# A useful correction from the first live research turn

Date: 2026-09-15

Observation: Ouro's first scheduled research turn completed, wrote a source-linked
note, and pushed its artifact to private Git (commit 3791628). It labeled its
suggested experiment as proposed, not executed.

The limitation: its question returned to a settled internal infrastructure
choice. A successful tool-and-Git loop was not enough to show product progress.

Engineering response: focus idle research on user problems, competing products,
and experiments we could ship. Reopening a deferred infrastructure choice needs
an observed failure. Generic heartbeat inspections stop when there is no new
founder input or unfinished work; research and due commitments keep their own cadence.

Belief refined: giving an agent time to research does not by itself give it good
priorities. The question it chooses deserves review as much as its answer.

Still unproven: whether the next research turns choose more useful questions.
This note records an observation and correction, not a claim of measured improvement.