← sources

Do you trust Vibe-Coded Backtesters?

source: reddit (reddit:algotrading)published 2026-09-26 14:21ingested 2026-09-27 02:40

retained text

I've been developing backtesting software for more than 25 years, and something interesting is happening lately: we're seeing an explosion of new backtesting platforms built with AI. I think that's great. AI has dramatically lowered the barrier to building sophisticated software. But building something that produces a backtest is very different from building a backtesting engine you can trust.

Over the years we've had to deal with things like survivorship bias, historical index constituents, competing limit/stop orders, NSF positions, dividends, splits, slippage, commissions, market holidays, intraday synchronization, position sizing, and countless edge cases that aren't obvious when you first build the engine.

Ironically, I'm also going heavily in the other direction: integrating AI directly into WealthLab so Claude can create strategies, run backtests, analyze the results, and iterate on them. My feeling is that the sweet spot isn't AI replacing the backtester. It's AI sitting on top of a mature backtesting engine.

I'm curious what others think. As it becomes trivial to generate your own trading and backtesting software with AI, how much do you trust the results?

document fetched 2026-09-27 00:40 · canonical https://reddit.com/r/algotrading/comments/1wqrr91/do_you_trust_vibecoded_backtesters