Skilirium

Adaptive skill diagnosis · pilot

Know which parts of a job you can already do.

Skilirium is an adaptive test. It maps what you know against the knowledge and skills an occupation needs, one small unit at a time, and shows you what to learn next.

Add Skilirium to Claude or another MCP app with this address

https://skilirium.com/mcp

You sign in with Google. Your results are kept under a pseudonymous learner ID; your name and email are stored separately and never with your results.

K-001 Recognises a modelling scope with no boundary: Not yet, where people leadK-002 Recognises a modelling standard named but not chosen: Not yetK-003 Recognises a model layer planned with no notation: Not yetK-004 Recognises model inputs the plan does not locate: Not yetK-005 Recognises a model review planned by its modeller: Not yet, where people leadK-006 Recognises a model with no repository or baseline: MasteredK-031 Plans the deliverables of a modelling assignment: Not yet, where people leadK-032 Chooses a modelling scheme and notation for an audience and purpose: Not yet, where people leadK-033 Sets the modelling standards a project follows: Not yet, where people leadK-034 Selects the tools and repository for a modelling project: Mastered K-101 Recognises a data requirement making several claims: Not yetK-102 Recognises a data requirement prescribing a design solution: Not yetK-103 Recognises a statement outside the scope of data requirements: Mastered, where people leadK-104 Recognises an ambiguous business term in a data requirement: MasteredK-105 Recognises untestable wording in a data requirement: Not yetK-106 Recognises an unverifiable data requirement: Not yetK-107 Recognises a reference with several possible referents: Not yetK-108 Recognises a data requirement without a named subject: Not yetK-109 Recognises a measure stated without unit or basis: Not yetK-110 Recognises a data requirement silent on current versus past values: Not yetK-111 Recognises an absolute statement with its exception unasked: Not yetK-112 Recognises a mandatory attribute without its moment of obligation: Not yetK-113 Recognises a business rule left inside another statement: Not yetK-114 Recognises content filed under the wrong requirement: Not yetK-115 Recognises an unsourced data requirement: MasteredK-116 Recognises a missing or meaningless requirement priority: Not yetK-117 Recognises a missing or wrong requirement type: Not yet, where people leadK-118 Recognises a status contradicting the requirement's content: MasteredK-119 Recognises a guess recorded as fact: MasteredK-120 Recognises a missing rationale for a non-obvious requirement: Not yet, where people leadK-121 Recognises duplicate data requirements: Not yetK-122 Recognises paraphrased data requirements: Not yetK-123 Recognises conflicting data requirements: Not yetK-124 Recognises inconsistent terminology across the register: MasteredK-125 Recognises a requirements register with unbounded scope: MasteredK-126 Recognises a requirements register describing no events: Not yetK-127 Recognises an uncovered identity requirement: Not yetK-128 Recognises an uncovered volume or retention requirement: Not yetK-129 Recognises an uncovered sensitivity or access requirement: Not yetK-130 Recognises an uncovered lineage requirement: Not yetK-131 Recognises undecided history requirements across the register: Not yetK-132 Recognises a requirements register serving no stated question: Not yetK-133 Recognises a missing or wrong requirement trace: MasteredK-134 Recognises unstable requirement identifiers: MasteredK-151 Explains singular and unambiguous requirement statements: MasteredK-152 Distinguishes a business requirement from a design solution: MasteredK-153 Explains the data requirement types of the register: MasteredK-154 Explains requirement attributes and traceability: MasteredK-155 Explains current value, history and point-in-time reproducibility: MasteredK-156 Explains the identity of a business thing: MasteredK-157 Explains the non-functional data concerns: MasteredK-158 Explains the scope, events and questions a data register must state: MasteredK-159 Explains referential and logical ambiguity in natural-language statements: MasteredK-160 Explains persistent identifiers: MasteredK-161 Explains verifiable requirement statements: MasteredK-162 Explains consistency within a requirements register: MasteredK-355 Explains Witt's classification of business rules: MasteredK-135 Retires a superseded register entry instead of deleting it: Mastered K-401 Recognises an ambiguous node label: Not yet, where people leadK-402 Recognises a node with an unclear boundary and no scope note: Not yet, where people leadK-403 Recognises overlapping sibling nodes without reciprocal scope notes: Not yet, where people leadK-404 