Code import - branch release/SIENTIAPDE-1646

laborious/activities/activities.py

@@ -0,0 +1,177 @@
+from sientia_do.observability.metrics_controller import MetricsController
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from typing import Any
+
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.observability.logger import Logger
+    from sientia_do.repository.minio_repository_sync import MinioRepository
+    from sientia_model.model_repository.mlflow_repository import SientiaMLflowRepository
+    from sientia_model.model_repository.plugin_store import PluginStore
+
+    from laborious.activities.api import API
+    from laborious.activities.gates import Gates
+    from laborious.activities.mlflow import MLFlow
+    from laborious.activities.model_metrics import ModelMetrics
+    from laborious.activities.opc import OPC
+    from laborious.activities.storage import Storage
+    from laborious.utils.connectors_config import build_mlflow_config
+
+
+class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
+    """
+    Central orchestrator for all Temporal activities used by Laborious workflows.
+
+    Composes Storage (Postgres + MinIO offload), MLFlow (wrapper-based inference and retrain
+    via ``SientiaMLflowRepository``), Gates (data quality and ML response filters), OPC exports,
+    drift/simple metrics, and PI Web API writes. The worker constructs one ``Activities`` instance
+    per process and registers its callables on multiple workers bound to different task queues.
+
+    MLflow connectivity: unless ``mlflow_repository`` is injected (tests only), this class builds
+    ``SientiaMLflowRepository`` from ``build_mlflow_config()`` so tracking credentials and URL
+    stay aligned with the rest of Laborious env-based configuration.
+
+    Attributes:
+        Inherits and exposes behaviour from mixins; the MLFlow mixin holds ``mlflow_repository``
+        and ``plugin_store`` after ``__init__``.
+    """
+
+    def __init__(
+        self,
+        postgres_config: dict[str, Any],
+        plugin_store: PluginStore,
+        minio_config: dict[str, Any],
+        opc_config: dict[str, Any],
+        pi_web_api_config: dict[str, Any],
+        logger: Logger,
+        notification_handler: NotificationHandler,
+        metrics_controller: MetricsController | None = None,
+        mlflow_repository: SientiaMLflowRepository | None = None,
+    ):
+        """
+        Wire Postgres, MinIO, MLflow, OPC, gates, metrics, and PI Web API into a single object.
+
+        A single ``MetricsController`` instance is created (or reused) and passed to MinIO,
+        MLflow repository, and all mixins so Prometheus and SDK metrics stay consistent.
+
+        Args:
+         - postgres_config: Host, port, credentials, db name, and pool bounds for Storage.
+         - plugin_store: ``PluginStore`` instance; the worker must call ``install_runtime`` before
+           activities run so wrapper code is importable.
+         - minio_config: Endpoint, keys, bucket, retention, and TLS flag for object storage payloads.
+         - opc_config: Map of OPC server id to connection settings for ``OPC`` mixin.
+         - pi_web_api_config: Base URL and auth for ``API`` mixin.
+         - logger: Structured logger used across all activities.
+         - notification_handler: Handler for alerts and persisted notifications.
+         - metrics_controller: Optional shared controller; if ``None``, a new one is created.
+         - mlflow_repository: Optional ``SientiaMLflowRepository`` for unit/e2e tests; in production
+           leave unset so the repository is built from environment via ``build_mlflow_config()``.
+
+        Raises:
+            Exception: If any parent ``__init__`` fails (e.g. invalid config keys).
+
+        Return:
+            None
+        """
+
+        mc = metrics_controller or MetricsController(logger=logger)
+
+        # Production path: one shared MLflow client for all model registry / tracking calls.
+        if mlflow_repository is None:
+            mlflow_cfg = build_mlflow_config()
+            mlflow_repository = SientiaMLflowRepository(
+                host=mlflow_cfg['url'],
+                username=mlflow_cfg['username'],
+                password=mlflow_cfg['password'],
+                logger=logger,
+                notification_handler=notification_handler,
+                metrics_controller=mc,
+            )
+
+        minio_repository = MinioRepository(
+            endpoint=minio_config['endpoint_url'],
+            access_key=minio_config['access_key'],
+            secret_key=minio_config['secret_key'],
+            bucket=minio_config['default_bucket'],
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+            secure=minio_config['secure'],
+        )
+
+        Storage.__init__(
+            self,
+            host=postgres_config['host'],
+            port=postgres_config['port'],
+            user=postgres_config['user'],
+            password=postgres_config['password'],
+            dbname=postgres_config['dbname'],
+            min_connections=postgres_config['min_connections'],
+            max_connections=postgres_config['max_connections'],
+            retention_hours=minio_config['retention_hours'],
+            minio_repository=minio_repository,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        MLFlow.__init__(
+            self,
+            mlflow_repository=mlflow_repository,
+            plugin_store=plugin_store,
+            minio_repository=minio_repository,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        Gates.__init__(
+            self,
+            minio_repository=minio_repository,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        OPC.__init__(
+            self,
+            opc_servers=opc_config,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        ModelMetrics.__init__(
+            self,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        API.__init__(
+            self,
+            base_url=pi_web_api_config['base_url'],
+            auth_type=pi_web_api_config['auth_type'],
+            auth_token=pi_web_api_config['auth_token'],
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+    def shutdown(self) -> None:
+        """
+        Close database pools, sync clients, and OPC sessions in a defined order.
+
+        Should be invoked on worker exit so connection pools and OPC sessions are released
+        cleanly before process termination.
+
+        Return:
+            None
+        """
+        Storage.close(self)
+        MLFlow.close(self)
+        Gates.close(self)
+        OPC.close(self)
+        ModelMetrics.close(self)
+        API.close(self)

laborious/activities/api.py

@@ -0,0 +1,305 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    import json
+    import traceback
+    from typing import Any
+
+    from pandas import DataFrame
+    from sientia_do.notifications.handlers import NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+    from sientia_do.repository.pi_web_api_client_sync import PIWebAPIClient
+
+    from laborious import metrics
+
+
+PI_WEB_API_PREDICTION_ERROR_CONFIDENCE = 13
+
+
+class API(SientiaMonitoring):
+    """
+    PI Web API operations for writing prediction data to PI Web API.
+
+    This class provides Temporal activities for interacting with the PI Web API
+    to write prediction and confidence values to industrial systems. It handles
+    error scenarios gracefully by setting error confidence values and sending
+    notifications when write operations fail.
+
+    The class implements comprehensive error handling for both prediction and
+    confidence value writes, ensuring that partial failures are properly
+    reported and handled.
