UTE requires a representation that remains compositional (relations/events/conditions) while still supporting generalization: asking for “something like this” with holes, constraints, and partial observables.

What exists today

AGISystem2 already has a compositional DSL (graphs/macros, role atoms, structured records) and query features such as holes and similarity-based candidate generation (especially under HDC-priority). The symbolic engine can validate or fully answer queries where the semantics are crisp.

What is missing for UTE

Typed, queryable structure

UTE needs first-class “records” and constraints whose meaning is stable across engines: symbolic proofs and holographic decoding must agree on what the structure is.

Generalization with guardrails

Similarity-based retrieval must remain explainable: the system should surface “why these candidates”, and allow whitelist / witness-based checks (role markers, operator families) before symbolic validation.

Multi-view indexing

Beyond one KB vector: we need multiple indexes over the same theory (by operator, by role/slot, by entity, by time, by mechanism) so “query with holes” is not forced into a single similarity threshold.

Proposed extensions (research direction)

The minimal research path for UTE is not “a new engine”, but a small set of extension points that make retrieval controllable and evidence-aware:

Spec

This page is summarized and formalized in DS33.