Recognises a meaning change without a history note: Not yet, where people leadK-405 Recognises a node without warrant: Not yetK-406 Recognises inconsistent node label forms: Not yetK-407 Recognises an abbreviation, jargon or system name as a node label: Not yet, where people leadK-408 Recognises a compound node combining two characteristics: Not yet, where people leadK-409 Recognises siblings divided by mixed characteristics: Not yet, where people leadK-410 Recognises a synonym kept as a separate node: Not yetK-411 Recognises a link failing the all-and-some test: MasteredK-412 Recognises an instance placed as a concept node: Not yet, where people leadK-413 Recognises a part-of link in a kind-of hierarchy: Not yetK-414 Recognises an unjustified polyhierarchy: Not yetK-415 Recognises an associative relationship shown as a hierarchy: Not yet, where people leadK-416 Recognises non-reciprocal hierarchy links: MasteredK-417 Recognises uneven levels of specificity among siblings: Not yet, where people leadK-418 Recognises a home-grown copy of an existing vocabulary: Not yet, where people leadK-420 Recognises a structural error in a taxonomy hierarchy: Not yet, where people leadK-423 Recognises a taxonomy node deleted while data still refers to it: MasteredK-424 Recognises a node code reused for a different concept: Not yetK-451 Explains hierarchical relations and the all-and-some test: MasteredK-452 Explains scope notes and history notes: MasteredK-453 Explains warrant for vocabulary terms: MasteredK-454 Explains label form conventions for vocabularies: MasteredK-455 Explains characteristics of division and node labels: MasteredK-456 Explains equivalence relations and preferred terms: MasteredK-457 Explains associative relations: MasteredK-458 Explains reuse of external vocabularies: MasteredK-421 Models a role as an association to the party playing it: MasteredK-422 Derives the broader concept from the head of a definition: Mastered K-501 Recognises an insertion anomaly in a logical data model: Not yetK-502 Recognises an update anomaly in a logical data model: MasteredK-503 Recognises a deletion anomaly in a logical data model: Not yetK-504 Recognises a repeating group in an entity: Not yetK-505 Recognises an attribute holding several facts: MasteredK-506 Recognises an attribute with several meanings: MasteredK-507 Recognises attributes held as name-value rows: Not yetK-508 Recognises a partial dependency on a composite key: Not yet, where people leadK-509 Recognises a transitive dependency between non-key attributes: MasteredK-510 Recognises a stored derivable attribute: MasteredK-511 Recognises a non-superkey determinant in a 3NF entity: Not yet, where people leadK-512 Recognises independent multi-valued facts in one entity: Not yetK-513 Recognises a join dependency kept in one entity: MasteredK-514 Recognises a lossy decomposition: MasteredK-515 Recognises a functional dependency lost in a decomposition: Mastered, where people leadK-516 Recognises attributes inapplicable to some instances: MasteredK-517 Recognises an unstable or meaningful primary key: MasteredK-518 Recognises an entity identified only by a system-assigned key: MasteredK-519 Recognises undeclared subtype constraints: MasteredK-520 Recognises independently changing histories in one entity: Not yet, where people leadK-521 Recognises an unresolved many-to-many relationship: Not yetK-522 Recognises a polymorphic association: MasteredK-523 Recognises a connection trap: MasteredK-524 Recognises a fan trap: MasteredK-525 Recognises a chasm trap: MasteredK-526 Recognises a redundant relationship: MasteredK-527 Recognises overlapping entities: Not yetK-528 Recognises one entity split by value into several: MasteredK-529 Recognises a relationship contradicting its data rule: MasteredK-530 Recognises a constraint implied by neither keys nor domains: Not yet, where people leadK-531 Traces a logical-model anomaly to its structural cause: Not yetK-532 Records the functional dependencies of an entity: Not yetK-533 Decomposes an entity along a functional dependency without loss: Not yet, where people leadK-534 Resolves a many-to-many relationship into an associative entity: Not yetK-535 Replaces a repeating group with a dependent entity: Not yetK-536 Introduces subtypes for attributes applying to only some instances: Not yetK-551 Explains functional dependency: MasteredK-552 Explains candidate key and prime attribute: MasteredK-553 Explains the insertion, update and deletion anomalies: MasteredK-554 Explains first normal form: MasteredK-555 Explains second normal form: MasteredK-556 Explains third normal form: MasteredK-557 Explains Boyce-Codd normal form: MasteredK-558 Explains multivalued dependency and fourth normal form: MasteredK-559 Explains join dependency and fifth normal form: MasteredK-560 Explains sixth normal form for temporal data: MasteredK-561 Explains domain-key normal form: MasteredK-562 Explains attribute domains: MasteredK-563 Explains lossless-join decomposition: MasteredK-564 Explains dependency preservation: MasteredK-565 Explains derived data and point-in-time snapshots: MasteredK-566 Explains supertype and subtype constraints: MasteredK-567 Explains relationship cardinality and optionality: MasteredK-568 Explains natural keys, surrogate keys and key stability: MasteredK-569 Explains the connection, fan and chasm traps: MasteredK-570 Explains an entity type as the realisation of one business concept: MasteredK-581 Expresses a logical model as a UML class diagram: Not yetK-582 Models data as fact types in a fact-based scheme: MasteredK-583 Maps an application model onto a canonical model: Not yet K-310 Recognises a constraint claimed by a rule and missing from the schema: Not yetK-313 Recognises a declared constraint the target platform does not enforce: MasteredK-601 Recognises an unmanaged duplicate column: MasteredK-602 Recognises logical entities merged into one table: Not yetK-603 Recognises a stored derived or summary value: MasteredK-604 Recognises a lossy or unconstrained vertical split: Not yet, where people leadK-605 Recognises an entity split into clone tables by value: Not yetK-606 Recognises rolled-up subtypes losing conditional constraints: Not yetK-607 Recognises rolled-down subtypes losing supertype constraints: Not yet, where people leadK-608 Recognises a hierarchy encoded redundantly: Not yetK-609 Recognises content held outside the database by reference: Not yetK-610 Recognises a surrogate key without natural-key uniqueness: Not yet, where people leadK-611 Recognises a surrogate value with meaning, reused or renumbered: Not yet, where people leadK-612 Recognises an undeclared foreign key: MasteredK-613 Recognises a referential action destroying what should survive: Not yetK-616 Recognises a changeable value list fixed in the DDL: Not yet, where people leadK-617 Recognises a physical domain differing from the logical domain: MasteredK-618 Recognises binary floating point used for an exact quantity: Not yetK-620 Recognises a business fact held in physical row order: Not yet, where people leadK-621 Recognises an index set not derived from access paths: Not yetK-622 Recognises an overwrite where history is required: MasteredK-623 Recognises inconsistent validity periods in a versioned table: MasteredK-624 Recognises a reference to a versioned table by business key alone: MasteredK-625 Recognises a soft delete the constraints do not understand: Not yet, where people leadK-626 Recognises a fact table of mixed grain: Not yet, where people leadK-627 Recognises a measure without stated additivity: Mastered, where people leadK-628 Recognises a nullable foreign key on a reporting join path: Not yet, where people leadK-705 Recognises history kept without its constraints checked: MasteredK-629 Recognises a sentinel value standing in for null: MasteredK-630 Recognises a null with a business meaning: MasteredK-651 Explains controlled denormalisation: MasteredK-652 Explains the implementation options for subtypes: MasteredK-653 Explains referential integrity and referential actions: MasteredK-654 Explains exact and approximate numeric types: MasteredK-655 Explains null semantics and three-valued logic: MasteredK-656 Explains value sets held as reference data: MasteredK-657 Explains index design from access paths: MasteredK-658 Explains history strategies and validity-period conventions: MasteredK-659 Explains grain and additivity in dimensional models: MasteredK-660 Explains vertical and horizontal partitioning of tables: MasteredK-661 Explains consistency of content stored by reference: MasteredK-662 Explains the absence of row order in relational tables: MasteredK-641 Routes a defect to the deliverable responsible for it: MasteredK-671 Designs a star schema for a stated grain: Not yetK-672 Designs a Data Vault model for auditable history: MasteredK-673 Designs a document structure for a stated access pattern: MasteredK-674 Designs views hiding physical structure from consumers: Not yet, where people lead K-801 Recognises generated SQL edited by hand: MasteredK-802 Recognises DDL older than its model: Not yetK-803 Recognises an object in the SQL missing from the model: Not yetK-804 Recognises generator warnings carried without a decision: Not yetK-806 Recognises a configured dialect left unregenerated: Not yetK-808 Recognises a CHECK constraint parsed and ignored: Not yetK-809 Recognises an expression running on one engine only: MasteredK-810 Recognises a truncated identifier taken as the real name: Not yetK-811 Recognises a reserved word used as an identifier: Not yet, where people leadK-812 