+    """
+
+    def __init__(
+        self,
+        base_url: str,
+        auth_type: str,
+        auth_token: str,
+        logger: Logger,
+        notification_handler: NotificationHandler,
+        metrics_controller: MetricsController,
+    ) -> None:
+        """
+        Initialize API activity with PI Web API client.
+
+        Args:
+            base_url (str): Base URL of the PI Web API server
+            auth_type (str): Authentication type ('basic' or 'bearer')
+            auth_token (str): Authentication token
+            logger (Logger): Logger instance for operation logging
+            notification_handler (NotificationHandler): Handler for system notifications
+            metrics_controller (MetricsController): Controller for metrics collection
+        """
+        SientiaMonitoring.__init__(self, logger, notification_handler, metrics_controller)
+        self.pi_web_api_client = PIWebAPIClient(
+            base_url=base_url,
+            auth_config={
+                'type': auth_type,
+                'token': auth_token,
+            },
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+            headers_config={
+                'Content-Type': 'application/json',
+                'Accept': 'application/json',
+                'x-requested-with': 'piwebapistreams',
+                'User-Agent': 'Aig-Laborious-Agent/1.0',
+            },
+        )
+
+    def get_pi_web_api_core_labels(
+        self,
+        metadata: dict[str, Any],
+        operation_type: str = 'write_pi_web_api_data',
+    ) -> dict[str, Any]:
+        """
+        Generate core labels for PI Web API metrics.
+
+        PI Web API metrics in laborious use the shared ``CORE_LABELS`` from
+        ``sientia_do``, which includes ``operation_type``. For this reason,
+        operation_type must always be present in emitted labels.
+
+        Args:
+         - metadata (dict[str, Any]): Workflow execution metadata used to derive labels.
+         - operation_type (str): Operation type label for metric cardinality.
+
+        Return:
+         dict[str, Any]: Core labels dictionary including operation_type.
+        """
+        return super().get_core_labels(
+            metadata=metadata,
+            operation_type=operation_type,
+        )
+
+    def close(self) -> None:
+        """
+        Close the PI Web API client and shutdown monitoring services.
+
+        This method properly closes all connections and resources associated
+        with the PI Web API client and monitoring services.
+        """
+        self.pi_web_api_client.close()
+        SientiaMonitoring.shutdown(self)
+
+    def process_pi_web_api_response(
+        self,
+        response_data: list[dict[str, Any]],
+        tags: dict[str, str],
+        core_labels: dict[str, str],
+        metadata: dict[str, Any],
+    ) -> tuple[int, str]:
+        """
+        Process the response data from PI Web API write operation.
+
+        Validates that all tags were successfully written, emits metrics for each tag
+        (success or error), and returns the appropriate prediction confidence value.
+        Sets error confidence if any tag write fails or if the number of written tags
+        doesn't match the expected count.
+
+        Args:
+            - response_data (dict[str, Any]): The response data from the PI Web API write operation.
+            - tags (dict[str, str]): The tags that were written to the PI Web API.
+            - core_labels (dict[str, str]): The core labels of the workflow execution.
+            - metadata (dict[str, Any]): The metadata of the workflow execution.
+        Returns:
+            int: Prediction confidence value (0 for success, 13 for errors)
+        """
+
+        # Convert tags from name:webid to webid:name
+        tags = {w: t for t, w in tags.items()}
+
+        tag_names = list[str](tags.values())
+
+        confidence = 0
+
+        message = ''
+
+        # Evaluate response for each tag
+        written_tags = []
+        for item in response_data:
+            web_id = item.get('WebId')
+            if not web_id:
+                self.error('The response did not contain some WebIds', metadata)
+                continue
+            errors = item.get('Errors', [])
+            tag_name = tags.get(web_id)
+            if not tag_name:
+                self.error(
+                    f'The response did not contain the tag name for WebId {web_id}', metadata
+                )
+                continue
+            if errors:
+                self.error(
+                    f'Error writing tag {tag_name}:{web_id} to PI Web API: {errors}', metadata
+                )
+                self.emit_metric_sync(
+                    metric_object=metrics.PI_WEB_API_PREDICTION_WRITTEN_ERROR_COUNT,
+                    tags={
+                        **core_labels,
+                        'tag_name': tag_name,
+                    },
+                )
+                confidence = PI_WEB_API_PREDICTION_ERROR_CONFIDENCE
+            else:
+                self.emit_metric_sync(
+                    metric_object=metrics.PI_WEB_API_PREDICTION_WRITTEN_COUNT,
+                    tags={
+                        **core_labels,
+                        'tag_name': tag_name,
+                    },
+                )
+                written_tags.append(tag_name)
+
+        if len(written_tags) != len(tag_names):
+            message = f'The number of written tags does not match the number of tag names: Expected {tag_names} tags, but {written_tags} tags were written.'
+
+            self.error(
+                f'{message}\nResponse:\n {json.dumps(response_data, indent=4)}\nTags:\n {json.dumps(tags, indent=4)}',
+                metadata,
+            )
+
+            self.send_notification(
+                metadata=metadata,
+                notification_id='WRITE_PI_WEB_API_PREDICTION_ERROR',
+                message=f'The number of written tags does not match the number of tag names: Expected {tag_names} tags, but {written_tags} tags were written.\nResponse:\n {json.dumps(response_data, indent=4)}\nTags:\n {json.dumps(tags, indent=4)}',
+                block='write_pi_web_api_data',
+                level=NotificationLevel.ERROR,
+            )
+            confidence = PI_WEB_API_PREDICTION_ERROR_CONFIDENCE
+
+        return confidence, message
+
+    @activity.defn(name='write_pi_web_api_data')
+    def write_pi_web_api_data(self, input_data: dict[str, Any]) -> dict[Any, Any]:
+        """
+        Write prediction and confidence data to PI Web API.
+
+        Writes prediction values and confidence scores to PI Web API using configured
+        web IDs. Processes responses to validate writes and emit metrics. Handles errors
+        gracefully by setting error confidence values when writes fail and sending
+        notifications for both prediction and confidence write errors.
+
+        Args:
+            input_data (dict[str, Any]): The input data containing:
+                - metadata (dict[str, Any]): Workflow execution metadata
+                - pi_web_api_output_config (dict[str, Any]): PI Web API configuration with:
+                    - endpoint (str): PI Web API endpoint URL
+                    - prediction_tags (dict[str, str]): Mapping of tag names to web IDs for predictions
+                    - confidence_tags (dict[str, str]): Mapping of tag names to web IDs for confidence
+                - data (dict[str, Any]): Prediction data, its a dataframe converted to dict.
