MOMENTS RESEARCH · 05
Governed Recursive Self-Improvement
A long-term research direction for improving parts of an agent’s reasoning, tools and learning loops under verifiable objectives and continuous oversight.
Long-term research direction
Questions guiding our workROLE IN THE MOMENTS LOOP
The frontier beyond the operating loop: recursive improvement remains bounded by evaluation, sandboxed execution and rollback.
Text · Image · Audio · Video · Behavior
- 01Perceive
- 02Model
- 03Plan & Act
- 04Assess
- 05Evolve
Outcomes · Capability · Uncertainty · Safety
OPEN QUESTIONS
What we are trying to understand.
- 01
Which parts of a system can be improved while keeping the objective stable and verifiable?
- 02
How can recursive changes be isolated, audited and compared with a known baseline?
- 03
What forms of human oversight remain effective as iteration accelerates?
RESEARCH BOUNDARIES
How the work stays grounded.
- RSI is a research direction, not a product claim.
- Every proposed change must remain observable, testable and reversible.
- Sandboxing, independent evaluation and human rollback are foundational constraints.
PUBLIC RECORD
Methods, results and lessons from the work.
Our research notes will explain what we explored, what the evidence shows and what remains open.
No publications yet
Explore the research questions above while we prepare work to share.