Curiosity has to earn its place.

An attractive mechanism meets its controls

An isolated digital-biology curiosity model improved stable synthetic prediction by 7.97%. But reversal performance worsened by 6.24%, and pure-noise prediction worsened by 2.17%.

These results are not contradictory. A mechanism can exploit a stable pattern while adapting poorly when that pattern changes or when there is no meaningful pattern to learn.

A useful rejection

A follow-up adaptation candidate passed four of six promotion gates and was rejected. The research therefore did not justify integrating it into live Owen.

This is part of the project’s development, not an embarrassing footnote. A nature-inspired idea needs a measurable benefit against appropriate baselines. Keeping the failed controls in the same account as the improvement prevents a selective success story.

The experiment was standalone and synthetic. Its percentages do not measure a general increase or decrease in Owen’s intelligence.