+        Returns:
+            dict[Any, Any]: Data dictionary with potentially modified confidence values
+                If prediction write fails, prediction_confidence is set to error value (13)
+        """
+        metadata = input_data['metadata']
+        data = DataFrame(input_data['data'])
+        pi_web_api_output_config = input_data['pi_web_api_output_config']
+
+        self.info(f'Writing data to PI Web API... config: {pi_web_api_output_config}', metadata)
+
+        raw_prediction_tags = pi_web_api_output_config['prediction_tags']
+        raw_confidence_tags = pi_web_api_output_config['confidence_tags']
+        prediction_tags = list[str](raw_prediction_tags.values())
+        confidence_tags = list(raw_confidence_tags.values())
+
+        core_labels = self.get_pi_web_api_core_labels(metadata)
+
+        prediction_value = data.head(1)['prediction'].values[0]
+        confidence_value = data.head(1)['prediction_confidence'].values[0]
+
+        try:
+            prediction_response = self.pi_web_api_client.write_value(
+                web_ids=prediction_tags,
+                value={
+                    'Timestamp': data.head(1)['timestamp'].values[0],
+                    'Value': prediction_value,
+                },
+                metadata=metadata,
+            )
+
+            confidence, message = self.process_pi_web_api_response(
+                response_data=prediction_response,
+                tags=raw_prediction_tags,
+                core_labels=core_labels,
+                metadata=metadata,
+            )
+
+            # Preserve incoming confidence/comments on successful PI writes.
+            # Only downgrade confidence or override comments when PI response
+            # explicitly reports a problem (e.g. partial write mismatch).
+            if confidence != 0:
+                data['prediction_confidence'] = confidence
+            if message:
+                data['comments'] = message
+
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id='WRITE_PI_WEB_API_PREDICTION_ERROR',
+                message=f'Error writing prediction data to PI Web API: {e}\n Tags: {raw_prediction_tags}',
+                block='write_pi_web_api_data',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+
+            self.error(trace, metadata)
+
+            data['prediction_confidence'] = PI_WEB_API_PREDICTION_ERROR_CONFIDENCE
+            data['comments'] = str(e)
+
+            return data.to_dict()
+
+        try:
+            confidence_response = self.pi_web_api_client.write_value(
+                web_ids=confidence_tags,
+                value={
+                    'Timestamp': data.head(1)['timestamp'].values[0],
+                    'Value': float(confidence_value),
+                },
+                metadata=metadata,
+            )
+
+            self.process_pi_web_api_response(
+                response_data=confidence_response,
+                tags=raw_confidence_tags,
+                core_labels=core_labels,
+                metadata=metadata,
+            )
+
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id='WRITE_PI_WEB_API_CONFIDENCE_ERROR',
+                message=f'Error writing confidence data to PI Web API: {e}\n Tags: {raw_confidence_tags}',
+                block='write_pi_web_api_data',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+
+        return data.to_dict()

laborious/activities/gates.py

@@ -0,0 +1,812 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    import traceback
+    from collections.abc import Callable, Mapping
+    from typing import Any
+
+    from pandas import DataFrame
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+    from sientia_do.repository.minio_repository_sync import MinioRepository
+    from sientia_do.utils.formatters import create_sample_dict
+
+    from laborious import metrics
+    from laborious.utils.dataframe_debug import build_dataframe_debug_message
+    from laborious.utils.filters.conditional_filters import (
+        filter_empty_data,
+        filter_specific_variables_null_values,
+    )
+    from laborious.utils.filters.mlflow_filters import api_error_filter, nan_values_filter
+    from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
+
+# Strongly-typed filter function signatures
+InputFilterFunc = Callable[[DataFrame, dict[str, Any]], bool]
+ResponseFilterFunc = Callable[[dict[str, Any], dict[str, Any]], bool]
+ContentFilterFunc = Callable[[DataFrame, dict[str, Any]], bool]
+
+# Input filter function mappings
+input_filter_functions: dict[str, InputFilterFunc] = {
+    'SPECIFIC_VARIABLES_NULL_VALUES': filter_specific_variables_null_values,
+    'EMPTY_DATA': filter_empty_data,
+}
+
+# Confidence mappings kept separate from function maps to avoid Union types
+input_path_confidence: Mapping[str, int] = {
+    'STOP': -1,
+    'CONTINUE': 2,
+    'REPEAT': -1,
+}
+
+# MLFlow response filter function mappings
+mlflow_response_filter_functions: dict[str, ResponseFilterFunc] = {
+    'API_ERROR': api_error_filter,
+}
+
+mlflow_response_path_confidence: Mapping[str, int] = {
+    'STOP': -1,
+    'CONTINUE': 10,
+    'REPEAT': -1,
+}
+
+# MLFlow content filter function mappings
+mlflow_content_filter_functions: dict[str, ContentFilterFunc] = {
+    'NAN_VALUES': nan_values_filter,
+    'EMPTY_DATA': filter_empty_data,
+}
+
+mlflow_content_path_confidence: Mapping[str, int] = {
+    'STOP': -1,
+    'CONTINUE': 18,
+    'REPEAT': -1,
+}
+
+
+class Gates(SientiaMonitoring):
+    """
+    Data quality gates and filtering activities for the Laborious system.
+
+    This class implements comprehensive data quality validation and filtering
+    mechanisms that can be applied at different stages of the prediction pipeline.
+    It provides configurable filters with policy-based decision making to ensure
+    data integrity and quality throughout the ML workflow.
+
+    The class supports multiple filter types and implements a flexible policy
+    system that can be configured for different validation requirements. Each
+    filter returns a path decision (STOP, CONTINUE, REPEAT) along with confidence
+    scores and detailed comments for monitoring and debugging.
+
+    Attributes:
+        input_filter_functions (dict): Mapping of input filter names to functions
+        mlflow_response_filter_functions (dict): Mapping of MLFlow response filter names to functions
+        mlflow_content_filter_functions (dict): Mapping of MLFlow content filter names to functions
+    """
+
+    minio_repository: MinioRepository | None = None
+    _MAX_DEBUG_DATAFRAME_ROWS = 100
+
+    def __init__(
+        self,
+        minio_repository: MinioRepository | None = None,
+        logger: Logger | None = None,
+        notification_handler: NotificationHandler | None = None,
+        metrics_controller: MetricsController | None = None,
+    ):
+        """
+        Initialize data quality gates with logging and notification capabilities.
