Method API

Look up arguments and options when registering a function. See Connect a Python function for the full flow.

python_method

from decision_layer.methods import Column, DataInput, Scope, python_method

method = python_method(
    function,
    name="my_team.analysis",
    version="1.0.0",
    inputs={"data": DataInput(columns={...})},
)

Argument	Meaning
First argument	Python function to execute
name, version	Required Method name and version
inputs	Required mapping from data arguments to DataInput
output	Result shape: default "auto", or "table", "estimate"
requires_packages	Required package versions, such as {"statsmodels": ">=0.14,<0.16"}
checks	Additional data checks
option_checks	Option checks executed before querying
adapt	Optional function converting the returned value to a supported shape
provides	Capability names used by Recipes and analysis goals

Ordinary function arguments become calculation options, with types and defaults inferred from the signature. Use parameters for explicit overrides. The docstring supplies the Method description.

requires_packages checks installed versions; it does not install packages at runtime.

DataInput and Column

Each inputs key names a function argument receiving a DataFrame. Declare a separate input for each table.

inputs={
    "data": DataInput(
        columns={
            "value": Column(source="measure", dtype="number"),
            "group": Column(source="dimension"),
        },
        minimum_rows=10,
    ),
}

Column option	Meaning
source	Required: "measure", "dimension", or a list allowing both
dtype	Require "number", "string", "boolean" or "time"; omitted means unrestricted
multiple=True	Select several metrics or dimensions; default False
required=False	Optional input; default True
role="entity_key"	Key identifying each row, such as customer or order ID
description	Explanation of the input

An entity_key must be one required dimension declared as an entity key by the semantic provider. Each DataInput allows one. Selecting a key does not establish statistical independence.

Function arguments matching declared columns receive DataFrame column names. Multiple columns receive names such as confounders_0, confounders_1 and are passed as a list. Select actual Cube references through bindings at execution time.

Common DataInput option	Meaning
minimum_rows	Minimum row count; default 1
format	Default "dataframe"; "records" provides a list of row dicts
time_granularity	Query by "day", "week", "month", "quarter" or "year"
period_parameter	Option naming a comparison date range
period_role="comparison"	Query a separately selected comparison period
filters_parameter	Option supplying additional query filters

Return values

Function result	Adaptation
DataFrame or list of row dicts	Table with default settings
Number	Estimate with default settings
Dict	One-row table with default settings
{"value": number, ...}	Estimate with details when output="estimate"

output="table" and output="estimate" require the selected shape; mismatches raise an error. Convert unsupported external-library objects inside the function or through adapt.

Checks and errors

Decision Layer checks required inputs, types and declared data conditions. Each Method does not need its own handling for those checks.

• checks run after querying and before calculation. Arguments use the DataInput names.
• option_checks receive only calculation options and run before querying.
• Returning False or raising ValueError or AssertionError records a failed check and stops calculation. None or True passes.
• Exceptions in the calculation function itself propagate as Python errors. Programming failures are not successful results.

Local sessions

API	Purpose
session.register(method)	Add a Method to the session
session.list_methods()	List registered manifests
session.find_methods(...)	Search Methods and their input requirements
await session.run(name, bindings=..., params=..., scope=...)	Execute locally and inspect the result

bindings selects semantic references; params supplies calculation options. Use Scope(date_range=(start, end), time_dimension=date_reference) for dates.

Inspect trial.result and trial.attempts for results and query attempts. Trials are not automatically stored on the server. Continue with Register in a project for service use.

For advanced control over queries, see the SDK examples. See the implementation for the complete registration signature.