Recognises a type meaning something else on another engine: Not yetK-813 Recognises an index the target engine cannot create or run: Not yetK-814 Recognises a foreign-key cycle taken as resolved: MasteredK-816 Recognises one decimal scale for every currency: Not yetK-817 Recognises a length taken from sample data: Not yetK-819 Recognises a collation changing what a key means: Not yet, where people leadK-820 Recognises a default inventing a business fact: Not yetK-821 Recognises a restricted column without access design: Not yetK-822 Recognises greenfield DDL aimed at a populated database: Not yetK-823 Recognises a drop script used as a rollback: Not yetK-824 Recognises an environment baked into the script: Not yetK-825 Recognises SQL reviewed without being run: Not yet, where people leadK-826 Recognises undetected drift between database and model: Not yetK-827 Recognises an event timestamp stored without a time zone: MasteredK-828 Recognises a business date stored with a time of day: Not yetK-356 Explains rule enforcement mechanisms and platform enforcement: MasteredK-851 Explains generated artefacts and their single source of truth: MasteredK-852 Explains SQL dialect differences: MasteredK-853 Explains currency minor units: MasteredK-854 Explains time zones in timestamps and business dates: MasteredK-855 Explains collation and case or accent sensitivity: MasteredK-856 Explains access design for classified data: MasteredK-857 Explains schema migration and rollback: MasteredK-858 Explains environment-independent deployment: MasteredK-859 Explains testing DDL by execution and detecting drift: Mastered K-1001 Recognises an incomplete data model scorecard: MasteredK-1002 Recognises a scorecard score with no evidence: Not yet, where people leadK-1003 Recognises a model reviewed by its modeller: Not yetK-1004 Recognises a review decision contradicting its total: Not yetK-1005 Recognises a review finding with no owner or action: Mastered, where people leadK-1006 Recognises a scorecard score contradicting the validator: Not yet, where people leadK-1007 Recognises a review of a version other than the one released: MasteredK-1031 Scores a data model against the Data Model Scorecard: Mastered K-1101 Recognises an inferred structure presented as declared: Not yet, where people leadK-1102 Recognises reverse-engineered objects not marked as-is: Not yetK-1103 Recognises a dead database object carried into the model: Not yetK-1104 Recognises a guessed meaning written as a definition: Not yet, where people leadK-1105 Recognises a reverse-engineering source with no environment or date: Not yet, where people leadK-1106 Recognises a physical workaround lifted into the logical model: Not yet, where people leadK-1131 Reverse-engineers a logical model from a database schema: Mastered K-1201 Recognises a model change with no reason: Not yetK-1202 Recognises a model change with no impact analysis: MasteredK-1203 Recognises a breaking model change released as a minor version: Not yetK-1204 Recognises a change to a deployed structure with no migration: Not yet, where people leadK-1205 Recognises a released model version with no baseline: MasteredK-1206 Recognises a model change released without approval: MasteredK-1231 Releases a model change as a new model version: Mastered Build the conceptual model and taxonomies builds on Capture data requirements Derive the logical data model builds on Build the conceptual model and taxonomies Design the physical data model builds on Derive the logical data model Generate and review DDL builds on Design the physical data model Review the data model against the scorecard builds on Design the physical data model Review the data model against the scorecard builds on Plan the modelling and set its standards Reverse-engineer a model from an existing database builds on Design the physical data model Maintain and version data models builds on Generate and review DDL Maintain and version data models builds on Review the data model against the scorecard Plan the modelling and set its standards: Mastered, 6 knobits, where people lead Capture data requirements: Mastered, 34 knobits Build the conceptual model and taxonomies: Mastered, 22 knobits, where people lead Derive the logical data model: Mastered, 30 knobits Design the physical data model: Mastered, 30 knobits, where people lead Generate and review DDL: Mastered, 28 knobits Review the data model against the scorecard: Mastered, 7 knobits, where people lead Reverse-engineer a model from an existing database: Mastered, 6 knobits, where people lead Maintain and version data models: Mastered, 6 knobits
  • Mastered by the AI model
  • Partly
  • Not yet
  • Where people lead
  • Not tested
  • Builds on (a prerequisite)
  • Both mastered
  • The activity’s knobits, each lit when mastered

