acryl-datahub-cloud 0.3.11rc0__py3-none-any.whl → 0.3.16.1rc0__py3-none-any.whl

acryl_datahub_cloud/sdk/assertion_input/assertion_input.py

@@ -0,0 +1,1470 @@
+"""
+This file contains the AssertionInput class and related classes, which are used to
+validate and represent the input for creating an Assertion in DataHub.
+"""
+
+import inspect
+import random
+import string
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from datetime import datetime
+from enum import Enum
+from typing import (
+    Callable,
+    Collection,
+    Literal,
+    Optional,
+    Type,
+    TypeAlias,
+    TypeVar,
+    Union,
+    cast,
+)
+
+import pydantic
+import pytz
+import tzlocal
+from avrogen.dict_wrapper import DictWrapper
+from croniter import croniter
+from pydantic import BaseModel, Extra, ValidationError
+
+from acryl_datahub_cloud.sdk.entities.assertion import (
+    Assertion,
+    AssertionActionsInputType,
+    AssertionInfoInputType,
+    TagsInputType,
+)
+from acryl_datahub_cloud.sdk.entities.monitor import Monitor
+from acryl_datahub_cloud.sdk.errors import (
+    SDKUsageError,
+    SDKUsageErrorWithExamples,
+)
+from datahub.emitter.enum_helpers import get_enum_options
+from datahub.emitter.mce_builder import make_ts_millis, parse_ts_millis
+from datahub.metadata import schema_classes as models
+from datahub.metadata.urns import AssertionUrn, CorpUserUrn, DatasetUrn, TagUrn
+from datahub.sdk import Dataset
+from datahub.sdk.entity_client import EntityClient
+
+# TODO: Import ASSERTION_MONITOR_DEFAULT_TRAINING_LOOKBACK_WINDOW_DAYS from datahub_executor.config
+ASSERTION_MONITOR_DEFAULT_TRAINING_LOOKBACK_WINDOW_DAYS = 60
+
+DEFAULT_NAME_PREFIX = "New Assertion"
+DEFAULT_NAME_SUFFIX_LENGTH = 8
+
+
+DEFAULT_DAILY_SCHEDULE: models.CronScheduleClass = models.CronScheduleClass(
+    cron="0 0 * * *",  # Every day at midnight, matches the UI default
+    timezone=str(
+        tzlocal.get_localzone()
+    ),  # User local timezone, matches the UI default
+)
+DEFAULT_SCHEDULE: models.CronScheduleClass = DEFAULT_DAILY_SCHEDULE
+
+# Legacy aliases - all now point to daily schedule for consistency
+DEFAULT_HOURLY_SCHEDULE: models.CronScheduleClass = DEFAULT_DAILY_SCHEDULE
+DEFAULT_EVERY_SIX_HOURS_SCHEDULE: models.CronScheduleClass = DEFAULT_DAILY_SCHEDULE
+
+TYPE_CLASS_NAME_TO_TYPE_MAP = {
+    "StringTypeClass": "STRING",
+    "NumberTypeClass": "NUMBER",
+    "BooleanTypeClass": "BOOLEAN",
+    "DateTypeClass": "DATE",
+    "TimeTypeClass": "TIME",
+    "NullTypeClass": "NULL",
+}
+
+
+class AbstractDetectionMechanism(BaseModel, ABC):
+    type: str
+
+    class Config:
+        extra = Extra.forbid
+
+
+class _InformationSchema(AbstractDetectionMechanism):
+    type: Literal["information_schema"] = "information_schema"
+
+
+class _AuditLog(AbstractDetectionMechanism):
+    type: Literal["audit_log"] = "audit_log"
+
+
+# Keep this in sync with the allowed field types in the UI, currently in
+# datahub-web-react/src/app/entity/shared/tabs/Dataset/Validations/assertion/builder/constants.ts: LAST_MODIFIED_FIELD_TYPES
+LAST_MODIFIED_ALLOWED_FIELD_TYPES = [models.DateTypeClass(), models.TimeTypeClass()]
+
+
+class _LastModifiedColumn(AbstractDetectionMechanism):
+    type: Literal["last_modified_column"] = "last_modified_column"
+    column_name: str
+    additional_filter: Optional[str] = None
+
+
+# Keep this in sync with the allowed field types in the UI, currently in
+# datahub-web-react/src/app/entity/shared/tabs/Dataset/Validations/assertion/builder/constants.ts: HIGH_WATERMARK_FIELD_TYPES
+HIGH_WATERMARK_ALLOWED_FIELD_TYPES = [
+    models.NumberTypeClass(),
+    models.DateTypeClass(),
+    models.TimeTypeClass(),
+]
+
+
+class _HighWatermarkColumn(AbstractDetectionMechanism):
+    type: Literal["high_watermark_column"] = "high_watermark_column"
+    column_name: str
+    additional_filter: Optional[str] = None
+
+
+class _DataHubOperation(AbstractDetectionMechanism):
+    type: Literal["datahub_operation"] = "datahub_operation"
+
+
+class _Query(AbstractDetectionMechanism):
+    # COUNT(*) query
+    type: Literal["query"] = "query"
+    additional_filter: Optional[str] = None
+
+
+class _AllRowsQuery(AbstractDetectionMechanism):
+    # For column-based assertions, this is the default detection mechanism.
+    type: Literal["all_rows_query"] = "all_rows_query"
+    additional_filter: Optional[str] = None
+
+
+class _AllRowsQueryDataHubDatasetProfile(AbstractDetectionMechanism):
+    # Used for column-based assertions.
+    type: Literal["all_rows_query_datahub_dataset_profile"] = (
+        "all_rows_query_datahub_dataset_profile"
+    )
+
+
+class _ChangedRowsQuery(AbstractDetectionMechanism):
+    # Used for column-based assertions.