+
+        Args:
+            logger: Logger instance for observability and debugging
+            notification_handler: Notification handler for alerts and monitoring
+
+        Raises:
+            Exception: If BaseActivity initialization fails
+        """
+        self.minio_repository = minio_repository
+        SientiaMonitoring.__init__(
+            self,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+        )
+
+    def close(self) -> None:
+        """
+        Close the gates activity and clean up resources.
+        """
+
+        if self.minio_repository is not None:
+            try:
+                self.minio_repository.close()
+            finally:
+                self.minio_repository = None
+        SientiaMonitoring.shutdown(self)
+
+    def __del__(self):
+        self.close()
+
+    def _debug_dataframe(self, message: str, data: Any, metadata: dict[str, Any]) -> None:
+        """
+        Log dataframe content only when row count is below the configured threshold
+
+        Args:
+         - message (str): Base log message to identify the dataframe in logs
+         - data (Any): Dataframe-like payload to be logged
+         - metadata (dict[str, Any]): Workflow metadata for contextual logging
+        """
+        self.debug(
+            build_dataframe_debug_message(
+                message=message,
+                data=data,
+                max_rows=self._MAX_DEBUG_DATAFRAME_ROWS,
+            ),
+            metadata,
+        )
+
+    @staticmethod
+    def _read_filter_entry(config: dict[str, Any]) -> tuple[str, dict[str, Any]]:
+        """
+        Read filter policy/config keys in a case-insensitive way.
+
+        Args:
+            config (dict[str, Any]): Filter configuration dictionary.
+
+        Return:
+            tuple[str, dict[str, Any]]: Parsed policy and config payload.
+        """
+        normalized = {str(key).upper(): value for key, value in config.items()}
+        policy = normalized['POLICY']
+        filter_config = normalized.get('CONFIG', {})
+        return policy, filter_config
+
+    @activity.defn(name='input_gate')
+    def input_gate(self, input_data: dict[str, Any]) -> tuple[str | None, int, str]:
+        """
+        Apply input data quality filters and validation.
+
+        This activity validates input data quality using configurable filters
+        before proceeding with ML operations. It applies multiple filter types
+        and returns a path decision based on the filter results and configured
+        policies.
+
+        The method implements a comprehensive filtering system that:
+        1. Applies configured filters to input data
+        2. Evaluates filter results against policy configurations
+        3. Determines appropriate path decisions (STOP, CONTINUE, REPEAT)
+        4. Provides confidence scores and detailed comments
+        5. Handles errors gracefully with notification integration
+
+        Args:
+            input_data: Configuration and data for input validation
+                Required keys:
+                    - metadata (dict): Workflow execution metadata
+                    - filters (dict): Filter configuration and policies
+                    - data (dict): Input data to validate
+                    - path_priority (list[str]): Priority order for path decisions
+
+        Returns:
+            tuple: (path_flag, confidence, comment)
+                - path_flag (str | None): Decision path (STOP, CONTINUE, REPEAT, or None)
+                - confidence (int): Confidence score for the decision
+                - comment (str): Detailed explanation of the decision
+
+        Raises:
+            Exception: If filter execution fails or configuration is invalid
+        """
+        metadata = input_data['metadata']
+
+        self.info('Performing input gate...', metadata)
+
+        filters = input_data['filters']
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+        path_priority = input_data['path_priority']
+
+        filter_output = []
+
+        self._debug_dataframe('Input data:', data, metadata)
+        self.debug(f'Filters: {filters}', metadata)
+
+        # Apply each configured filter
+        for fil, config in filters.items():
+            if fil not in input_filter_functions:
+                self.error(f'Filter {fil} not found', metadata)
+                continue
+            policy, filter_config = self._read_filter_entry(config)
+            try:
+                if input_filter_functions[fil](data, filter_config):
+                    self.debug(f'Data not passed the input filter {fil}:{config}', metadata)
+                    filter_output.append(policy)
+            except Exception as e:
+                trace = traceback.format_exc()
+                self.send_notification(
+                    metadata=metadata,
+                    notification_id=f'INTPUT_GATE_ERROR__{fil}',
+                    message=f'Error in filter {fil}:{config}: \n {e}',
+                    block='input_gate',
+                    level=NotificationLevel.ERROR,
+                    attachment_content=trace,
+                )
+
+        for path_flag in path_priority:
+            if path_flag in filter_output:
+                self.info(f'Input gate result: {path_flag}', metadata)
+                return path_flag, input_path_confidence[path_flag], 'Input data with bad quality'
+
+        self.info('Nothing was filtered by the input gate', metadata)
+
+        del data
+
+        return None, 0, ''
+
+    @activity.defn(name='mlflow_response_gate')
+    def mlflow_response_gate(self, input_data: dict[str, Any]) -> tuple[str | None, int, str]:
+        """
+        Validate MLFlow API response quality and integrity.
+
+        This activity validates MLFlow API responses to ensure they meet quality
+        standards before proceeding with further processing. It applies response-specific
+        filters and determines appropriate path decisions based on response quality.
