RISGRA® Hyper-graph

Concepts & capabilities

Schema count as at: March 2025

Hyper-Relations: 1,842

Entities: 489

Attributes: 223

Rules: 642

  • RISGRA® – is a “Symbolic” (foundational) hyper-graph model.

    Symbolic models bring real-world “Judgement”, to counter-balance & complement the “Predictive” power of Neural AI and Agentic AI models (LLM + ML).

    • Regulated organisations are grappling with the challenge of how to capitalise on the immense Predictive power of Neural AI and Agentic AI models, to create sustainable Enterprise Value;

    • Concurrently there is a realisation that integration into operations of “black box” AI capability will fail to meet multiple regulatory hurdles, which require complex, non-linear decision processes to be explainable, repeatable and demonstrably fair to customers;

    • Symbolic hyper-graph models (Ontologies) such as RISGRA® are foundational representations of real-world, curated knowledge that counter-balance and provide the guard rails necessary to enable the controlled implementation and compliant adoption of Neural AI and Agentic AI [ = Neuro-Symbolic AI ];

    • Organisations will need to rapidly develop bespoke, proprietary Neuro-Symbolic AI capabilities that are internally governed & controlled within the corporate perimeter;

    • RISGRA® models are purpose-built to support this fast moving transition.

  • Integral to RISGRA® graph-schema are constructs that support the granular mapping (Bayesian Directed Acyclic Graph (DAG)) of risk transmission pathways, from: horizon risks; through crystallisation of risk incidents; to networked impacts; and extending to long-tail effects.

  • Many practitioners acknowledge that to-date, time as a dimension has been largely absent from graph languages and has been relatively unsophisticated in risk modelling more generally.

    In collaboration with a niche graph partner spun out from the Alan Turing Institute, RISGRA® supports the integration of temporal data, enabling the modelling of: tipping points; cascade effects; temporary points of equilibrium and more.

  • At the heart of RISGRA® is a comprehensive schema (Ontology) - a coded database design that describes all RISGRA® concepts and constructs, and the many nuanced relationships between them.

    Schema is “good medicine” in any database, it:

    • ensures that the model is coherent and has design integrity

    • provides guard rails that ensure quality, reliability and that outputs are explainable

    • provides the foundation for inference and deductive reasoning, which unveils hidden risks and opportunities by linking seemingly unrelated data points.

  • RISGRA® scope includes both Ontology (for which read “Schema”, and for that please see Schema section above) and Taxonomy.

    Taxonomy - RISGRA® includes a representative Risk Universe that comprises c.250 (and growing) distinct risk types, including classifications, definitions, and c.44,000 (theoretical maximum) causal inter-relationships (risk transmission pathways).

    Building on the baseline risk taxonomy, multiple sector-specific variants of Risk Universe can be developed, for example an Aviation Risk Universe is currently under development.

  • Traditional databases are transactional, focusing on storing discrete data points. Knowledge “property” graphs rely on RDF triples (subject-predicate-object) to represent relationships between things.

    RDF triples lack power and expressivity, often requiring “reification” - the act of artificially turning relationships (edges) into entities - in an attempt to represent more complex structures. This compromise undermines the integrity of the risk model and restricts its power and analytical depth.

    RISGRA® capitalizes on a new, highly expressive hyper-graph language, that naturally supports modelling the complex and nuanced interfaces in the Enterprise Risk domain within a regulated organisation, and along it’s value chain.

  • Hyper-relations are integral to the design of RISGRA® - where Relations can participate in other Relations.

    This enables the accurate modelling of:

    • the complex, nuanced, multi-layered relationships & dependencies between risk management constructs;

    • non-linear risk transmission pathways across the whole value-chain.

  • Many longer-established graph platforms not only persist with RDF triples at their core, but also carry legacy back-end technology and associated performance penalty.

    RISGRA® partners with contemporary graph platforms that are optimised to ingest and analyse data orders-of-magnitude faster than the long-established graph products.

    For example, throughput at 4.6 million updates a minute (c.77,000 per second), compare with averages closer to 6 updates per second for legacy platforms (=> Yes, that’s c.13,000x faster!)

    This performance advantage is further underpinned by reduced time taken for messaging during analysis by over 10x.

  • Traditional database models, and many property graph models lack flexibility and are expensive to update.

    In contrast, the design characteristics of the hyper-graph language used for RISGRA® ensures that making changes to RISGRA® schema is a relatively trivial exercise.

    This a key strategic advantage when developing the model through successive releases, or when creating bespoke versions of RISGRA® for end-user organisations.