Understand Methods
A Method is an analytical capability applied to data. A data scientist registers a function, and users and AI reuse it with the metrics they need.
A Python Method executes a calculation function. A Skill Method provides analysis guidance for AI. This guide explains the Python function path.
Starting catalog
The default catalog contains four Methods.
| Type | Method | Purpose |
|---|---|---|
| Skill | Metric lookup | Query selected metrics and dates |
| Skill | Group breakdown | Inspect groups and investigate a selected group |
| Python | OLS regression | Analyze associations while accounting for other variables |
| Python | Matching | Compare groups with similar observed covariates |
Select inputs in the console to test a Python Method. Skills are driven by an MCP-connected AI; starting one on the web waits for the AI's queries. Statistical Methods require their dependencies and suitable entity-level data.
The Method detail shows execution records for its exact version. Registration and requirement checks do not execute an analysis. Existing Recipes can still resolve their previously registered Methods.
Define the calculation, then select the data
Registration defines how to calculate. Execution selects which data to analyze.
| When | What is selected | Example |
|---|---|---|
| Register a Method | Function and input requirements | OLS requiring a numeric outcome and explanatory variables |
| Execute an analysis | Metrics, variables and dates | Customer purchases, ad exposures and February data |
The semantic layer owns definitions and aggregations. The Method calculates using data queried from those definitions.
When AI requests an analysis
- AI discovers a registered Method and reads its input requirements.
- It selects metrics and dimensions from the semantic catalog, asking about ambiguous inputs or dates.
- Decision Layer validates the inputs, queries the data and passes it to the function.
- The server records inputs, queries and results in a Run. AI reads and explains the result.
Web, REST and MCP share the same execution code. A local notebook session lets a developer inspect querying and calculation without creating a server Run.
See an example
Connect a Python function shows ranking and OLS with code and results.
Continue with Register in a project to make a function available in the service. If existing Methods cover the analysis, compose them in a Recipe.
See Method API to look up registration arguments and input options.