How Skills Atlas Entries Are Written and Checked

The Skills Atlas describes what a capability involves in practice: the decisions a person makes, the result of their work, and the conditions under which a method can fail. A label such as “Pandas” or “RAG pipeline design” cannot answer those questions on its own.

The current entries follow the explanatory approach used in the AI Glossary, with the emphasis shifted from defining a concept to understanding a skill.

What an entry should explain

Every skill starts with a short definition that makes sense outside its page, including in search results. The main explanation then develops four questions:

What is it? The mechanism, scope, and concepts needed to understand the capability.
What does the work involve? The decisions, inputs, workflow, and deliverable expected from someone applying it.
What would it look like? A concrete illustrative scenario with a task and a way to inspect the outcome.
Where are the limits? Common mistakes, important distinctions, failure conditions, and appropriate quality checks.

These sections serve different purposes. An algorithm entry should explain how the method operates and how its assumptions affect the result. A tool entry should show the operations that matter and the limits of its execution model. A governance entry should identify the decisions and evidence involved without presenting a checklist as proof of compliance.

Definitions and competencies

Knowing the name of an algorithm is different from being able to use it. Logistic regression, for example, involves preparing features, fitting the model, interpreting coefficients, choosing a decision threshold, and checking performance on appropriate data.

The Atlas therefore connects an explanation to the work it supports. It does not certify that a reader possesses a skill, or turn familiarity with a product into a measure of proficiency. The learning graph offers a route through related knowledge; a person's prior experience may justify a different route.

Sources and examples

References should support the specific mechanism or practice described. For an algorithm, this can mean an original paper and the documentation of an implementation. For a library, it means documentation of the relevant operations rather than a vendor homepage. Each source has a short note describing what it contributes to the entry.

Examples are illustrative. They show how a decision could be made and checked; they are not customer case studies or evidence of a measured gain. An entry should not invent performance figures, adoption claims, or guarantees to make a method sound more valuable.

Why the research-status badges were removed

Early versions used agreement between research runs to help assemble the inventory. That agreement did not establish a useful explanation, a sound implementation, or the correctness of an individual claim. The public entries now focus on definitions, practice, limitations, and primary references. Readers can inspect those directly without translating a model-run score into an assumed quality judgment.

The current map

The current edition contains 466 skills across 15 domains, connected by 460 ontology relations and 330 prerequisite relations.

Stable skill identities keep existing links and learning relationships usable as descriptions are revised. Changes to a definition need a check against adjacent entries: a rewrite of LoRA should not silently absorb quantization, and an explanation of retrieval should not claim that retrieved material establishes the truth of an answer.

The map remains an editorial guide to GenAI and data-science work. Use an entry to understand a capability, follow its sources to investigate further, and test an implementation under the conditions of the task you actually face.