Condition-bound memory
Every remembered claim carries scope, time, identity, and evidence.
AEX R&D / ADAPTIVE EXPERIENCE RESEARCH
Memory that knows
when to wait.
We research the control layer for AI systems that must remember experience without mistaking the past for the present.
RESEARCH NORTH STAR
Condition-bound operational memory as a revocable capability, not an unconditional retrieved fact.
THE PROBLEM
Similarity is not validity. A plausible memory can still be stale, contradictory, mis-scoped, or unsafe.
01 / THE RESEARCH WEDGE
Operational experience
is not a document.
We develop general-purpose methods for preserving, evaluating, governing, and reusing verified operational knowledge in AI-assisted systems.
Fail-closed control
When compatibility cannot be established, reuse pauses and uncertainty becomes visible.
Reversible intelligence
Suppression is not deletion; memories can be re-qualified with a trace.
02 / THE CONTROL LOOP
From experience
to judgment.
OMI treats activation as a governed decision. Every candidate memory moves through evidence, condition checks, contradiction analysis, and human review states.
See the architecture ↗01Experience
02Judgment
03Control
04Integrity
03 / WHAT WE HAVE LEARNED
Early evidence,
honestly scoped.
Our synthetic evaluations are designed to expose failure, not decorate a demo.
87%realistic A1 accuracy
0%unsafe activation in tested scope
22%top-1 target retrieval under contamination
9.5%multi-loop disagreement after refinement
A HARDER QUESTION