What we commit to, on every record.
forwardlab exists to make an AI trading track record you can verify instead of trust. These are the rules that make the record mean something — the same for the house strategies and for yours.
Pre-registered before it runs
Every strategy is written as a spec and locked — a content hash with a timestamp — before its first live decision. Editing a locked strategy forks a new one; the original record always stands.
Forward-only, never backfit
The clock starts at lock. A model decides on data it could not have seen, and no performance score exists until enough live trades have accrued. There is no backtest left to curve-fit.
Losses stay on the record
Nothing is cherry-picked, deleted, or dressed up after the fact. Missing data stays missing, drawdowns stay visible, and each strategy is measured against a benchmark and a naive baseline exactly as it ran.
Every decision, in the model's own words
Each trade is journaled with the model's verbatim reasoning and its exact version. You can judge the process — a coherent thesis versus a lucky guess — not just the outcome.
A refusal is the product working
When a signal can't be computed point-in-time, or a data feed is stale, we decline to reason rather than fabricate — and we say so, in the open.
Paper only — never a solicitation
Every record is forward-tested on live market data with simulated fills. Nothing here is investment advice or a recommendation to buy or sell any security.