+    type: Literal["changed_rows_query"] = "changed_rows_query"
+    column_name: str
+    additional_filter: Optional[str] = None
+
+
+class _DatasetProfile(AbstractDetectionMechanism):
+    type: Literal["dataset_profile"] = "dataset_profile"
+
+
+class _SchemaMetadata(AbstractDetectionMechanism):
+    type: Literal["schema_metadata"] = "schema_metadata"
+
+
+# Operators that require a single value numeric parameter
+SINGLE_VALUE_NUMERIC_OPERATORS = [
+    models.AssertionStdOperatorClass.EQUAL_TO,
+    models.AssertionStdOperatorClass.NOT_EQUAL_TO,
+    models.AssertionStdOperatorClass.GREATER_THAN,
+    models.AssertionStdOperatorClass.LESS_THAN,
+    models.AssertionStdOperatorClass.GREATER_THAN_OR_EQUAL_TO,
+    models.AssertionStdOperatorClass.LESS_THAN_OR_EQUAL_TO,
+]
+
+# Operators that require a single value parameter
+SINGLE_VALUE_OPERATORS = [
+    models.AssertionStdOperatorClass.CONTAIN,
+    models.AssertionStdOperatorClass.END_WITH,
+    models.AssertionStdOperatorClass.START_WITH,
+    models.AssertionStdOperatorClass.REGEX_MATCH,
+    models.AssertionStdOperatorClass.IN,
+    models.AssertionStdOperatorClass.NOT_IN,
+] + SINGLE_VALUE_NUMERIC_OPERATORS
+
+# Operators that require a numeric range parameter
+RANGE_OPERATORS = [
+    models.AssertionStdOperatorClass.BETWEEN,
+]
+
+# Operators that require no parameters
+NO_PARAMETER_OPERATORS = [
+    models.AssertionStdOperatorClass.NULL,
+    models.AssertionStdOperatorClass.NOT_NULL,
+    models.AssertionStdOperatorClass.IS_TRUE,
+    models.AssertionStdOperatorClass.IS_FALSE,
+]
+
+
+# Keep these two lists in sync:
+_DETECTION_MECHANISM_CONCRETE_TYPES = (
+    _InformationSchema,
+    _AuditLog,
+    _LastModifiedColumn,
+    _HighWatermarkColumn,
+    _DataHubOperation,
+    _Query,
+    _DatasetProfile,
+    _AllRowsQuery,
+    _ChangedRowsQuery,
+    _AllRowsQueryDataHubDatasetProfile,
+    _SchemaMetadata,
+)
+_DetectionMechanismTypes = Union[
+    _InformationSchema,
+    _AuditLog,
+    _LastModifiedColumn,
+    _HighWatermarkColumn,
+    _DataHubOperation,
+    _Query,
+    _DatasetProfile,
+    _AllRowsQuery,
+    _ChangedRowsQuery,
+    _AllRowsQueryDataHubDatasetProfile,
+    _SchemaMetadata,
+]
+
+_DETECTION_MECHANISM_TYPES_WITH_ADDITIONAL_FILTER = (
+    _LastModifiedColumn,
+    _HighWatermarkColumn,
+    _Query,
+    _AllRowsQuery,
+    _ChangedRowsQuery,
+)
+
+DEFAULT_DETECTION_MECHANISM: _DetectionMechanismTypes = _InformationSchema()
+
+
+class DetectionMechanism:
+    # To have a more enum-like user experience even with sub parameters, we define the detection mechanisms as class attributes.
+    # The options with sub parameters are the classes themselves so that parameters can be applied, and the rest are already instantiated instances of the classes.
+    INFORMATION_SCHEMA = _InformationSchema()
+    AUDIT_LOG = _AuditLog()
+    LAST_MODIFIED_COLUMN = _LastModifiedColumn
+    HIGH_WATERMARK_COLUMN = _HighWatermarkColumn
+    DATAHUB_OPERATION = _DataHubOperation()
+    QUERY = _Query
+    ALL_ROWS_QUERY = _AllRowsQuery
+    CHANGED_ROWS_QUERY = _ChangedRowsQuery
+    ALL_ROWS_QUERY_DATAHUB_DATASET_PROFILE = _AllRowsQueryDataHubDatasetProfile()
+    DATASET_PROFILE = _DatasetProfile()
+
+    _DETECTION_MECHANISM_EXAMPLES = {
+        "Information Schema from string": "information_schema",
+        "Information Schema from DetectionMechanism": "DetectionMechanism.INFORMATION_SCHEMA",
+        "Audit Log from string": "audit_log",
+        "Audit Log from DetectionMechanism": "DetectionMechanism.AUDIT_LOG",
+        "Last Modified Column from dict": {
+            "type": "last_modified_column",
+            "column_name": "last_modified",
+            "additional_filter": "last_modified > '2021-01-01'",
+        },
+        "Last Modified Column from DetectionMechanism": "DetectionMechanism.LAST_MODIFIED_COLUMN(column_name='last_modified', additional_filter='last_modified > 2021-01-01')",
+        "High Watermark Column from dict": {
+            "type": "high_watermark_column",
+            "column_name": "id",
+            "additional_filter": "id > 1000",
+        },
+        "High Watermark Column from DetectionMechanism": "DetectionMechanism.HIGH_WATERMARK_COLUMN(column_name='id', additional_filter='id > 1000')",
+        "DataHub Operation from string": "datahub_operation",
+        "DataHub Operation from DetectionMechanism": "DetectionMechanism.DATAHUB_OPERATION",
+        "Query from string": "query",
+        "Query from dict": {
+            "type": "query",
+            "additional_filter": "id > 1000",
+        },
+        "Query from DetectionMechanism (with optional additional filter)": "DetectionMechanism.QUERY(additional_filter='id > 1000')",
+        "Dataset Profile from string": "dataset_profile",
+        "Dataset Profile from DetectionMechanism": "DetectionMechanism.DATASET_PROFILE",
+        "All Rows Query from string": "all_rows_query",
+        "All Rows Query from DetectionMechanism": "DetectionMechanism.ALL_ROWS_QUERY",
+        "All Rows Query from DetectionMechanism (with optional additional filter)": "DetectionMechanism.ALL_ROWS_QUERY(additional_filter='id > 1000')",
+        "Changed Rows Query from dict (with optional additional filter)": {
+            "type": "changed_rows_query",
+            "column_name": "id",
+            "additional_filter": "id > 1000",
+        },
+        "Changed Rows Query from DetectionMechanism": "DetectionMechanism.CHANGED_ROWS_QUERY(column_name='id')",
+        "Changed Rows Query from DetectionMechanism (with optional additional filter)": "DetectionMechanism.CHANGED_ROWS_QUERY(column_name='id', additional_filter='id > 1000')",
+        "All Rows Query DataHub Dataset Profile from string": "all_rows_query_datahub_dataset_profile",
+        "All Rows Query DataHub Dataset Profile from DetectionMechanism": "DetectionMechanism.ALL_ROWS_QUERY_DATAHUB_DATASET_PROFILE",
+    }
+
+    @staticmethod
+    def parse(
+        detection_mechanism_config: Optional[
+            Union[str, dict[str, str], _DetectionMechanismTypes]
+        ] = None,
+        default_detection_mechanism: _DetectionMechanismTypes = DEFAULT_DETECTION_MECHANISM,
+    ) -> _DetectionMechanismTypes:
+        if detection_mechanism_config is None:
+            return default_detection_mechanism
+        if isinstance(detection_mechanism_config, _DETECTION_MECHANISM_CONCRETE_TYPES):
+            return detection_mechanism_config
+        elif isinstance(detection_mechanism_config, str):
+            return DetectionMechanism._try_parse_from_string(detection_mechanism_config)
+        elif isinstance(detection_mechanism_config, dict):
+            return DetectionMechanism._try_parse_from_dict(detection_mechanism_config)
+        else:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid detection mechanism: {detection_mechanism_config}",
+                examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+            )
+
+    @staticmethod
+    def _try_parse_from_string(
+        detection_mechanism_config: str,
+    ) -> _DetectionMechanismTypes:
+        try:
+            return_value = getattr(
+                DetectionMechanism, detection_mechanism_config.upper()
+            )
+            if inspect.isclass(return_value) and issubclass(
+                return_value, pydantic.BaseModel
+            ):
+                try:
+                    # We try to instantiate here to let pydantic raise a helpful error
+                    # about which parameters are missing
+                    return_value = return_value()
+                except ValidationError as e:
+                    raise SDKUsageErrorWithExamples(
+                        msg=f"Detection mechanism type '{detection_mechanism_config}' requires additional parameters: {e}",
+                        examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+                    ) from e
+            return return_value
+        except AttributeError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid detection mechanism type: {detection_mechanism_config}",
+                examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+            ) from e
+
+    @staticmethod
+    def _try_parse_from_dict(
+        detection_mechanism_config: dict[str, str],
+    ) -> _DetectionMechanismTypes:
+        # Make a copy of the dictionary to avoid mutating the original
+        config_copy = detection_mechanism_config.copy()
+
+        try:
+            detection_mechanism_type = config_copy.pop("type")
+        except KeyError as e:
+            raise SDKUsageErrorWithExamples(
+                msg="Detection mechanism type is required if using a dict to create a DetectionMechanism",
+                examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+            ) from e
+        try:
+            detection_mechanism_obj = getattr(
+                DetectionMechanism, detection_mechanism_type.upper()
+            )
+        except AttributeError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid detection mechanism type: {detection_mechanism_type}",
+                examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+            ) from e
+
+        try:
+            return detection_mechanism_obj(**config_copy)
+        except TypeError as e:
+            if "object is not callable" not in e.args[0]:
+                raise e
+            if config_copy:
+                # If we are here in the TypeError case, the detection mechanism is an instance of a class,
+                # not a class itself, so we can't instantiate it with the config dict.
