Backtest results vary with different chunk_size values — potential inconsistency in synthesized bar data

[ReCodeLife]

Hello,

I'm encountering a potentially significant issue while running backtests using this library and would appreciate your insight.

Issue Description

I'm performing backtesting using quote/tick data that is aggregated into minute-level bars. Due to the large volume of raw data, I enabled the chunk_size parameter to process the data in chunks. However, I've observed that the final backtest PnL (or cumulative returns) differs depending on the value of chunk_size.

Upon closer inspection, it appears that the synthesized minute bars themselves are not identical across different chunk_size settings—for example, slight discrepancies exist in OHLC values or volume for certain bars. Since my strategy logic depends on these bars, even minor differences can lead to divergent trade signals and performance outcomes.

Questions

1. Is this behavior intended (e.g., due to edge-case handling at chunk boundaries during resampling), or could it indicate an unintended inconsistency?
2. If intentional, could the documentation clarify how chunk_size might affect data aggregation and backtest reproducibility?
3. If unintentional, is there a known workaround (e.g., avoiding chunking for critical runs), or is a fix planned?

Thank you very much for your excellent work on this project! I’d be grateful for any clarification or guidance.

[ReCodeLife]

Hi @cjdsellers

I’d really appreciate it if you could take a look at this issue when you have time. The inconsistency in bar synthesis across different chunk_size values is affecting backtest reproducibility, and I’m not sure whether it’s expected behavior or a subtle bug. Thanks in advance!