Research

Research-first direction, architecture publication, and governance references for Elora's non-neural behaviour detection, commit-boundary control, and evidence-rich operator model.

Current Research Program

The current research program focuses on observing and analysing AI inference behaviour under constrained compute conditions, with the aim of understanding how instability, drift, and inefficient generation emerge during runtime.

Initial work explores how non-neural machine learning and bounded governance regulation applied prior to generation can influence these behaviours, particularly in relation to token usage, runtime characteristics, and overall stability.

This research has been formalised in the following preprint.

AI Inference Behaviour Under Constrained Compute: Observability and Control in Resource-Limited Systems

Read Preprint

Technical Disclosure Governance Model Pack Architecture Layers Ecosystem Collaboration Guided Research Report

Open Standards Direction

Elora favors transparent architectural publication and interoperability over closed governance primitives. The project publishes governance and security architecture as references while keeping sensitive runtime implementation contracts private for operational security.

Read Cryptographic Foundations

Research Areas

Core research and disclosure materials are grouped below for fast access across behaviour detection, governance control, Model Exams, and capability expansion research.

Elora Learning System (ERIS)

How Elora Runtime Intelligence System (ERIS) uses bounded non-neural/classical ML and statistical methods to learn runtime/system behaviour, pressure signatures, and degradation posture for read-only runtime-intelligence visibility under policy-bound governance controls.

Open Learning System

Elora Behavioural Learning System (EBLS) In Development

EBLS is the model-behaviour learning domain: suitability by task/profile, governance-adherence scoring, and pattern outcomes from observer/model-exam evidence. It remains policy-gated and does not bypass commit-boundary authority.

See Governance Boundary

Technical Disclosure

Dated milestones, standards baseline, terminology, and capability chronology for defensive publication.

Open Technical Disclosure

Governance Model

Control-plane semantics, commit boundary admissibility, operator risk workflow, and behaviour-governance mapping.

Open Governance Model Pack

Architecture Layers

Layered responsibility model from ingress trust to behaviour detection, evidence capture, and replay.

Open Architecture Layers

Commit Boundary Model

Proposal to commit to execution as the core authorization design of Elora's governance runtime.

Open Architecture Narrative Foundations Section

Prior Art Position

Public chronology of implementation milestones and evidence method for open architecture protection.

Open Prior Art Position

Ecosystem Collaboration

Documented project-to-project dialogue and architecture pattern exploration with explicit non-ownership framing.

Open Collaboration Page

Publications

Peer-reviewed and preprint outputs, including collaboration-linked research contributions.

Open Publications

Upcoming Research Project

Interactive Control Study: The Wrestling Game

Planned research explores using a model as Elora's "hands" to execute game actions while Elora evaluates instruction-following, timing, and policy-safe behaviour under dynamic pressure, as part of capability research beyond standard inference control planes.

This will run as a controlled program with safe telemetry boundaries, deterministic scoring, and reproducible scenario playback.