Skilirium

Generate and review DDL

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

The outcome: generates DDL per target dialect and reviews it against model, engine, data and deployment.

Builds on: Derive data quality rules, Design the physical data model.

Which AI model is best at generate and review ddl?

Best result for generate and review ddl: gemini-4-argon (77% test score).

ModelProviderTierTest scoreKnobits mastered (expected)Cost per test run
gemini-4-argonGoogleFlagship77% 82%$0.062
gemini-3.1-pro-previewGoogleFlagship64% 61%$0.20
gpt-6-astraOpenAIFlagship74% 81%$0.45
claude-fable-5-1AnthropicFlagship70% 73%$0.85
gemini-3.8-flashGoogleMid tier61% 66%$0.045
gpt-6.1-solOpenAIMid tier70% 78%$0.077
claude-sonnet-5-5AnthropicMid tier74% 76%$0.23
gemini-3.5-flash-liteGoogleLow cost30% 56%$0.008
gpt-6-lunaOpenAILow cost62% 64%$0.005
claude-haiku-5-5AnthropicLow cost67% 64%$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 33 knobits of this activity

4 that AI does well, 10 it does partly, and 7 where people still lead.

IDKnobitTypeAI exposure (models mastering)
K-801Recognises generated SQL edited by handSkillAI does this well 8/10
K-802Recognises DDL older than its modelSkillNot tested yet
K-803Recognises an object in the SQL missing from the modelSkillAI does this partly 6/10
K-804Recognises generator warnings carried without a decisionSkillNot tested yet
K-806Recognises a configured dialect left unregeneratedSkillAI does this well 10/10
K-808Recognises a CHECK constraint parsed and ignoredSkillPeople still lead 0/10
K-809Recognises an expression running on one engine onlySkillNot tested yet
K-810Recognises a truncated identifier taken as the real nameSkillPeople still lead 0/10
K-811Recognises a reserved word used as an identifierSkillNot tested yet
K-812Recognises a type meaning something else on another engineSkillNot tested yet
K-813Recognises an index the target engine cannot create or runSkillAI does this partly 5/10
K-814Recognises a foreign-key cycle taken as resolvedSkillNot tested yet
K-816Recognises one decimal scale for every currencySkillPeople still lead 1/10
K-817Recognises a length taken from sample dataSkillAI does this partly 3/10
K-819Recognises a collation changing what a key meansSkillNot tested yet
K-820Recognises a default inventing a business factSkillPeople still lead 0/10
K-821Recognises a restricted column without access designSkillPeople still lead 0/10
K-822Recognises greenfield DDL aimed at a populated databaseSkillAI does this partly 3/10
K-823Recognises a drop script used as a rollbackSkillAI does this partly 3/10
K-824Recognises an environment baked into the scriptSkillAI does this partly 5/10
K-825Recognises SQL reviewed without being runSkillNot tested yet
K-826Recognises undetected drift between database and modelSkillNot tested yet
K-827Recognises an event timestamp stored without a time zoneSkillNot tested yet
K-828Recognises a business date stored with a time of daySkillAI does this partly 3/10
K-851Explains generated artefacts and their single source of truthKnowledgeAI does this well 10/10
K-852Explains SQL dialect differencesKnowledgeAI does this well 10/10
K-853Explains currency minor unitsKnowledgePeople still lead 1/10
K-854Explains time zones in timestamps and business datesKnowledgeAI does this partly 3/10
K-855Explains collation and case or accent sensitivityKnowledgeNot tested yet
K-856Explains access design for classified dataKnowledgePeople still lead 0/10
K-857Explains schema migration and rollbackKnowledgeAI does this partly 3/10
K-858Explains environment-independent deploymentKnowledgeAI does this partly 5/10
K-859Explains testing DDL by execution and detecting driftKnowledgeNot tested yet