Quantitative teams buy new datasets to find alpha. Those datasets rarely record when each fact became public, so simulations end up using information that did not exist yet. Lemma stores every record with both dates, never overwrites a correction, and runs reproducible backtests on top.
Two runs of one strategy on the same synthetic dataset: 562 bitemporal records, 8 issuers, 349 trading days, identical seed and identical 3bp costs. The only difference is which date decides whether a record is visible. One run uses the event date, the mistake most pipelines make by default. The other uses the publication date, which is what the market could actually see.
A Sharpe above two in a backtest should raise suspicion rather than expectations, and that is the point of showing it. The dataset is synthetic and the size of the gap belongs to this dataset alone. What does not depend on the data is the direction of the effect, and the fact that it stays invisible until capital is at risk. Re-run either simulation against the same data snapshot and the result checksum matches, which is what makes a claim auditable a year later.
No record enters a simulation before it was knowable. Everything else exists to serve that rule.
Alternative and financial feeds arrive in whatever shape the vendor sends. Each record is parsed and resolved to a stable issuer identity that survives renames, mergers and reused tickers.
Every record is stored with its event time and its knowledge time. Corrections are appended, never written over, so any past state of the record can be replayed exactly.
Each backtest runs in its own isolated environment, provisioned for that run and destroyed when it ends. Nothing leaks between runs and compute is charged by use.
API and SDK inside the customer's own stack, a console for working on screen, and distribution through data marketplaces so procurement is not the bottleneck.
We do not sell signals, we do not manage money and we never need to see a strategy. Lemma sits underneath the research process, not next to it.
Systematic funds, multi-strategy desks and institutional asset managers that onboard new datasets regularly and have to defend a track record to an allocator, an auditor or a regulator.
The demo is a working console on synthetic data: a bitemporal catalogue, an as-of explorer that shows exactly which records are excluded, and a sandbox that reproduces a run and verifies it. Leave an address and I will send it.
We are exhibiting at Web Summit Lisbon 2026. If you are attending, message me through the app and I will walk you through the engine at the stand.