Benchmarks
Performance tests, throughput measurements, and the methodology behind them.
Overview
The infrastructure is built for quantitative research workloads — not one-off backtests. The benchmark data below reflects real execution performance on production infrastructure. All results are reproducible.
Reference Benchmark
The primary benchmark is BTC/USDC over 5 years at 15M timeframe. This represents approximately 119,000 candles — a meaningful workload for evaluating infrastructure performance.
Execution Time by Timeframe
All measurements use BTC/USDC over a 5-year period. Shorter timeframes have more candles and therefore take longer.
| Timeframe | Candles (5y) | Approx. execution time |
|---|---|---|
| 1D | ~1,243 | ~5ms |
| 4H | ~7,449 | ~34ms |
| 2H | ~14,896 | ~51ms |
| 1H | ~29,789 | ~67ms |
| 30M | ~59,573 | ~104ms |
| 15M | ~119,137 | ~262ms |
| 5M | ~357,396 | ~1.3s |
Times are approximate and may vary based on strategy complexity (number of indicators, lookback periods) and current infrastructure load.
#Methodology
What is measured
- Total wall-clock time from API request receipt to result availability.
- Data is pre-loaded and cached on the infrastructure. Cold-start times are not included.
- Measurements exclude network latency between client and API server.
- Strategy uses configuration.maxOpenPositions: 1 — a single open position at a time. Without a cap, positions can pile up on fine-grained timeframes (many more candles for a fixed % stop-loss to resolve), which dominates runtime independently of caching — capping it reflects typical usage.
- Each row is a single execution of the same strategy, with a warm data cache.
Test strategy used
Benchmarks use a standard strategy: EMA 9/21 crossover + RSI confirmation + volume spike, on 1H timeframe. This is a representative strategy with medium computational complexity (3 indicators, 3 conditions).
Reproducibility
All benchmark results are reproducible. Given the same strategy, asset pair, and date range, the execution time will be within the published ranges. You can verify this by running your own benchmarks via the API and comparing against the published numbers.
Throughput & Concurrency
The infrastructure supports concurrent execution. Multiple backtests can run simultaneously without blocking each other. This is designed for:
- Parameter optimization loops (running 100 variations of the same strategy in parallel).
- AI agent workflows that generate and test multiple hypotheses simultaneously.
- Multi-asset testing (running the same strategy on BTC, ETH, and SOL in parallel).
- Multi-timeframe testing (running the same strategy across 15M, 1H, and 4H simultaneously).