+                # In this case, the config dict should be empty after the type is popped.
+                # If it is not empty, we raise an error.
+                raise SDKUsageErrorWithExamples(
+                    msg=f"Invalid additional fields specified for detection mechanism '{detection_mechanism_type}': {config_copy}",
+                    examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+                ) from e
+            return detection_mechanism_obj
+        except ValidationError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid detection mechanism type '{detection_mechanism_type}': {config_copy} {e}",
+                examples=DetectionMechanism._DETECTION_MECHANISM_EXAMPLES,
+            ) from e
+
+
+DetectionMechanismInputTypes: TypeAlias = Union[
+    str, dict[str, str], _DetectionMechanismTypes, None
+]
+
+
+class InferenceSensitivity(Enum):
+    HIGH = "high"
+    MEDIUM = "medium"
+    LOW = "low"
+
+    @staticmethod
+    def parse(
+        sensitivity: Optional[
+            Union[
+                str,
+                int,
+                "InferenceSensitivity",
+                models.AssertionMonitorSensitivityClass,
+            ]
+        ],
+    ) -> "InferenceSensitivity":
+        if sensitivity is None:
+            return DEFAULT_SENSITIVITY
+        EXAMPLES = {
+            "High sensitivity from string": "high",
+            "High sensitivity from enum": "InferenceSensitivity.HIGH",
+            "Medium sensitivity from string": "medium",
+            "Medium sensitivity from enum": "InferenceSensitivity.MEDIUM",
+            "Low sensitivity from string": "low",
+            "Low sensitivity from enum": "InferenceSensitivity.LOW",
+            "Sensitivity from int (1-3: low, 4-6: medium, 7-10: high)": "10",
+        }
+
+        if isinstance(sensitivity, InferenceSensitivity):
+            return sensitivity
+        if isinstance(sensitivity, models.AssertionMonitorSensitivityClass):
+            sensitivity = sensitivity.level
+        if isinstance(sensitivity, int):
+            if (sensitivity < 1) or (sensitivity > 10):
+                raise SDKUsageErrorWithExamples(
+                    msg=f"Invalid inference sensitivity: {sensitivity}",
+                    examples=EXAMPLES,
+                )
+            elif sensitivity < 4:
+                return InferenceSensitivity.LOW
+            elif sensitivity < 7:
+                return InferenceSensitivity.MEDIUM
+            else:
+                return InferenceSensitivity.HIGH
+        try:
+            return InferenceSensitivity(sensitivity)
+        except ValueError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid inference sensitivity: {sensitivity}",
+                examples=EXAMPLES,
+            ) from e
+
+    @staticmethod
+    def to_int(sensitivity: "InferenceSensitivity") -> int:
+        return {
+            InferenceSensitivity.HIGH: 10,
+            InferenceSensitivity.MEDIUM: 5,
+            InferenceSensitivity.LOW: 1,
+        }[sensitivity]
+
+
+DEFAULT_SENSITIVITY: InferenceSensitivity = InferenceSensitivity.MEDIUM
+
+TIME_WINDOW_SIZE_EXAMPLES = {
+    "Recommended: Time window size from objects": "TimeWindowSize(unit=CalendarInterval.MINUTE, multiple=10)",
+    "Time window size from object": "TimeWindowSize(unit='MINUTE', multiple=10)",
+    "Time window size from models.TimeWindowSizeClass": "models.TimeWindowSizeClass(unit='MINUTE', multiple=10)",
+    "Time window size from dict": '{"unit": "MINUTE", "multiple": 10}',
+}
+
+
+class CalendarInterval(str, Enum):
+    MINUTE = "MINUTE"
+    HOUR = "HOUR"
+    DAY = "DAY"
+
+
+class TimeWindowSize(BaseModel):
+    unit: Union[CalendarInterval, str]
+    multiple: int
+
+
+TimeWindowSizeInputTypes: TypeAlias = Union[
+    models.TimeWindowSizeClass,
+    models.FixedIntervalScheduleClass,
+    TimeWindowSize,
+    dict[str, Union[str, int]],  # {"unit": "MINUTE", "multiple": 10}
+]
+
+
+def _try_parse_calendar_interval(
+    config: Union[str, CalendarInterval],
+) -> CalendarInterval:
+    if isinstance(config, CalendarInterval):
+        return config
+    try:
+        return CalendarInterval(config.upper())
+    except ValueError as e:
+        raise SDKUsageErrorWithExamples(
+            msg=f"Invalid calendar interval: {config}",
+            examples=TIME_WINDOW_SIZE_EXAMPLES,
+        ) from e
+
+
+def _try_parse_time_window_size(
+    config: TimeWindowSizeInputTypes,
+) -> models.TimeWindowSizeClass:
+    if isinstance(config, models.TimeWindowSizeClass):
+        return config
+    elif isinstance(config, models.FixedIntervalScheduleClass):
+        return models.TimeWindowSizeClass(
+            unit=_try_parse_and_validate_schema_classes_enum(
+                config.unit, models.CalendarIntervalClass
+            ),
+            multiple=config.multiple,
+        )
+    elif isinstance(config, TimeWindowSize):
+        return models.TimeWindowSizeClass(
+            unit=_try_parse_and_validate_schema_classes_enum(
+                _try_parse_and_validate_schema_classes_enum(
+                    config.unit, CalendarInterval
+                ).value,
+                models.CalendarIntervalClass,
+            ),
+            multiple=config.multiple,
+        )
+    elif isinstance(config, dict):
+        if "unit" not in config or "multiple" not in config:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid time window size: {config}",
+                examples=TIME_WINDOW_SIZE_EXAMPLES,
+            )
+        try:
+            multiple = int(config["multiple"])
+        except ValueError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid time window size: {config}",
+                examples=TIME_WINDOW_SIZE_EXAMPLES,
+            ) from e
+        return models.TimeWindowSizeClass(
+            unit=_try_parse_calendar_interval(str(config["unit"])),
+            multiple=multiple,
+        )
+    else:
+        raise SDKUsageErrorWithExamples(
+            msg=f"Invalid time window size: {config}",
+            examples=TIME_WINDOW_SIZE_EXAMPLES,
+        )
+
+
+class FixedRangeExclusionWindow(BaseModel):
+    type: Literal["fixed_range_exclusion_window"] = "fixed_range_exclusion_window"
+    start: datetime
+    end: datetime
+
+
+ExclusionWindowTypes: TypeAlias = Union[
+    FixedRangeExclusionWindow,
+    # Add other exclusion window types here as they are added to the SDK.
