Evidence over claims.

We publish failures as readily as successes. BALANCED mode is off in v0 because it didn't clear its own bar — not because it's convenient to hide.

We don't sell before we've proven. A policy reaches production only after a frozen, pre-registered, prospective test clears its GO threshold.

We separate benchmarks from production. The public ledger counts real outcomes only — shadow runs and tests are excluded.

We keep the scope small. Better one category measured honestly than ten claimed loosely.