
  "system_info": {
    "timestamp": "2026-01-19T10:30:00",
    "platform": "Windows-10-10.0.19045-SP0",
    "processor": "Intel64 Family 6 Model 142 Stepping 12, GenuineIntel",
    "python_version": "3.11.0",
    "numpy_version": "1.24.0",
    "cpu_count": 8,
    "modules_available": true,
    "pqc_available": false
  },
  "results": {
    "mean_ms": 0.023,
    "p95_ms": 0.045,
    "p99_ms": 0.067,
    "min_ms": 0.015,
    "max_ms": 0.089,
    "n_trials": 1000
  }
}

Why Evidence Files?

1. Reproducibility: Captures exact system configuration and test parameters
2. Trend Analysis: Compare performance across different hardware/versions
3. Compliance: Provides audit trail for performance claims
4. Debugging: Helps diagnose performance regressions

Test Categories

Primitive Benchmarks

• layer1_complex_state.json - Complex state construction
• layer4_poincare_embedding.json - Hyperbolic embedding
• layer5_hyperbolic_distance.json - Distance computation
• layer6_breathing_transform.json - Breathing dynamics
• layer14_audio_axis.json - Audio telemetry

System Benchmarks

• memory_footprint.json - Memory usage tracking
• concurrent_operations.json - Concurrency speedup metrics

PQC Benchmarks (Optional)

• mlkem768_keygen.json - ML-KEM key generation

Notes

• Evidence files are not committed to git (see .gitignore)
• Tests are opt-in via -m perf marker or SCBE_RUN_PERF=1
• Performance thresholds are hardware-dependent - adjust as needed
• Concurrency tests report metrics rather than asserting (GIL limitations)