+]
+
+FIXED_RANGE_EXCLUSION_WINDOW_EXAMPLES = {
+    "Exclusion Window from datetimes": {
+        "start": "datetime(2025, 1, 1, 0, 0, 0)",
+        "end": "datetime(2025, 1, 2, 0, 0, 0)",
+    },
+    "Exclusion Window from strings": {
+        "start": "2025-01-01T00:00:00",
+        "end": "2025-01-02T00:00:00",
+    },
+    "Exclusion Window from object": "ExclusionWindow(start=datetime(2025, 1, 1, 0, 0, 0), end=datetime(2025, 1, 2, 0, 0, 0))",
+}
+FixedRangeExclusionWindowInputTypes: TypeAlias = Union[
+    dict[str, datetime],
+    dict[str, str],
+    list[dict[str, datetime]],
+    list[dict[str, str]],
+    FixedRangeExclusionWindow,
+    list[FixedRangeExclusionWindow],
+]
+
+ExclusionWindowInputTypes: TypeAlias = Union[
+    models.AssertionExclusionWindowClass,
+    list[models.AssertionExclusionWindowClass],
+    FixedRangeExclusionWindowInputTypes,
+    # Add other exclusion window types here as they are added to the SDK.
+]
+
+IterableExclusionWindowInputTypes: TypeAlias = Union[
+    list[dict[str, datetime]],
+    list[dict[str, str]],
+    list[FixedRangeExclusionWindow],
+    list[models.AssertionExclusionWindowClass],
+]
+
+
+def _try_parse_exclusion_window(
+    config: Optional[ExclusionWindowInputTypes],
+) -> Union[FixedRangeExclusionWindow, list[FixedRangeExclusionWindow], None]:
+    if config is None:
+        return []
+    if isinstance(config, dict):
+        return [FixedRangeExclusionWindow(**config)]
+    if isinstance(config, FixedRangeExclusionWindow):
+        return [config]
+    elif isinstance(config, models.AssertionExclusionWindowClass):
+        assert config.fixedRange is not None
+        return [
+            FixedRangeExclusionWindow(
+                start=parse_ts_millis(config.fixedRange.startTimeMillis),
+                end=parse_ts_millis(config.fixedRange.endTimeMillis),
+            )
+        ]
+    elif isinstance(config, list):
+        return _try_parse_list_of_exclusion_windows(config)
+    else:
+        raise SDKUsageErrorWithExamples(
+            msg=f"Invalid exclusion window: {config}",
+            examples=FIXED_RANGE_EXCLUSION_WINDOW_EXAMPLES,
+        )
+
+
+def _try_parse_list_of_exclusion_windows(
+    config: IterableExclusionWindowInputTypes,
+) -> Union[list[FixedRangeExclusionWindow], None]:
+    if all(isinstance(item, models.AssertionExclusionWindowClass) for item in config):
+        exclusion_windows = []
+        for item in config:
+            assert isinstance(item, models.AssertionExclusionWindowClass)
+            assert item.fixedRange is not None
+            exclusion_windows.append(
+                FixedRangeExclusionWindow(
+                    start=parse_ts_millis(item.fixedRange.startTimeMillis),
+                    end=parse_ts_millis(item.fixedRange.endTimeMillis),
+                )
+            )
+        return exclusion_windows
+    else:
+        exclusion_windows = []
+        for item in config:
+            if isinstance(item, dict):
+                try:
+                    exclusion_windows.append(FixedRangeExclusionWindow(**item))
+                except ValidationError as e:
+                    raise SDKUsageErrorWithExamples(
+                        msg=f"Invalid exclusion window: {item}",
+                        examples=FIXED_RANGE_EXCLUSION_WINDOW_EXAMPLES,
+                    ) from e
+            elif isinstance(item, FixedRangeExclusionWindow):
+                exclusion_windows.append(item)
+            elif item is None:
+                pass
+            else:
+                raise SDKUsageErrorWithExamples(
+                    msg=f"Invalid exclusion window: {item}",
+                    examples=FIXED_RANGE_EXCLUSION_WINDOW_EXAMPLES,
+                )
+    return exclusion_windows
+
+
+class AssertionIncidentBehavior(Enum):
+    RAISE_ON_FAIL = "raise_on_fail"
+    RESOLVE_ON_PASS = "resolve_on_pass"
+
+
+ASSERTION_INCIDENT_BEHAVIOR_EXAMPLES = {
+    "Raise on fail from string": "raise_on_fail",
+    "Raise on fail from enum": "AssertionIncidentBehavior.RAISE_ON_FAIL",
+    "Resolve on pass from string": "resolve_on_pass",
+    "Resolve on pass from enum": "AssertionIncidentBehavior.RESOLVE_ON_PASS",
+}
+
+AssertionIncidentBehaviorInputTypes: TypeAlias = Union[
+    str,
+    list[str],
+    AssertionIncidentBehavior,
+    list[AssertionIncidentBehavior],
+    None,
+]
+
+
+def _try_parse_incident_behavior(
+    config: AssertionIncidentBehaviorInputTypes,
+) -> Union[AssertionIncidentBehavior, list[AssertionIncidentBehavior], None]:
+    if config is None:
+        return []
+    if isinstance(config, str):
+        try:
+            return [AssertionIncidentBehavior(config)]
+        except ValueError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid incident behavior: {config}",
+                examples=ASSERTION_INCIDENT_BEHAVIOR_EXAMPLES,
+            ) from e
+    if isinstance(config, AssertionIncidentBehavior):
+        return [config]
+    elif isinstance(config, list):
+        incident_behaviors = []
+        for item in config:
+            if isinstance(item, str):
+                try:
+                    incident_behaviors.append(AssertionIncidentBehavior(item))
+                except ValueError as e:
+                    raise SDKUsageErrorWithExamples(
+                        msg=f"Invalid incident behavior: {item}",
+                        examples=ASSERTION_INCIDENT_BEHAVIOR_EXAMPLES,
+                    ) from e
+            elif isinstance(item, AssertionIncidentBehavior):
+                incident_behaviors.append(item)
+            else:
+                raise SDKUsageErrorWithExamples(
+                    msg=f"Invalid incident behavior: {item}",
+                    examples=ASSERTION_INCIDENT_BEHAVIOR_EXAMPLES,
+                )
+        return incident_behaviors
+    else:
+        raise SDKUsageErrorWithExamples(
+            msg=f"Invalid incident behavior: {config}",
+            examples=ASSERTION_INCIDENT_BEHAVIOR_EXAMPLES,
+        )
+
+
+def _generate_default_name(prefix: str, suffix_length: int) -> str:
+    return f"{prefix}-{''.join(random.choices(string.ascii_letters + string.digits, k=suffix_length))}"
+
+
+TRAINING_DATA_LOOKBACK_DAYS_EXAMPLES = {
+    "Training data lookback days from int": ASSERTION_MONITOR_DEFAULT_TRAINING_LOOKBACK_WINDOW_DAYS,
+    f"Training data lookback days from None (uses default of {ASSERTION_MONITOR_DEFAULT_TRAINING_LOOKBACK_WINDOW_DAYS} days)": None,
+}
+
+
+def _try_parse_training_data_lookback_days(
+    training_data_lookback_days: Optional[int],
+) -> int:
+    if training_data_lookback_days is None:
+        return ASSERTION_MONITOR_DEFAULT_TRAINING_LOOKBACK_WINDOW_DAYS
+    if isinstance(training_data_lookback_days, str):
+        try:
+            training_data_lookback_days = int(training_data_lookback_days)
+        except ValueError as e:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid training data lookback days: {training_data_lookback_days}",
+                examples=TRAINING_DATA_LOOKBACK_DAYS_EXAMPLES,
+            ) from e
+    if not isinstance(training_data_lookback_days, int):
+        raise SDKUsageErrorWithExamples(
+            msg=f"Invalid training data lookback days: {training_data_lookback_days}",
+            examples=TRAINING_DATA_LOOKBACK_DAYS_EXAMPLES,
+        )
+    if training_data_lookback_days < 0:
+        raise SDKUsageError("Training data lookback days must be non-negative")
+    return training_data_lookback_days
+
+
+def _validate_cron_schedule(schedule: str, timezone: str) -> None:
+    """We are using the POSIX.1-2017 standard for cron expressions.
