Changelog.

Research rarely ends with the first successful experiment. A result raises another question; a failure can expose a bad assumption. We want Optimus to carry an investigation across those turns, maintaining the objective and the evidence behind its decisions as the work continues.

We’re working toward longer autonomous investigations in which Optimus can inspect a failed run, revise its approach, and choose a useful next experiment. That also means recognizing when the evidence is insufficient, when further runs are unlikely to help, and when a researcher should make the decision. Researchers should be able to follow the reasoning, set limits, and redirect the work without rebuilding its context.

Trying more candidates helps only if they test different ideas. We want Optimus to investigate alternatives that change the assumptions of an experiment, the learning algorithm, or the information available to a learner. Small variations around one familiar approach can leave a better direction unexplored.

Our research focuses on how to allocate effort across these alternatives: which hypotheses deserve an initial test, which results justify a deeper investigation, and when to return to a direction that looked unpromising. Comparisons need consistent evaluation conditions so that an apparent gain reflects a better system. The aim is to explore several promising directions while preserving what has already worked.

An idea from one field can suggest a different way to approach a problem in another. We want Optimus to identify connections between mechanisms, such as how a system explores under uncertainty or learns from delayed feedback, and make the assumptions behind those connections explicit.

A convincing analogy still needs evidence. We’re investigating how Optimus can turn a proposed connection into a small, discriminating experiment, compare it with an appropriate baseline, and identify where the transfer breaks down. Success would mean finding ideas that remain useful after testing in the new domain.

We’re evolving Arkenos toward a richer workspace experience, with clearer ways to move between project files, experiments, running sessions, and results. You should be able to return to an investigation and understand what changed, what is still running, and which decisions need attention.

Cross-team collaboration and multiplayer are part of that direction. We plan to let colleagues work together in a shared project context, review results, and hand off an investigation with its history intact. People and agents will need clear ownership of changes and controls over who can view, edit, or execute work. We’re designing these interactions so teams can coordinate longer research efforts while keeping individual contributions and decisions visible.