04 APPLICATIONS

Where could this research
be tested next?

These are research directions and prototypes, not a catalogue of finished products. Each field states what exists today and what still needs to be validated.

STATUS LANGUAGE

Possibility and readiness are different claims.

Status is assigned from current evidence and can move backward when a counterexample appears.

RESEARCHQuestion and experiments exist

PROTOTYPEAn experimental implementation exists

VALIDATEDHeld under named conditions

PRODUCTAvailable for sustained real use

PROTOTYPE

Regulation Intelligence

Search and answer support for regulations, guides, tables, and referenced clauses.

Evidence today: A working regulation-search prototype, dual original-text and natural-anchor retrieval paths, and answer-experience reuse experiments.
Still required: Broader accuracy evaluation, source-update governance, access control, and sustained user validation.

PROTOTYPE

Experience-Based Retrieval

A retrieval path that can remember how differently worded questions reached the same grounded answer.

Evidence today: Question decomposition, natural-language anchors, answer-centered experience grouping, and measured complementarity tests.
Still required: Cross-domain reuse, stable precision boundaries, and comparison against simpler retrieval baselines.

RESEARCH

Intelligent NPC & Living World

Characters and worlds whose future behavior is affected by remembered interaction.

Evidence today: Bounded memory-owned behavior and world-observation experiments.
Still required: Long-horizon consistency, autonomous structure formation, safety boundaries, and evaluation in an actual game loop.

RESEARCH

Experience-Based Agent

An agent that preserves actual observations, actions, and outcomes across tasks and restarts.

Evidence today: One external observation reuse loop and exact post-restart action acknowledgement reconciliation.
Still required: Unfamiliar-goal transfer, negative outcome handling, authorization boundaries, and real workflow evaluation.

RESEARCH

Long-Term Relational AI

The original horizon: an intelligence and a person who remember and affect one another over time.

Evidence today: Architecture and component experiments only.
Still required: Nearly every system-level claim: continuity, identity, consent, correction, mutual influence, and long-term evaluation.

FOCUS

Do not turn every possibility into a product.

MIRA will use concrete applications to test the research. A field moves forward when it exposes a clear user problem and a measurable experiment—not because it sounds like a large market.

Discuss a technical collaboration