Hover or tap a star to see the activity; what it builds on lights up. The small dots are its knobits: hover or tap one to see it.

The data modeller’s constellation: each star an activity, the faint stars its knobits, the lines what builds on what. Lit here by gemini-4-argon, the AI model with the best results for this occupation; gold is where people still lead. Yours appears after a test.

A job, broken into knobits

An occupation needs competences, a competence has stages of work, and every stage rests on knobits: the smallest pieces of knowledge or skill that make sense to learn and test on their own. Knobits build on each other, so knowing a few answers tells a lot about the rest.

The pilot covers the data modeller and the data modelling competence: 307 knobits based on DAMA-DMBOK2, in eight stages.

The eight stages of data modelling

  1. Capture atomic, sourced and testable data requirements
  2. Define business terms in a glossary
  3. Document data rules and resolve rule conflicts
  4. Build a conceptual model and its taxonomies
  5. Derive a normalised logical model
  6. Design a physical model for the target platform
  7. Derive data quality rules
  8. Generate and review DDL

How a test works

  1. ConnectAdd the address above as a connector in Claude or another app that supports MCP.
  2. Sign inSign in with your Google account when the app asks. Choose a stage to test.
  3. AnswerQuestions appear one at a time. Each answer decides the most useful next question. Choosing “I don’t know” is better than guessing.
  4. See your mapThe result shows which knobits you have mastered, which you are ready to learn next, and how far you cover each stage.

A stage takes 10 to about 100 questions, depending on its size and how clear your answers are.

The same test for AI models

An AI model can take exactly the same test through the app’s tools. People and models are measured on the same knobits, so you can see where a model can do the work and where it still needs a person.

Your data stays yours

Skilirium follows the MyData principles: you decide what happens with your results. They are stored in the EU, are never sold or used for advertising, and you can export or delete them at any time. The public pages use Google Analytics to count visits, which sets analytics cookies; the results pages behind sign-in run no analytics and set only one cookie that keeps you signed in.

Read the privacy policy

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data modelling tests. Of its 282 benchmarked knobits, 124 are mastered by most models, and in 48 people still lead.

See the AI benchmarks · Which parts of a data modeller's job AI can do

Top models in data modelling

  1. gemini-4-argon · 90% of the competence
  2. claude-haiku-5-5 · 81% of the competence
  3. gpt-6-luna · 80% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data governance tests. Of its 158 benchmarked knobits, 33 are mastered by most models, and in 8 people still lead.