+
+        The method implements response validation that:
+        1. Applies MLFlow response-specific filters
+        2. Evaluates API response quality and integrity
+        3. Determines path decisions based on response validation results
+        4. Provides confidence scores and detailed validation comments
+        5. Handles API errors and response validation failures
+
+        Args:
+            input_data: Configuration and data for response validation
+                Required keys:
+                    - metadata (dict): Workflow execution metadata
+                    - filters (dict): Response filter configuration and policies
+                    - data (dict): MLFlow API response data to validate
+                    - type (str): Type of MLFlow operation (transform, predict)
+                    - path_priority (list[str]): Priority order for path decisions
+
+        Returns:
+            tuple: (path_flag, confidence, comment)
+                - path_flag (str | None): Decision path (STOP, CONTINUE, REPEAT, or None)
+                - confidence (int): Confidence score for the decision
+                - comment (str): Detailed explanation of the decision
+
+        Raises:
+            Exception: If response validation fails or configuration is invalid
+        """
+        metadata = input_data['metadata']
+        self.info('Performing mlflow response gate...', metadata)
+        raw_data = input_data['data']
+        filters = input_data['filters']
+
+        self.debug(
+            f'Input data: \n {create_sample_dict(raw_data, max_items=5, max_depth=5)}', metadata
+        )
+        self.debug(f'Filters: {filters}', metadata)
+
+        payload = MinioDataFramePayload.from_dict(raw_data)
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        gate_type = input_data['type']
+        path_priority = input_data['path_priority']
+
+        filter_output = []
+
+        comments = []
+
+        status = payload.status or {}
+
+        for fil, config in filters.items():
+            if fil not in mlflow_response_filter_functions:
+                continue
+            policy, filter_config = self._read_filter_entry(config)
+            try:
+                if mlflow_response_filter_functions[fil](status, filter_config):
+                    filter_output.append(policy)
+                    comments.append(status.get('message', 'Unknown MLFlow API error'))
+                    self.send_notification(
+                        metadata=metadata,
+                        notification_id=f'{gate_type.upper()}_GATE_RESPONSE_FILTER__{fil}',
+                        message=status.get('message', 'Unknown MLFlow API error'),
+                        block='mlflow_gate',
+                        level=NotificationLevel.ERROR,
+                        attachment_content=status.get('traceback'),
+                    )
+            except Exception as e:
+                trace = traceback.format_exc()
+                self.send_notification(
+                    metadata=metadata,
+                    notification_id=f'MLFLOW_GATE_RESPONSE_FILTER__{fil}',
+                    message=f'Error in filter {fil}:{config}: \n {e}',
+                    block='mlflow_gate',
+                    level=NotificationLevel.ERROR,
+                    attachment_content=trace,
+                )
+
+        for path_flag in path_priority:
+            if path_flag in filter_output:
+                self.info(f'Mlflow response gate result: {path_flag}', metadata)
+                return path_flag, mlflow_response_path_confidence[path_flag], ', '.join(comments)
+
+        self.info('Nothing was filtered by the mlflow response gate', metadata)
+
+        del data
+
+        return None, 0, ''
+
+    @activity.defn(name='mlflow_content_gate')
+    def mlflow_content_gate(self, input_data: dict[str, Any]) -> tuple[str | None, int, str]:
+        """
+        Validate MLFlow prediction content quality and integrity.
+
+        This activity validates the content of MLFlow predictions to ensure they
+        meet quality standards before export and persistence. It applies content-specific
+        filters and determines appropriate path decisions based on content quality.
+
+        The method implements content validation that:
+        1. Applies MLFlow content-specific filters
+        2. Evaluates prediction content quality and integrity
+        3. Determines path decisions based on content validation results
+        4. Provides confidence scores and detailed validation comments
+        5. Handles content validation failures and quality issues
+
+        Args:
+            input_data: Configuration and data for content validation
+                Required keys:
+                    - metadata (dict): Workflow execution metadata
+                    - filters (dict): Content filter configuration and policies
+                    - data (dict): MLFlow prediction content to validate
+                    - type (str): Type of MLFlow operation (transform, predict)
+                    - path_priority (list[str]): Priority order for path decisions
+
+        Returns:
+            tuple: (path_flag, confidence, comment)
+                - path_flag (str | None): Decision path (STOP, CONTINUE, REPEAT, or None)
+                - confidence (int): Confidence score for the decision
+                - comment (str): Detailed explanation of the decision
+
+        Raises:
+            Exception: If content validation fails or configuration is invalid
+        """
+        metadata = input_data['metadata']
+        self.info('Performing mlflow content gate...', metadata)
+
+        filters = input_data['filters']
+
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        gate_type = input_data['type']
+        path_priority = input_data['path_priority']
+
+        filter_output = []
+
+        self._debug_dataframe('Input data:', data, metadata)
+        self.debug(f'Filters: \n {filters}', metadata)
+
+        for fil, config in filters.items():
+            if fil not in mlflow_content_filter_functions:
+                continue
+            policy, filter_config = self._read_filter_entry(config)
+            try:
+                if mlflow_content_filter_functions[fil](data, filter_config):
+                    filter_output.append(policy)
+                    self.send_notification(
+                        metadata=metadata,
+                        notification_id=f'{gate_type.upper()}_GATE_CONTENT_FILTER__{fil}',
+                        message=f'Data not passed the content filter {fil}:{config}',
+                        block='mlflow_gate',
+                        level=NotificationLevel.WARNING,
+                        attachment_content=data.to_string(),
+                    )
+            except Exception as e:
+                trace = traceback.format_exc()
+                self.send_notification(
+                    metadata=metadata,
+                    notification_id=f'MLFLOW_GATE_CONTENT_FILTER__{fil}',
+                    message=f'Error in filter {fil}:{config}: \n {e}',
+                    block='mlflow_gate',
+                    level=NotificationLevel.ERROR,
+                    attachment_content=trace,
+                )
+
+        for path_flag in path_priority:
+            if path_flag in filter_output:
+                self.info(f'Mlflow content gate result: {path_flag}', metadata)
+                return (
+                    path_flag,
+                    mlflow_content_path_confidence[path_flag],
+                    'Transformed data not passed the content filter',
+                )
+
+        self.info('Nothing was filtered by the mlflow content gate', metadata)
+
+        del data
+
+        return None, 0, ''
+
+    def get_prediction_store_policy(
+        self, prediction_store_policy: str, metadata: dict[str, Any]
+    ) -> tuple[str, int]:
+        """
+        Parse and validate prediction store policy configuration.
+
+        This method parses prediction store policy strings in the format 'type:value'
+        and validates them against allowed policy types and values. It provides
+        sensible defaults for invalid configurations and logs policy validation
+        failures for operational monitoring.
+
+        Supported Policy Types:
+        - 'lts': Latest timestamp - sorts data by timestamp descending
+        - 'erl': Earliest timestamp - sorts data by timestamp ascending
+
+        Args:
+            prediction_store_policy (str): Policy string in format 'type:value'
+            metadata (dict[str, Any]): Context metadata for logging and notifications
+
+        Returns:
+            tuple[str, int]: (policy_type, policy_value)
+                - policy_type (str): Validated policy type ('lts' or 'erl')
+                - policy_value (int): Number of rows to retain
+        """
+        policy_elements = prediction_store_policy.split(':')
+
+        if len(policy_elements) < 2:
+            self.error(
+                f'Invalid prediction store policy: {prediction_store_policy}, using default policy',
+                metadata,
+            )
+            return 'lts', 1
+
+        policy_type = policy_elements[0]
+        policy_value = policy_elements[1]
+
+        # If the policy_type is not lts or erl, we use the default policy
+        # If the policty_value is not a number or 0, we use the default policy
+        if (
+            policy_type not in ['lts', 'erl']
+            or not policy_value.isdigit()
+            or int(policy_value) == 0
+        ):
+            self.error(
+                f'Invalid prediction store policy: {prediction_store_policy}, using default policy',
+                metadata,
+            )
+            return 'lts', 1
+
+        return policy_type, int(policy_value)
+
+    @activity.defn(name='format_transformed_data')
+    def format_transformed_data(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
+        """
+        Format transformed data for storage and export operations.