+
+    Note: We are using the croniter library for cron parsing which is different from executor, which uses apscheduler, so there is a risk of mismatch here.
+    """
+    try:
+        # Validate timezone - pytz.timezone() raises UnknownTimeZoneError for invalid timezones
+        # Skip timezone validation when empty
+        if timezone:
+            pytz.timezone(timezone)
+
+        # Validate 5-field cron expression only (POSIX.1-2017 standard)
+        fields = schedule.strip().split()
+        if len(fields) != 5:
+            raise ValueError("POSIX.1-2017 requires exactly 5 fields")
+
+        # POSIX.1-2017 specific validation: Sunday must be 0, not 7
+        # However croniter accepts 7 as Sunday, so custom check is needed here.
+        # Check the day-of-week field (5th field, index 4)
+        dow_field = fields[4]
+        if "7" in dow_field:
+            # Check if 7 appears as a standalone value or in ranges
+            import re
+
+            # Match 7 as standalone, in lists, or in ranges
+            if re.search(r"\b7\b|7-|,7,|^7,|,7$|-7\b", dow_field):
+                raise ValueError(
+                    "POSIX.1-2017 standard: Sunday must be represented as 0, not 7"
+                )
+
+        # Validate cron expression - croniter constructor validates the expression
+        croniter(schedule)
+
+    except Exception as e:
+        raise SDKUsageError(
+            f"Invalid cron expression or timezone: {schedule} {timezone}, please use a POSIX.1-2017 compatible cron expression and timezone."
+        ) from e
+
+
+def _try_parse_schedule(
+    schedule: Optional[Union[str, models.CronScheduleClass]],
+) -> Optional[models.CronScheduleClass]:
+    if schedule is None:
+        return None
+    if isinstance(schedule, str):
+        _validate_cron_schedule(schedule, "UTC")
+        return models.CronScheduleClass(
+            cron=schedule,
+            timezone="UTC",
+        )
+    if isinstance(schedule, models.CronScheduleClass):
+        _validate_cron_schedule(schedule.cron, schedule.timezone)
+        return schedule
+
+
+FieldSpecType = Union[models.FreshnessFieldSpecClass, models.SchemaFieldSpecClass]
+
+
+T = TypeVar("T")
+
+
+def _try_parse_and_validate_schema_classes_enum(
+    value: Union[str, T],
+    enum_class: Type[T],
+) -> T:
+    if isinstance(value, enum_class):
+        return value
+    assert isinstance(value, str)
+    if value.upper() not in get_enum_options(enum_class):
+        raise SDKUsageError(
+            f"Invalid value for {enum_class.__name__}: {value}, valid options are {get_enum_options(enum_class)}"
+        )
+    return getattr(enum_class, value.upper())
+
+
+def get_gms_type_if_criteria_unchanged(
+    criteria_parameters: Union[str, int, float],
+    gms_type_info: Optional[Union[models.AssertionStdParameterTypeClass, tuple]],
+) -> Optional[models.AssertionStdParameterTypeClass]:
+    """
+    Get the GMS type for criteria, but only if the user hasn't changed the value.
+
+    When updating an assertion, we want to preserve the stored type metadata only
+    if the criteria value wasn't changed. If the user provided a new value, we
+    should infer the type from their input instead.
+
+    Args:
+        criteria_parameters: The criteria value (user-provided or extracted from GMS).
+        gms_type_info: Type info from GMS in format (value, type).
+
+    Returns:
+        The GMS type if criteria is unchanged, None if user provided new criteria.
+    """
+    if gms_type_info is None:
+        return None
+
+    # Validate format: (value, type) where neither is a tuple
+    if not (
+        isinstance(gms_type_info, tuple)
+        and len(gms_type_info) >= 2
+        and not isinstance(gms_type_info[0], tuple)
+        and not isinstance(gms_type_info[1], tuple)
+    ):
+        return None
+
+    gms_value, gms_type = gms_type_info[0], gms_type_info[1]
+
+    # If values match, criteria was extracted (not user-provided) - use GMS type
+    if criteria_parameters == gms_value:
+        return gms_type
+
+    # Values differ - user provided new criteria, caller should infer type
+    return None
+
+
+def get_gms_types_if_criteria_unchanged(
+    criteria_parameters: tuple,
+    gms_type_info: Optional[Union[models.AssertionStdParameterTypeClass, tuple]],
+) -> Optional[tuple]:
+    """
+    Get the GMS types for range criteria, but only if the user hasn't changed the values.
+
+    Args:
+        criteria_parameters: The range values (min, max).
+        gms_type_info: Type info from GMS in format ((min_val, max_val), (min_type, max_type)).
+
+    Returns:
+        The GMS types if criteria is unchanged, None if user provided new criteria.
+    """
+    if gms_type_info is None:
+        return None
+
+    # Validate format: ((min_val, max_val), (min_type, max_type))
+    if not (
+        isinstance(gms_type_info, tuple)
+        and len(gms_type_info) == 2
+        and isinstance(gms_type_info[0], tuple)
+        and isinstance(gms_type_info[1], tuple)
+    ):
+        return None
+
+    gms_values, gms_types = gms_type_info[0], gms_type_info[1]
+
+    # If values match, criteria was extracted (not user-provided) - use GMS types
+    if criteria_parameters == gms_values:
+        return gms_types
+
+    # Values differ - user provided new criteria, caller should infer types
+    return None
+
+
+@dataclass(frozen=True)
+class DatasetSourceType:
+    """
+    DatasetSourceType is used to represent a dataset source type.
+    It is used to check if a source type is valid for a dataset type and assertion type.
