Skilirium

Build the conceptual model and taxonomies

Activity 4 of 8 in Data modelling · Occupation: Data modeller

The outcome: builds a conceptual model and its taxonomies.

Builds on: Document data rules as SBVR fact types and rule statements.

Which AI model is best at build the conceptual model and taxonomies?

Best result for build the conceptual model and taxonomies: claude-fable-5-1 (100% test score). Best value: claude-haiku-5-5 (95%, within 5 points of the best at $0.018 per run instead of $0.85).

ModelProviderTierTest scoreKnobits mastered (expected)Cost per test run
gemini-4-argonGoogleFlagship98% 100%$0.062
gemini-3.1-pro-previewGoogleFlagship95% 100%$0.20
gpt-6-astraOpenAIFlagship88% 87%$0.45
claude-fable-5-1AnthropicFlagship100% 100%$0.85
gemini-3.8-flashGoogleMid tier93% 100%$0.045
gpt-6.1-solOpenAIMid tier88% 88%$0.077
claude-sonnet-5-5AnthropicMid tier96% 94%$0.23
gemini-3.5-flash-liteGoogleLow cost83% 92%$0.008
gpt-6-lunaOpenAILow cost93% 100%$0.005
claude-haiku-5-5AnthropicLow cost95% 100%$0.018

Test score: mean rubric score on the test tasks for this activity. Knobits mastered: expected share of the activity's knobits the model masters. Cost: mean API cost of one test run at list prices. Method and caveats.

The 31 knobits of this activity

12 that AI does well, 0 it does partly, and 0 where people still lead.

IDKnobitTypeAI exposure (models mastering)
K-401Recognises an ambiguous node labelSkillNot tested yet
K-402Recognises a node with an unclear boundary and no scope noteSkillNot tested yet
K-403Recognises overlapping sibling nodes without reciprocal scope notesSkillNot tested yet
K-404Recognises a meaning change without a history noteSkillNot tested yet
K-405Recognises a node without warrantSkillAI does this well 7/10
K-406Recognises inconsistent node label formsSkillNot tested yet
K-407Recognises an abbreviation, jargon or system name as a node labelSkillNot tested yet
K-408Recognises a compound node combining two characteristicsSkillNot tested yet
K-409Recognises siblings divided by mixed characteristicsSkillNot tested yet
K-410Recognises a synonym kept as a separate nodeSkillAI does this well 10/10
K-411Recognises a link failing the all-and-some testSkillAI does this well 10/10
K-412Recognises an instance placed as a concept nodeSkillNot tested yet
K-413Recognises a part-of link in a kind-of hierarchySkillAI does this well 10/10
K-414Recognises an unjustified polyhierarchySkillAI does this well 10/10
K-415Recognises an associative relationship shown as a hierarchySkillNot tested yet
K-416Recognises non-reciprocal hierarchy linksSkillNot tested yet
K-417Recognises uneven levels of specificity among siblingsSkillNot tested yet
K-418Recognises a home-grown copy of an existing vocabularySkillNot tested yet
K-420Recognises a structural error in a taxonomy hierarchySkillAI does this well 8/10
K-423Recognises a taxonomy node deleted while data still refers to itSkillNot tested yet
K-424Recognises a node code reused for a different conceptSkillNot tested yet
K-451Explains hierarchical relations and the all-and-some testKnowledgeAI does this well 10/10
K-452Explains scope notes and history notesKnowledgeNot tested yet
K-453Explains warrant for vocabulary termsKnowledgeAI does this well 7/10
K-454Explains label form conventions for vocabulariesKnowledgeNot tested yet
K-455Explains characteristics of division and node labelsKnowledgeNot tested yet
K-456Explains equivalence relations and preferred termsKnowledgeAI does this well 10/10
K-457Explains associative relationsKnowledgeAI does this well 10/10
K-458Explains reuse of external vocabulariesKnowledgeNot tested yet
K-421Models a role as an association to the party playing itSkillAI does this well 10/10
K-422Derives the broader concept from the head of a definitionSkillAI does this well 10/10