+
+        This method formats transformed data from MLFlow model transformations
+        into a standardized format suitable for database storage. It converts
+        wide-format data (columns as variables) into long-format (melted)
+        with proper timestamp handling and model identification.
+
+        The formatting process includes:
+        1. Converting input data dictionary to DataFrame
+        2. Extracting timestamps from DataFrame index
+        3. Resetting index to create sequential row numbers
+        4. Melting data from wide format to long format (variable-value pairs)
+        5. Adding model_id for data lineage tracking
+
+        Args:
+            input_data (dict): Input data containing:
+                - metadata (dict): Workflow execution metadata
+                - data (dict[str, Any]): Transformed data to format (DataFrame-compatible dict)
+                - model_id (str): Unique identifier for the ML model
+
+        Returns:
+            dict: Formatted data dictionary with keys:
+                - timestamp (dict): Timestamp values indexed by row number
+                - variable (dict): Variable names indexed by row number
+                - value (dict): Variable values indexed by row number
+                - model_id (dict): Model identifiers indexed by row number
+        """
+        metadata = input_data['metadata']
+
+        model_id = input_data['model_id']
+
+        self.info('Formatting transformed data...', metadata)
+
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        data['timestamp'] = data.index
+        data = data.reset_index(drop=True)
+
+        data = data.melt(id_vars='timestamp', var_name='variable', value_name='value')
+        data['model_id'] = model_id
+
+        return MinioDataFramePayload.from_dataframe(
+            dataframe=data,
+            minio_repo=self.minio_repository,
+            model_name=input_data['model_name'],
+            operation='transform',
+            workflow_metadata=metadata,
+            last_timestamp=payload.last_timestamp,
+            logger=self.logger,
+        )
+
+    @activity.defn(name='format_prediction')
+    def format_prediction(self, input_data: dict[str, Any]) -> dict:
+        """
+        Format prediction data according to configured storage policies.
+
+        This method formats prediction data for storage and export operations.
+        It applies timestamp-based sorting policies, adds metadata fields,
+        and ensures data consistency before persistence. The method supports
+        multiple storage policies for flexible data retention strategies.
+
+        If only one row is present, we use the last timestamp as the timestamp
+
+        Storage Policies:
+        - 'lts:N': Latest timestamp - retains N most recent predictions
+        - 'erl:N': Earliest timestamp - retains N oldest predictions
+
+        Args:
+            input_data (dict): Input data containing:
+                - data (dict[str, Any]): Raw prediction data to format
+                - timestamp (str): Timestamp of the data
+                - model_id (str): Unique identifier for the ML model
+                - prediction_confidence (float): Confidence score for the prediction
+                - prediction_store_policy (str): Storage policy in format 'type:value'
+
+        Returns:
+            dict: Formatted prediction data ready for storage and export
+        """
+        metadata = input_data['metadata']
+        last_timestamp = input_data['timestamp']
+        prediction_store_policy = input_data['prediction_store_policy']
+        self.info('Formatting prediction...', metadata)
+
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        # Create timestamp column from index and reset index
+        data['timestamp'] = data.index
+        data = data.reset_index(drop=True)
+
+        self.debug(f'Prediction store policy: {prediction_store_policy}', metadata)
+        self._debug_dataframe('Prediction data:', data, metadata)
+
+        policy_type, policy_value = self.get_prediction_store_policy(
+            prediction_store_policy, metadata
+        )
+
+        # If data has no timestamp, we use the default timestamp and not sort the data
+        self.info(
+            f'Sorting data by timestamp and applying policy: {policy_type}:{policy_value}', metadata
+        )
+
+        # If policy_type is lts, we need to sort the data by timestamp descending and take the first policy_value rows
+        if policy_type == 'lts':
+            self.debug('Sorting data by timestamp descending', metadata)
+            data = data.sort_values(by='timestamp', ascending=False)
+        # If policy_type is erl, we need to sort the data by timestamp ascending and take the first policy_value rows
+        elif policy_type == 'erl':
+            self.debug('Sorting data by timestamp ascending', metadata)
+            data = data.sort_values(by='timestamp', ascending=True)
+        else:
+            self.error(f'Invalid policy type: {policy_type}, using default policy', metadata)
+            raise ValueError(f'Invalid policy type: {policy_type}')
+
+        int_policy_value = int(policy_value)
+
+        data = data.head(int_policy_value)
+        if int_policy_value == 1:
+            data['timestamp'] = last_timestamp
+
+        data['model_id'] = input_data['model_id']
+        data['prediction_confidence'] = input_data['prediction_confidence']
+        data['prediction_status'] = 'Good'
+        data['comments'] = ''
+        data = data.sort_values(by='timestamp', ascending=False)
+        data = data.reset_index(drop=True)
+
+        self.info(f'Prediction formatted: {len(data)} rows', metadata)
+        self._debug_dataframe('Prediction data:', data, metadata)
+
+        return data.to_dict()
+
+    @activity.defn(name='format_default_prediction')
+    def format_default_prediction(self, input_data: dict[str, Any]) -> dict:
+        """
+        Create and format default prediction data for error conditions.
+
+        This method generates default prediction data when the main prediction
+        pipeline encounters errors or quality issues. It creates a standardized
+        data structure with zero values for predictions and useful metadata
+        for operational monitoring and debugging.
+
+        The default prediction serves as a fallback mechanism to:
+        1. Maintain data pipeline continuity during failures
+        2. Provide operational visibility into prediction quality issues
+        3. Enable downstream systems to handle error conditions gracefully
+        4. Support debugging and troubleshooting efforts
+
+        Args:
+            input_data (dict): Input data containing:
+                - timestamp (str): Timestamp for the default prediction
+                - model_id (str): Unique identifier for the ML model
+                - prediction_confidence (float): Confidence score (typically low for errors)
+                - comment (str): Error description or operational comment
+
+        Returns:
+            dict: Formatted default prediction data with error indicators
+        """
+
+        metadata = input_data['metadata']
+        self.debug('Formatting default prediction...', metadata)
+
+        data = DataFrame(
+            {
+                'prediction': [0],
+                'response_time': [0],
+                'timestamp': [input_data['timestamp']],
+                'model_id': [input_data['model_id']],
+                'prediction_confidence': [input_data['prediction_confidence']],
+                'prediction_status': ['Bad'],
+                'comments': [input_data['comment']],
+            }
+        )
+
+        self.info(f'Default prediction formatted: {data.size} rows', metadata)
+        return data.to_dict()
+
+    @activity.defn(name='format_retrain_report')
+    def format_retrain_report(self, input_data: dict[str, Any]) -> dict:
+        """
+        Format retrain report data for storage and audit trail maintenance.