+
+    Args:
+        source_type: The source type (e.g. information schema, field value, etc. aka detection mechanism)
+        platform: The platform of the dataset as a string OR "all" for all platforms.
+        assertion_type: The assertion type as a models.AssertionTypeClass string e.g. models.AssertionTypeClass.FRESHNESS OR "all" for all assertion types.
+
+    Example:
+    DatasetSourceType(
+        source_type=_InformationSchema,
+        platform="databricks",
+        assertion_type="all",
+    )
+    This means that the source type _InformationSchema is invalid for the dataset type "databricks" and assertion type "all".
+    "all" in this example means that the source type is invalid for all assertion types.
+    """
+
+    source_type: Type[_DetectionMechanismTypes]
+    platform: str
+    assertion_type: Union[models.AssertionTypeClass, str]
+
+
+INVALID_SOURCE_TYPES = {
+    # Add exceptions here if a source type (detection mechanism) is invalid for a dataset type and assertion type.
+    DatasetSourceType(
+        source_type=_InformationSchema,
+        platform="databricks",
+        assertion_type="all",
+    )
+}
+
+
+def _is_source_type_valid(
+    dataset_source_type: DatasetSourceType,
+    invalid_source_types: set[DatasetSourceType] = INVALID_SOURCE_TYPES,
+) -> bool:
+    for invalid in invalid_source_types:
+        if invalid.source_type == dataset_source_type.source_type:
+            # If both platform and assertion type are "all", the source type is invalid for all combinations
+            if invalid.platform == "all" and invalid.assertion_type == "all":
+                return False
+            # If platform matches and assertion type is "all", the source type is invalid for all assertion types on that platform
+            if (
+                invalid.platform == dataset_source_type.platform
+                and invalid.assertion_type == "all"
+            ):
+                return False
+            # If platform is "all" and assertion type matches, the source type is invalid for all platforms for that assertion type
+            if (
+                invalid.platform == "all"
+                and invalid.assertion_type == dataset_source_type.assertion_type
+            ):
+                return False
+            # If both platform and assertion type match exactly, the source type is invalid
+            if (
+                invalid.platform == dataset_source_type.platform
+                and invalid.assertion_type == dataset_source_type.assertion_type
+            ):
+                return False
+    return True
+
+
+class _HasSmartAssertionInputs:
+    """
+    A class that contains the common inputs for smart assertions.
+    This is used to avoid code duplication in the smart assertion inputs.
+
+    Args:
+        sensitivity: The sensitivity to be applied to the assertion.
+        exclusion_windows: The exclusion windows to be applied to the assertion. If not provided, no exclusion windows will be applied.
+        training_data_lookback_days: The training data lookback days to be applied to the assertion.
+    """
+
+    def __init__(
+        self,
+        *,
+        sensitivity: Optional[Union[str, InferenceSensitivity]] = None,
+        exclusion_windows: Optional[ExclusionWindowInputTypes] = None,
+        training_data_lookback_days: Optional[int] = None,
+    ):
+        self.sensitivity = InferenceSensitivity.parse(sensitivity)
+        self.exclusion_windows = _try_parse_exclusion_window(exclusion_windows)
+        self.training_data_lookback_days = _try_parse_training_data_lookback_days(
+            training_data_lookback_days
+        )
+
+    def _convert_exclusion_windows(
+        self,
+    ) -> list[models.AssertionExclusionWindowClass]:
+        """
+        Convert exclusion windows into AssertionExclusionWindowClass objects including generating display names for them.
+
+        Returns:
+            A list of AssertionExclusionWindowClass objects.
+
+        Raises:
+            SDKUsageErrorWithExamples: If an exclusion window is of an invalid type.
+        """
+        exclusion_windows: list[models.AssertionExclusionWindowClass] = []
+        if self.exclusion_windows:
+            for window in self.exclusion_windows:
+                if not isinstance(window, FixedRangeExclusionWindow):
+                    raise SDKUsageErrorWithExamples(
+                        msg=f"Invalid exclusion window type: {window}",
+                        examples=FIXED_RANGE_EXCLUSION_WINDOW_EXAMPLES,
+                    )
+                # To match the UI, we generate a display name for the exclusion window.
+                # See here for the UI code: https://github.com/acryldata/datahub-fork/blob/acryl-main/datahub-web-react/src/app/entityV2/shared/tabs/Dataset/Validations/assertion/builder/steps/inferred/common/ExclusionWindowAdjuster.tsx#L31
+                # Copied here for reference: displayName: `${dayjs(startTime).format('MMM D, h:mm A')} - ${dayjs(endTime).format('MMM D, h:mm A')}`,
+                generated_display_name = f"{window.start.strftime('%b %-d, %-I:%M %p')} - {window.end.strftime('%b %-d, %-I:%M %p')}"
+                exclusion_windows.append(
+                    models.AssertionExclusionWindowClass(
+                        type=models.AssertionExclusionWindowTypeClass.FIXED_RANGE,  # Currently only fixed range is supported
+                        displayName=generated_display_name,
+                        fixedRange=models.AbsoluteTimeWindowClass(
+                            startTimeMillis=make_ts_millis(window.start),
+                            endTimeMillis=make_ts_millis(window.end),
+                        ),
+                    )
+                )
+        return exclusion_windows
+
+    def _convert_sensitivity(self) -> models.AssertionMonitorSensitivityClass:
+        """
+        Convert sensitivity into an AssertionMonitorSensitivityClass.
+
+        Returns:
+            An AssertionMonitorSensitivityClass with the appropriate sensitivity.
+        """
+        return models.AssertionMonitorSensitivityClass(
+            level=InferenceSensitivity.to_int(self.sensitivity),
+        )
+
+
+class _AssertionInput(ABC):
+    def __init__(
+        self,
+        *,
+        dataset_urn: Union[str, DatasetUrn],
+        entity_client: EntityClient,  # Needed to get the schema field spec for the detection mechanism if needed
+        urn: Optional[
+            Union[str, AssertionUrn]
+        ] = None,  # Can be None if the assertion is not yet created
+        display_name: Optional[str] = None,
+        enabled: bool = True,
+        schedule: Optional[Union[str, models.CronScheduleClass]] = None,
+        detection_mechanism: DetectionMechanismInputTypes = None,
+        incident_behavior: Optional[AssertionIncidentBehaviorInputTypes] = None,
+        tags: Optional[TagsInputType] = None,
+        source_type: str = models.AssertionSourceTypeClass.NATIVE,  # Verified on init to be a valid enum value
+        created_by: Union[str, CorpUserUrn],
+        created_at: datetime,
+        updated_by: Union[str, CorpUserUrn],
+        updated_at: datetime,
+        default_detection_mechanism: _DetectionMechanismTypes = DEFAULT_DETECTION_MECHANISM,
+    ):
+        """
+        Create an AssertionInput object.
+
+        Args:
+            dataset_urn: The urn of the dataset to be monitored.
+            entity_client: The entity client to be used for creating the assertion.
+            urn: The urn of the assertion. If not provided, a random urn will be generated.
+            display_name: The display name of the assertion. If not provided, a random display name will be generated.
+            enabled: Whether the assertion is enabled. Defaults to True.
+            detection_mechanism: The detection mechanism to be used for the assertion.
