FactorHelm

What we won't claim

A research system earns trust by what it refuses to say. These are the claims we won't make until the evidence supports them, and some never.

That we have proven alpha.Not without enough prospective, independently reconcilable evidence. We don't have it yet.
That our checks are independent certification.The research, the validation and the decision are under common control. Until an outside party reproduces our results, we say "separate", not "independent".
That agents arguing with each other produces truth.Two agents built on the same model can share the same mistake.
That more data or more agents make results better.Each one has to beat a simpler baseline to stay.
That the system improves itself from daily data.Models change only through versioned releases with frozen evaluations and a person's approval.
That a historical replay is a track record.Replays are labelled as reconstructions everywhere they appear.
That a research approval is permission to trade.FactorHelm doesn't place orders. Execution sits outside it.
That we manage money.We don't, and won't before the legal, risk, operational and custody requirements are met.
That your data trains our models.Not by default, and not without a separate written agreement.
That data volume is a moat.Rows collected say nothing about whether a thesis was right.
That backtests predict the future.They show what a rule would have done, under assumptions we list.
That a decision record removes human accountability.Every decision names the person who made it.

Words we use carefully

Our validation path runs apart from the research it tests, but the same company builds both.

A thesis that passes its gates is qualified for a stated use. Whether it was right is decided later, by the outcome.

"Long reversion" and "short fade" describe the direction of a research thesis. They are not orders.

A replay is reconstructed after the fact. A call is recorded before the outcome exists.