Launched this week
Reamer
Quantitative Research Infrastructure
9 followers
Quantitative Research Infrastructure
9 followers
REAMER is local-first quantitative research infrastructure for systematic trading. Write a strategy in Python; a C++ engine replays it with deterministic synthetic ticks, verified against a 282-check execution spec. Monte Carlo robustness, tick-level replay. Multi-asset (FX, futures, equities), plus direct Databento loading (no CSV timezone bugs). Free GUI; licensed SDK, free test license on request. Proves a strategy is real before it risks anything.
Hey everyone.
I built REAMER because I kept not trusting my own backtests. A strategy would look great in a notebook, and I'd have no real way to know if that was a genuine edge or curve-fit noise, or whether the fills were even realistic. Existing tools either made that hard to check or didn't try.
REAMER is local-first quantitative research infrastructure: write a strategy in Python, a C++ engine replays it with deterministic synthetic ticks, and every fill is checked against a published execution spec (282 conformance checks). Monte Carlo testing catches overfitting before you risk anything, and tick-level replay in the GUI shows exactly which tick filled a stop or triggered a bracket on the same seeded run. Multi-asset (FX, futures, equities), 10.3-15.6x faster than Backtrader and QuantConnect LEAN on identical workloads.
One thing worth saying up front: it's research-only today — it proves a strategy is real; it doesn't place trades or connect to a broker. That's the step before deployment, not instead of it.
If you're already pulling data from Databento, there's a direct loader — mostly just saves you a CSV export/import step and a specific class of DST bug, nothing dramatic.
The GUI's free, no signup. The Python SDK needs a license to run — happy to send a free, time-limited one to anyone here who wants to actually try it on their own data.
Would genuinely love feedback, especially from anyone who's been burned by a backtest that lied to them.
the local-first angle is refreshing, no more babysitting cloud jobs just to backtest a 50-line strategy. tick-level replay against a 282-check spec is a serious flex, exactly what I needed after CSV timezone headaches on Databento data. solid.
Thanks @nihatlajx. Really appreciate that. The execution model and local-first workflow were the two things I was unwilling to compromise on from day one.
the determinism angle is what caught me, validating a Python strategy against 282 checks in a C++ engine is the kind of rigor that should be table stakes but rarely is. love the databento integration too, no more fighting timezone offsets at 2am.
Thanks @tugay865925. Appreciate it. Execution drift between research assumptions and implementation has been one of the biggest problems I've run into. Making execution behavior explicit and deterministic was the core design goal from the beginning.