+            incident_behavior: The incident behavior to be applied to the assertion. Accepts:
+                - String values: "raise_on_fail", "resolve_on_pass"
+                - Enum values: AssertionIncidentBehavior.RAISE_ON_FAIL, AssertionIncidentBehavior.RESOLVE_ON_PASS
+                - Lists of any of the above values
+                - None (default behavior)
+            tags: The tags to be applied to the assertion.
+            source_type: The source type of the assertion. Defaults to models.AssertionSourceTypeClass.NATIVE.
+            created_by: The actor that created the assertion.
+            created_at: The timestamp of the assertion creation.
+            updated_by: The actor that last updated the assertion.
+            updated_at: The timestamp of the assertion last update.
+        """
+        self.dataset_urn = DatasetUrn.from_string(dataset_urn)
+        self.entity_client = entity_client
+        self.urn = AssertionUrn(urn) if urn else None
+        self.display_name = (
+            display_name
+            if display_name is not None
+            else _generate_default_name(DEFAULT_NAME_PREFIX, DEFAULT_NAME_SUFFIX_LENGTH)
+        )
+        self.enabled = enabled
+        self.schedule = _try_parse_schedule(schedule)
+        self.detection_mechanism = DetectionMechanism.parse(
+            detection_mechanism, default_detection_mechanism
+        )
+        if not _is_source_type_valid(
+            DatasetSourceType(
+                source_type=type(self.detection_mechanism),
+                platform=self.dataset_urn.platform,
+                assertion_type=self._assertion_type(),
+            )
+        ):
+            raise SDKUsageError(
+                f"Invalid source type: {self.detection_mechanism} for dataset type: {self.dataset_urn.platform} and assertion type: {self._assertion_type()}"
+            )
+        self.incident_behavior = _try_parse_incident_behavior(incident_behavior)
+        self.tags = tags
+        if source_type not in get_enum_options(models.AssertionSourceTypeClass):
+            raise SDKUsageError(
+                msg=f"Invalid source type: {source_type}, valid options are {get_enum_options(models.AssertionSourceTypeClass)}",
+            )
+        self.source_type = source_type
+        self.created_by = created_by
+        self.created_at = created_at
+        self.updated_by = updated_by
+        self.updated_at = updated_at
+        self.cached_dataset: Optional[Dataset] = None
+
+    def to_assertion_and_monitor_entities(self) -> tuple[Assertion, Monitor]:
+        """
+        Convert the assertion input to an assertion and monitor entity.
+
+        Returns:
+            A tuple of (assertion, monitor) entities.
+        """
+        assertion = self.to_assertion_entity()
+        monitor = self.to_monitor_entity(assertion.urn)
+        return assertion, monitor
+
+    def to_assertion_entity(self) -> Assertion:
+        """
+        Convert the assertion input to an assertion entity.
+
+        Returns:
+            The created assertion entity.
+        """
+        on_success, on_failure = self._convert_incident_behavior()
+        filter = self._create_filter_from_detection_mechanism()
+
+        return Assertion(
+            id=self.urn,
+            info=self._create_assertion_info(filter),
+            description=self.display_name,
+            on_success=on_success,
+            on_failure=on_failure,
+            tags=self._convert_tags(),
+            source=self._convert_source(),
+            last_updated=self._convert_last_updated(),
+        )
+
+    def _convert_incident_behavior(
+        self,
+    ) -> tuple[
+        Optional[AssertionActionsInputType],
+        Optional[AssertionActionsInputType],
+    ]:
+        """
+        Convert incident behavior to on_success and on_failure actions.
+
+        Returns:
+            A tuple of (on_success, on_failure) actions.
+        """
+        if not self.incident_behavior:
+            return None, None
+
+        behaviors = (
+            [self.incident_behavior]
+            if isinstance(self.incident_behavior, AssertionIncidentBehavior)
+            else self.incident_behavior
+        )
+
+        on_success: Optional[AssertionActionsInputType] = [
+            models.AssertionActionClass(
+                type=models.AssertionActionTypeClass.RESOLVE_INCIDENT
+            )
+            for behavior in behaviors
+            if behavior == AssertionIncidentBehavior.RESOLVE_ON_PASS
+        ] or None
+
+        on_failure: Optional[AssertionActionsInputType] = [
+            models.AssertionActionClass(
+                type=models.AssertionActionTypeClass.RAISE_INCIDENT
+            )
+            for behavior in behaviors
+            if behavior == AssertionIncidentBehavior.RAISE_ON_FAIL
+        ] or None
+
+        return on_success, on_failure
+
+    def _create_filter_from_detection_mechanism(
+        self,
+    ) -> Optional[models.DatasetFilterClass]:
+        """
+        Create a filter from the detection mechanism if it has an additional filter.
+
+        Returns:
+            A DatasetFilterClass if the detection mechanism has an additional filter, None otherwise.
+        """
+        if not isinstance(
+            self.detection_mechanism,
+            _DETECTION_MECHANISM_TYPES_WITH_ADDITIONAL_FILTER,
+        ):
+            return None
+
+        additional_filter = self.detection_mechanism.additional_filter
+        if not additional_filter:
+            return None
+
+        return models.DatasetFilterClass(
+            type=models.DatasetFilterTypeClass.SQL,
+            sql=additional_filter,
+        )
+
+    def _convert_tags(self) -> Optional[TagsInputType]:
+        """
+        Convert the tags input into a standardized format.
+
+        Tag names are automatically converted to tag URNs using TagUrn constructor. For example:
+        - "my_tag" becomes "urn:li:tag:my_tag"
+        - "urn:li:tag:my_tag" remains unchanged
+
+        Returns:
+            A list of tags or None if no tags are provided.
+
+        Raises:
+            SDKUsageErrorWithExamples: If the tags input is invalid.
+        """
+        if not self.tags:
+            return None
+
+        if isinstance(self.tags, str):
+            return [str(TagUrn(self.tags))]
+        elif isinstance(self.tags, list):
+            return [
+                str(TagUrn(tag)) if isinstance(tag, str) else tag for tag in self.tags
+            ]
+        else:
+            raise SDKUsageErrorWithExamples(
+                msg=f"Invalid tags: {self.tags}",
+                examples={
+                    "Tags from string (tag name)": "my_tag_1",
+                    "Tags from string (tag URN)": "urn:li:tag:my_tag_1",
+                    "Tags from list (mixed)": cast(
+                        Collection[str],
+                        [
+                            "my_tag_1",
+                            "urn:li:tag:my_tag_2",
+                        ],
+                    ),
+                },
+            )
+
+    def _convert_source(self) -> models.AssertionSourceClass:
+        """
+        Convert the source input into a models.AssertionSourceClass.
+        """
+        return models.AssertionSourceClass(
+            type=self.source_type,
+            created=models.AuditStampClass(
+                time=make_ts_millis(self.created_at),
+                actor=str(self.created_by),
+            ),
+        )
+
+    def _convert_last_updated(self) -> tuple[datetime, str]:
+        """
+        Convert the last updated input into a tuple of (datetime, str).
+
+        Validation is handled in the Assertion entity constructor.
+        """
+        return (self.updated_at, str(self.updated_by))
+
+    def to_monitor_entity(self, assertion_urn: AssertionUrn) -> Monitor:
+        """
+        Convert the assertion input to a monitor entity.