+
+        This method formats model retraining operation results into a standardized
+        report format suitable for database storage and operational monitoring.
+        It captures retraining status, timestamps, and model version information
+        for comprehensive audit trails and operational visibility.
+
+        The formatting process includes:
+        1. Extracting retraining experiment response data
+        2. Capturing model update report information (version, MLflow IDs)
+        3. Formatting timestamps and status information
+        4. Conditionally including version information for successful retrains
+
+        Args:
+            input_data (dict): Input data containing:
+                - metadata (dict): Workflow execution metadata
+                - experiment_response (dict): Retraining experiment response containing:
+                    - success (bool): Retraining operation success status
+                    - timestamp (str): Timestamp of the retraining operation
+                    - message (str): Status message or error description
+                - update_report (dict): Model update report containing:
+                    - version (str): New model version identifier
+                    - mlflow_run_id (str): MLflow run identifier
+                    - mlflow_experiment_id (str): MLflow experiment identifier
+                - model_id (str): Unique identifier for the ML model
+                - model_name (str): Name of the ML model
+
+        Returns:
+            dict: Formatted retrain report dictionary with keys:
+                - model_id (dict): Model identifiers indexed by row number
+                - model_name (dict): Model names indexed by row number
+                - timestamp (dict): Retraining timestamps indexed by row number
+                - status (dict): Retraining status messages indexed by row number
+                - version (dict, optional): Model versions indexed by row number
+                    Only included if experiment_response['success'] is True
+                - mlflow_run_id (dict, optional): MLflow run IDs indexed by row number
+                    Only included if experiment_response['success'] is True
+                - mlflow_experiment_id (dict, optional): MLflow experiment IDs indexed by row number
+                    Only included if experiment_response['success'] is True
+        """
+        metadata = input_data['metadata']
+        self.info('Formatting retrain report...', metadata)
+
+        experiment_response = input_data['experiment_response']
+        update_report = input_data['update_report']
+        model_id = input_data['model_id']
+        model_name = input_data['model_name']
+
+        report = DataFrame(
+            {
+                'model_id': [model_id],
+                'model_name': [model_name],
+                'timestamp': [experiment_response['timestamp']],
+                'status': [experiment_response['message']],
+            }
+        )
+
+        if experiment_response['success']:
+            # Retrain was successfull
+            report['version'] = update_report['version']
+            report['mlflow_run_id'] = update_report['mlflow_run_id']
+            report['mlflow_experiment_id'] = update_report['mlflow_experiment_id']
+
+        self._debug_dataframe('Retrain report:', report, metadata)
+
+        return report.to_dict()
+
+    @activity.defn(name='write_metrics')
+    def write_metrics(self, input_data: dict[str, Any]):
+        """
+        Write prediction performance metrics to Prometheus monitoring system.
+
+        This method records comprehensive metrics for prediction operations,
+        enabling operational monitoring, performance analysis, and alerting.
+        It tracks prediction counts, confidence levels, and response times
+        for each model and pipeline combination.
+
+        Metrics Recorded:
+        1. Prediction Count: Incremental counter for successful predictions
+        2. Confidence Monitor: Current confidence level for predictions
+        3. Response Time Monitor: Histogram of prediction response times
+
+        Args:
+            input_data (dict): Input data containing:
+                - metadata (dict[str, Any]): Workflow execution metadata
+                - prediction (dict[str, Any]): Prediction data with metrics
+
+        Raises:
+            Exception: If metrics writing fails or configuration is invalid
+        """
+        metadata = input_data['metadata']
+        prediction = DataFrame(input_data['prediction'])
+        prediction_confidence = prediction['prediction_confidence'].values[0]
+        response_time = prediction['response_time'].values[0]
+        opc_metrics = input_data['opc_metrics']
+
+        self.info(f'Writing metrics for model {metadata["model_name"]}', metadata)
+
+        core_tags = {
+            'pod_id': self.pod_id,
+            'runtime': self.runtime,
+            'operation_type': 'predict',
+            'model_name': metadata['model_name'],
+            'workflow_name': metadata['workflow_name'],
+        }
+        self.emit_metric_sync(
+            metric_object=metrics.PREDICTIONS_WRITTEN_COUNT,
+            tags=core_tags,
+        )
+
+        self.emit_metric_sync(
+            metric_object=metrics.PREDICTION_CONFIDENCE_MONITOR,
+            method='set',
+            tags=core_tags,
+            value=prediction_confidence,
+        )
+
+        self.emit_metric_sync(
+            metric_object=metrics.PREDICTION_RESPONSE_TIME_MONITOR,
+            method='observe',
+            tags=core_tags,
+            value=response_time,
+        )
+
+        for server_id, tags in opc_metrics.items():
+            for tag, response_time in tags.items():
+                if response_time is not None:
+                    self.emit_metric_sync(
+                        metric_object=metrics.PREDICTION_OPC_WRITING_RESPONSE_TIME_MONITOR,
+                        method='observe',
+                        tags={
+                            **core_tags,
+                            'opc_server_id': server_id,
+                            'tag': tag,
+                        },
+                        value=response_time,
+                    )
+
+                    self.emit_metric_sync(
+                        metric_object=metrics.PREDICTION_OPC_WRITING_COUNT,
+                        tags={
+                            **core_tags,
+                            'opc_server_id': server_id,
+                            'tag': tag,
+                        },
+                    )
+
+        self.info(f'Metrics written for model {metadata["model_name"]}', metadata)

laborious/activities/mlflow.py

@@ -0,0 +1,755 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    import tempfile
+    import traceback
+    from datetime import datetime
+    from pathlib import Path
+    from shutil import rmtree
+    from typing import Any
+
+    import mlflow
+    import numpy as np
+    import pandas as pd
+    from pandas import DataFrame, to_datetime
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+    from sientia_do.repository.minio_repository_sync import MinioRepository
+    from sientia_do.temporal.constants import (
+        DATETIME_FORMAT,
+        DATETIME_FORMAT_MS_WITH_TZ,
+        DATETIME_FORMAT_WITH_TZ,
+        now,
+    )
+    from sientia_do.utils.formatters import create_sample_dict
+    from sientia_model.model_repository.mlflow_repository import SientiaMLflowRepository
+    from sientia_model.model_repository.plugin_store import PluginStore
+
+    from laborious.utils.dataframe_debug import build_dataframe_debug_message
+    from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
+
+
+class MLFlow(SientiaMonitoring):
+    """
+    Temporal activities that talk to MLflow through ``SientiaMLflowRepository`` and ``SientiaModel`` wrappers.
