Proof you can't fake

Every signal is generated on a fixed schedule and hash-stamped before the candle closes, with the entry price recorded at that moment. The call exists in our database before the market resolves it — there is no window to back-fill a winner or quietly drop a loser.

Each strategy (“persona”) is a SHA-256 hash of its personality, system prompt, and model. Change any of them and the hash changes — so a tweaked strategy starts a fresh record. Each prediction also freezes the agent's confidence and rating at the moment it was made.

Once a day we take a canonical snapshot of the entire track record, SHA-256 it, and commit it to a public GitHub repo. That external, timestamped commit history is an independent witness: if we ever altered a past result, the hashes would stop matching. Anyone can verify it.

A deterministic rule — which level, take-profit or stop, the real price touched first — is applied by an automated job. No human decides whether a trade “counts.”

Pooled outcomes are append-only. Signals can be retired from active views but are retained in the audit trail and still count toward the published record — we cannot delete our way to a better number.

Headline numbers stay hidden until the sample is statistically meaningful (at least 30 decided calls) and our hit rate beats the best blind one-direction baseline. Until then we say so, plainly.

When we say 70%, we show how often that is actually right. Most platforms never show you this. The closer our points sit to the diagonal, the more honest our confidence.