+
+        Args:
+            assertion_urn: The URN of the assertion to monitor.
+
+        Returns:
+            A Monitor entity configured with the assertion input parameters.
+        """
+        return Monitor(
+            id=(self.dataset_urn, assertion_urn),
+            info=self._create_monitor_info(
+                assertion_urn=assertion_urn,
+                status=self._convert_monitor_status(),
+                schedule=self._convert_schedule(),
+            ),
+        )
+
+    def _convert_monitor_status(self) -> models.MonitorStatusClass:
+        """
+        Convert the enabled flag into a MonitorStatusClass.
+
+        Returns:
+            A MonitorStatusClass with ACTIVE or INACTIVE mode based on the enabled flag.
+        """
+        return models.MonitorStatusClass(
+            mode=models.MonitorModeClass.ACTIVE
+            if self.enabled
+            else models.MonitorModeClass.INACTIVE,
+        )
+
+    def _get_schema_field_spec(self, column_name: str) -> models.SchemaFieldSpecClass:
+        """
+        Get the schema field spec for the detection mechanism if needed.
+        """
+        # Only fetch the dataset if it's not already cached.
+        # Also we only fetch the dataset if it's needed for the detection mechanism.
+        if self.cached_dataset is None:
+            self.cached_dataset = self.entity_client.get(self.dataset_urn)
+
+        # Handle case where dataset doesn't exist
+        if self.cached_dataset is None:
+            raise SDKUsageError(
+                f"Dataset {self.dataset_urn} not found. Cannot validate column {column_name}."
+            )
+
+        # TODO: Make a public accessor for _schema_dict in the SDK
+        schema_fields = self.cached_dataset._schema_dict()
+        field = schema_fields.get(column_name)
+        if field:
+            return self._convert_schema_field_to_schema_field_spec(field)
+        else:
+            raise SDKUsageError(
+                msg=f"Column {column_name} not found in dataset {self.dataset_urn}",
+            )
+
+    def _convert_schema_field_to_schema_field_spec(
+        self, field: models.SchemaFieldClass
+    ) -> models.SchemaFieldSpecClass:
+        """
+        Convert a SchemaFieldClass to a SchemaFieldSpecClass.
+        """
+        type_class_name = field.type.type.__class__.__name__
+        try:
+            type = self._convert_schema_field_type_class_name_to_type(type_class_name)
+        except KeyError as e:
+            raise SDKUsageError(
+                msg=f"Invalid type: {type_class_name}. Must be one of {list(TYPE_CLASS_NAME_TO_TYPE_MAP.keys())}",
+            ) from e
+
+        return models.SchemaFieldSpecClass(
+            path=field.fieldPath,
+            type=type,
+            nativeType=field.nativeDataType,
+        )
+
+    def _convert_schema_field_type_class_name_to_type(
+        self, type_class_name: str
+    ) -> str:
+        """
+        Convert a type class name to a type.
+        """
+        return TYPE_CLASS_NAME_TO_TYPE_MAP[type_class_name]
+
+    def _validate_field_type(
+        self,
+        field_spec: models.SchemaFieldSpecClass,
+        column_name: str,
+        allowed_types: list[DictWrapper],
+        field_type_name: str,
+    ) -> None:
+        """
+        Validate that a field has an allowed type.
+
+        Args:
+            field_spec: The field specification to validate
+            column_name: The name of the column for error messages
+            allowed_types: List of allowed field types
+            field_type_name: Human-readable name of the field type for error messages
+
+        Raises:
+            SDKUsageError: If the field has an invalid type
+        """
+        allowed_type_names = [
+            self._convert_schema_field_type_class_name_to_type(t.__class__.__name__)
+            for t in allowed_types
+        ]
+        if field_spec.type not in allowed_type_names:
+            raise SDKUsageError(
+                msg=f"Column {column_name} with type {field_spec.type} does not have an allowed type for a {field_type_name} in dataset {self.dataset_urn}. "
+                f"Allowed types are {allowed_type_names}.",
+            )
+
+    @abstractmethod
+    def _create_monitor_info(
+        self,
+        assertion_urn: AssertionUrn,
+        status: models.MonitorStatusClass,
+        schedule: models.CronScheduleClass,
+    ) -> models.MonitorInfoClass:
+        """
+        Create a MonitorInfoClass with all the necessary components.
+
+        Args:
+            status: The monitor status.
+            schedule: The monitor schedule.
+        Returns:
+            A MonitorInfoClass configured with all the provided components.
+        """
+        pass
+
+    @abstractmethod
+    def _assertion_type(self) -> str:
+        """Get the assertion type."""
+        pass
+
+    @abstractmethod
+    def _create_assertion_info(
+        self, filter: Optional[models.DatasetFilterClass]
+    ) -> AssertionInfoInputType:
+        """Create assertion info specific to the assertion type."""
+        pass
+
+    @abstractmethod
+    def _convert_schedule(self) -> models.CronScheduleClass:
+        """Convert schedule to appropriate format for the assertion type."""
+        pass
+
+    @abstractmethod
+    def _get_assertion_evaluation_parameters(
+        self, source_type: str, field: Optional[FieldSpecType]
+    ) -> models.AssertionEvaluationParametersClass:
+        """Get evaluation parameters specific to the assertion type."""
+        pass
+
+    @abstractmethod
+    def _convert_assertion_source_type_and_field(
+        self,
+    ) -> tuple[str, Optional[FieldSpecType]]:
+        """Convert detection mechanism to source type and field spec."""
+        pass
+
+
+class _HasFreshnessFeatures:
+    def _create_field_spec(
+        self,
+        column_name: str,
+        allowed_types: list[DictWrapper],  # TODO: Use the type from the PDL
+        field_type_name: str,
+        kind: str,
+        get_schema_field_spec: Callable[[str], models.SchemaFieldSpecClass],
+        validate_field_type: Callable[
+            [models.SchemaFieldSpecClass, str, list[DictWrapper], str], None
+        ],
+    ) -> models.FreshnessFieldSpecClass:
+        """
+        Create a field specification for a column, validating its type.
+
+        Args:
+            column_name: The name of the column to create a spec for
+            allowed_types: List of allowed field types
+            field_type_name: Human-readable name of the field type for error messages
+            kind: The kind of field to create
+
+        Returns:
+            A FreshnessFieldSpecClass for the column
+
+        Raises:
+            SDKUsageError: If the column is not found or has an invalid type
+        """
+        SUPPORTED_KINDS = [
+            models.FreshnessFieldKindClass.LAST_MODIFIED,
+            models.FreshnessFieldKindClass.HIGH_WATERMARK,
+        ]
+        if kind not in SUPPORTED_KINDS:
+            raise SDKUsageError(
+                msg=f"Invalid kind: {kind}. Must be one of {SUPPORTED_KINDS}",
+            )
+
+        field_spec = get_schema_field_spec(column_name)
+        validate_field_type(field_spec, column_name, allowed_types, field_type_name)
+        return models.FreshnessFieldSpecClass(
+            path=field_spec.path,
+            type=field_spec.type,
+            nativeType=field_spec.nativeType,
+            kind=kind,
+        )