See the AI benchmarks · Which parts of a data governance manager's job AI can do

Top models in data governance

  1. gemini-4-argon · 81% of the competence
  2. claude-haiku-5-5 · 78% of the competence
  3. gpt-6-luna · 70% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data architecture tests. Of its 89 benchmarked knobits, 17 are mastered by most models, and in 0 people still lead.

See the AI benchmarks · Which parts of a data architect's job AI can do

Top models in data architecture

  1. claude-haiku-5-5 · 78% of the competence
  2. gemini-4-argon · 78% of the competence
  3. gpt-6-luna · 68% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data storage and operations tests. Of its 102 benchmarked knobits, 27 are mastered by most models, and in 2 people still lead.

See the AI benchmarks · Which parts of a database administrator's job AI can do

Top models in data storage and operations

  1. gemini-4-argon · 83% of the competence
  2. claude-haiku-5-5 · 82% of the competence
  3. gpt-6-luna · 77% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data security tests. Of its 107 benchmarked knobits, 42 are mastered by most models, and in 1 people still lead.

See the AI benchmarks · Which parts of a data security specialist's job AI can do

Top models in data security

  1. gpt-6-luna · 87% of the competence
  2. gemini-4-argon · 87% of the competence
  3. claude-haiku-5-5 · 86% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data integration and interoperability tests. Of its 104 benchmarked knobits, 26 are mastered by most models, and in 2 people still lead.

See the AI benchmarks · Which parts of a data integration specialist's job AI can do

Top models in data integration and interoperability

  1. gemini-4-argon · 89% of the competence
  2. claude-haiku-5-5 · 83% of the competence
  3. gpt-6-luna · 76% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the document and content management tests. Of its 86 benchmarked knobits, 29 are mastered by most models, and in 1 people still lead.

See the AI benchmarks · Which parts of a records and document manager's job AI can do

Top models in document and content management

  1. claude-haiku-5-5 · 83% of the competence
  2. gemini-4-argon · 81% of the competence
  3. gpt-6-luna · 76% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the reference and master data tests. Of its 84 benchmarked knobits, 22 are mastered by most models, and in 3 people still lead.

See the AI benchmarks · Which parts of a master data manager's job AI can do

Top models in reference and master data

  1. gemini-4-argon · 81% of the competence
  2. gpt-6-luna · 80% of the competence
  3. claude-haiku-5-5 · 80% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data warehousing and business intelligence tests. Of its 84 benchmarked knobits, 22 are mastered by most models, and in 0 people still lead.

See the AI benchmarks · Which parts of a data warehouse and bi developer's job AI can do

Top models in data warehousing and business intelligence

  1. gemini-4-argon · 82% of the competence
  2. claude-haiku-5-5 · 77% of the competence
  3. gpt-6-luna · 75% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the metadata management tests. Of its 100 benchmarked knobits, 16 are mastered by most models, and in 1 people still lead.

See the AI benchmarks · Which parts of a metadata manager's job AI can do

Top models in metadata management

  1. gemini-4-argon · 87% of the competence
  2. claude-haiku-5-5 · 79% of the competence
  3. gpt-6-luna · 68% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the data quality tests. Of its 164 benchmarked knobits, 52 are mastered by most models, and in 6 people still lead.

See the AI benchmarks · Which parts of a data quality analyst's job AI can do

Top models in data quality

  1. gemini-4-argon · 86% of the competence
  2. claude-haiku-5-5 · 82% of the competence
  3. gpt-6-luna · 75% of the competence

Can AI do the job?

4 AI models from Google, OpenAI and Anthropic took the skill assessment tests. Of its 49 benchmarked knobits, 13 are mastered by most models, and in 0 people still lead.

See the AI benchmarks · Which parts of a assessment specialist's job AI can do

Top models in skill assessment

  1. gemini-4-argon · 85% of the competence
  2. claude-haiku-5-5 · 83% of the competence
  3. gpt-6-luna · 51% of the competence

Pilot status

Skilirium is in its pilot. The questions were drafted with the help of AI and are being reviewed by data modelling experts, so treat a result as an indication rather than a certificate. Feedback is welcome at peepkungas@gmail.com.

Terms of use