Assumption

An Assumption returns rows that contradict a statement about the estate. It passes when no rows are returned.

Location and identity

Item	Location	Filename	Authored form
Lakehouse	Lakehouse/<Item>/assumptions/	Schema__Object.py	Python
Lakehouse	Lakehouse/<Item>/assumptions/	Schema.Object.sql	Spark SQL
Warehouse	Warehouse/<Item>/assumptions/	Schema.Object.sql	T-SQL

For the Lakehouse/Warehouse forms above, the file must sit directly under assumptions/. Semantic-model Assumptions instead use PowerBI/<project>/assumptions/<Model>/<Schema.Object>.dax; see DAX validations for that supported form. Assumption ID: Schema.Object, filename and Python class name must agree exactly, including case. Each identity component is one logical name: no forward slash, backslash, dot, colon or surrounding whitespace. Tests and Assumptions share one validation namespace. In a Warehouse that namespace also collides with Tables and Views of the same identity; a Lakehouse validation has no Tables/ area.

Metadata

Key	Required	Value and default
Assumption ID	Yes	One non-empty Schema.Object; the file contains no other ID key.
Description	Yes	Non-empty prose or one metadata reference.
Dependencies	No	YAML list of item-relative Schema.Object names. Absent: infer; present: replace inference; []: none.
Notes	No	Non-empty free text.
Revision notes	No	Non-empty YAML list of dated notes using one accepted date spelling throughout the document; see Common metadata.

Only these keys are accepted. Primary key is incompatible because an Assumption has no expected/actual pair to correlate. Lineage, Schema, Column notes, Unique keys, Foreign keys, Not null, Identity, Comparison columns, Incremental, Static, Prohibit rebuild, File key, Has load procedure, Delete percentage threshold, Update percentage threshold and Stability row threshold are also incompatible.

Dependencies are inferred from Python object imports or SQL relation references when Dependencies is absent. Spark SQL Assumptions may use inference; the explicit-dependency requirement for Spark SQL Tables and Views does not apply to validations. A declaration replaces inferred references and cannot name the Assumption itself or any Test or Assumption. Item graphs remain acyclic.

Python body

The module declares exactly one Weaver class directly inheriting weaver.Assumption. Its name equals the filename stem and it implements one synchronous read() method. read() returns a Spark DataFrame containing the violating rows; any returned row is a failure.

"""
Assumption ID: Parcel.HasDestination

Description: Every parcel has a destination.
"""

from weaver import Assumption


class Parcel__HasDestination(Assumption):
    def read(self):
        return self.spark.createDataFrame([], "ParcelId string, Destination string")

Path: Lakehouse/Quality/assumptions/Parcel__HasDestination.py.

SQL body

Setup statements that return no rows may precede the contract query. All setup must finish before the first result-producing query. The body then produces exactly one result set containing violating rows.

Spark SQL example:

/*
Assumption ID: Parcel.HasDestination

Description: Every parcel has a destination.
*/
select cast(null as string) as ParcelId
where false;

Path: Lakehouse/Quality/assumptions/Parcel.HasDestination.sql.

Warehouse T-SQL example:

/*
Assumption ID: Parcel.HasDestination

Description: Every parcel has a destination.
*/
select cast(null as varchar(50)) as ParcelId
where 1 = 0;

Path: Warehouse/Quality/assumptions/Parcel.HasDestination.sql.

Operations and managed state

• Check parses metadata, identity, class/method structure, SQL program shape, dependencies and cycles without running the Assumption.
• Build installs the declaration and its runnable form. Lakehouse Python is deployed as a module; Lakehouse Spark SQL is compiled to a module; Warehouse T-SQL is compiled to a stored procedure. Build does not evaluate it. A rebuilt Assumption receives current status Pending.
• Load does not run Assumptions.
• Test runs the selected installed definition without automatically running its data dependencies. Zero rows records Succeeded; returned rows record Failed; execution failures record Error; unavailable dependencies can record Blocked. The violation count is retained, while diagnostic rows are not persisted.

_.TestDictionary records the logical item, Schema.Object, type Assumption, resolved description and its reference, a null primary key and source signature. _.Registry records the compiled module or procedure with role Assumption, not a physical object at the logical Assumption ID. _.Dependency records resolved edges. _.TestStatus records the logical identity, test type, workflow, result, timing and current violation count as Failure count; _.Log keeps execution history.