Ontology

An ontology is a formal, explicit specification of the concepts in a domain and the relationships that can hold between them. Where a knowledge graph stores concrete facts about specific entities, an ontology defines the schema behind those facts: the classes of thing that exist (Miner, Hashboard, ASIC, Firmware), the properties they can have, and the constraints and rules that govern valid statements. In short, the ontology is the blueprint; the knowledge graph is the building. The term is borrowed from philosophy, where ontology is the study of what exists — computer science narrowed it to mean an engineered, machine-readable account of what exists in one domain.

Classes, properties, and axioms

A well-formed ontology declares three kinds of structure. First, a hierarchy of classes: an Antminer is a kind of ASIC miner, which is a kind of mining hardware. Second, properties that relate classes or attach values to them: a hashboard belongs to a miner, carries some number of chips, draws some wattage. Third, axioms and constraints that police validity: every hashboard belongs to exactly one miner; chips-per-board is a positive integer; a firmware targets a control-board platform. On the Semantic Web these declarations are commonly expressed in RDFS or OWL, and because they are machine-readable, software can perform automated reasoning: inferring facts nobody typed in (if an S19 is an Antminer and Antminers are ASIC miners, an S19 is an ASIC miner), detecting contradictions before they corrupt a dataset, and merging data published by different organisations without terminological ambiguity.

Ontology versus taxonomy versus schema

A taxonomy is just the class hierarchy — a tree of is-a relationships. An ontology is strictly richer: it adds arbitrary relationships between branches of the tree, cardinality rules, and logical axioms. A database schema is closer in spirit, but a schema constrains one application's tables, while an ontology aims to capture the domain itself so that any number of systems can share it. That shareability is the point: two teams who adopt the same ontology can exchange data with no mapping layer, because "domain", "board", and "chip" already mean the same thing to both.

Why sovereign reference work uses one

Building a consistent, citable reference for mining hardware depends on a shared vocabulary with fixed meanings. Without one, "chip" drifts between meaning a die, a package, and a board position; "domain" gets confused with a hostname; and structured data emitted across thousands of pages silently contradicts itself. An ontology pins these terms down — for instance, that voltage regulation on a hashboard is a property of a domain (a group of chips sharing a DC-DC converter), never of an individual chip — so that every page, dataset, and API response drawn from the same graph stays mutually consistent and machine-verifiable. The same discipline is what lets search engines and AI systems consume a site's structured data with confidence rather than guesswork.

Ontologies are built incrementally, not delivered from on high. The usual craft sequence is: enumerate the core classes from real questions users ask ("which miners use this chip?"), add only the relationships those questions require, then tighten constraints as contradictions surface in the data. Over-modelling is the classic failure — an ontology with hundreds of speculative classes nobody populates is worse than a small one that is actually enforced. The test of a good ontology is boring reliability: new facts slot in without debate about what they mean, and queries written last year keep returning correct answers as the graph grows underneath them.

Ontologies provide the semantic backbone for knowledge graphs and feed downstream systems such as semantic search and retrieval pipelines over vector databases. They are conceptually distinct from the statistical modelling inside a transformer, which learns fuzzy patterns from text rather than declared rules — the two are complementary, with the ontology supplying the hard guarantees a language model cannot.