Back to blog
6 min readPremium

What our engine is no longer allowed to decide on its own

methodologyproofenginetransparency

An engine that learns from its results looks virtuous. It makes a mistake, notices it, corrects itself. But there are two ways of learning. Adjusting a parameter from its results, under conditions written in advance, is learning. Changing its shape (which filters block, which rules are allowed to speak, which way they bet) from nothing but the decisions it produced itself is grading its own homework.

On 26 September, while examining another automatic setting of the same kind, we found seven nightly settings that were still changing the engine's shape from live results, against a rule written on 2 August. We switched them off. Here is what they did, what they had actually done, and what live results are still allowed to tune, because some things remain.

GeoPulse

Follow the markets with GeoPulse

GeoPulse correlates geopolitical events with financial markets using AI analysis of every event.

Create a free account