+
+    Models are resolved by registered name and the ``production`` alias (not by legacy stages or
+    separate transform/predict flavors). ``get_cached_model`` loads or reuses a wrapper; inference
+    uses ``wrapper.transform`` / ``wrapper.predict``; retrain uses ``wrapper.retrain`` or
+    ``wrapper.train`` plus ``store_model`` and registry promotion via ``promote_to_alias``.
+
+    Large inputs and outputs flow through ``MinioDataFramePayload`` when workflows offload parquet
+    to MinIO. On failure, transform/predict still return a payload with ``success: False`` and
+    error details for downstream gates.
+
+    Attributes:
+        mlflow_repository: Client for tracking, registry, artifact download, and run lifecycle.
+        plugin_store: Reference to the store (runtime is installed on the worker; reserved for
+            future store-backed helpers).
+    """
+
+    _MAX_DEBUG_DATAFRAME_ROWS = 100
+    _DEFAULT_MODEL_ALIAS = 'production'
+    _REFERENCE_ARTIFACT_CANDIDATES = ('evaluation_data.csv', 'test_data.csv')
+
+    def __init__(
+        self,
+        mlflow_repository: SientiaMLflowRepository,
+        plugin_store: PluginStore,
+        minio_repository: MinioRepository | None = None,
+        logger: Logger | None = None,
+        notification_handler: NotificationHandler | None = None,
+        metrics_controller: MetricsController | None = None,
+    ):
+        """
+        Attach shared MLflow and MinIO clients used by all ML activities in this mixin.
+
+        Args:
+         - mlflow_repository: Repository built by ``Activities`` (or injected in tests).
+         - plugin_store: Plugin store instance from worker bootstrap.
+         - minio_repository: MinIO client for ``MinioDataFramePayload`` upload/download.
+         - logger: Structured logger.
+         - notification_handler: Notifications on hard failures where applicable.
+         - metrics_controller: Shared metrics controller.
+
+        Return:
+            None
+        """
+
+        self.minio_repository = minio_repository
+        SientiaMonitoring.__init__(
+            self,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+        )
+        self.mlflow_repository = mlflow_repository
+        self.plugin_store = plugin_store
+
+    def close(self) -> None:
+        """
+        Release MinIO manager resources held by the mixin.
+
+        Return:
+            None
+        """
+        if self.minio_repository is not None:
+            try:
+                self.minio_repository.close()
+            finally:
+                self.minio_repository = None
+        SientiaMonitoring.shutdown(self)
+
+    def __del__(self):
+        self.close()
+
+    def _debug_dataframe(self, message: str, data: Any, metadata: dict[str, Any]) -> None:
+        """
+        Log dataframe content only when row count is below the configured threshold
+
+        Args:
+         - message (str): Base log message to identify the dataframe in logs
+         - data (Any): Dataframe-like object expected to expose shape and to_csv
+         - metadata (dict[str, Any]): Workflow metadata for contextual logging
+        """
+        self.debug(
+            build_dataframe_debug_message(
+                message=message,
+                data=data,
+                max_rows=self._MAX_DEBUG_DATAFRAME_ROWS,
+            ),
+            metadata,
+        )
+
+    def _detect_and_parse_datetime_index(self, data: pd.DataFrame, metadata: dict) -> pd.DataFrame:
+        """
+        Ensure the transform output index is homogeneous and encoded as ``DATETIME_FORMAT_WITH_TZ`` strings.
+
+        Accepts an all-string index (validated against the format), or all-``datetime`` /
+        ``Timestamp`` (naive timestamps are localized to UTC before formatting). Mixed element types
+        or unsupported types raise ``ValueError`` with a message logged at info level.
+
+        Args:
+         - data: DataFrame whose index carries the time dimension after transform.
+         - metadata: Workflow metadata for log correlation.
+
+        Return:
+            ``pd.DataFrame``: Same frame with a normalized string index; empty frames are returned as-is.
+        """
+
+        if data.empty:
+            self.info('Data is empty, skipping datetime index detection and parsing', metadata)
+            return data
+
+        index = data.index
+        index_type = type(index[0])
+
+        self.info(f'Index type: {index_type}', metadata)
+
+        message = (
+            f'Index must be all timestamp like column. Valid formats are: pandas Timestamp, datetime, '
+            f'string in format {DATETIME_FORMAT_WITH_TZ}.'
+        )
+
+        if not all(isinstance(i, index_type) for i in index):
+            types = map(str, map(type, index))
+            raise ValueError(f'{message}. Elements are {",".join(types)}')
+
+        if index_type is str:
+            try:
+                pd.to_datetime(data.index, format=DATETIME_FORMAT_WITH_TZ)
+            except ValueError as e:
+                raise ValueError(f'{message}. Unable to parse given date format: {e}') from e
+
+        elif index_type is datetime or index_type is pd.Timestamp:
+            idx = data.index
+            if hasattr(idx, 'tz') and idx.tz is None:
+                data.index = idx.tz_localize('UTC')  # type: ignore[attr-defined]
+
+            data.index = data.index.strftime(DATETIME_FORMAT_WITH_TZ)  # type: ignore[attr-defined]
+        else:
+            raise ValueError(f'{message}. Got {index_type}.')
+
+        return data
+
+    def _resolve_model_version_for_run(self, run_id: str) -> str:
+        """
+        Map an MLflow ``run_id`` to the latest registered model version that produced that run.
+
+        ``search_model_versions`` may return multiple versions if the model was registered more than
+        once for the same run; the highest numeric ``version`` wins so promotion targets the newest
+        artifact set.
+
+        Args:
+         - run_id: Run UUID from ``retrain_model`` / experiment payload.
+
+        Return:
+            str: Registry version string acceptable by ``promote_to_alias``.
+
+        Raises:
+            ValueError: If the filter returns no versions (model not registered for this run).
+        """
+
+        versions = self.mlflow_repository._client.search_model_versions(
+            filter_string=f"run_id='{run_id}'"
+        )
+        if not versions:
+            raise ValueError(f'No registered model version found for run_id={run_id}')
+        latest = max(versions, key=lambda v: int(v.version))
+        return str(latest.version)
+
+    def _resolve_model_alias(self, model_config: dict[str, Any] | None = None) -> str:
+        """
+        Resolve which MLflow alias should be used for model lookup/promotion.
+
+        Args:
+         - model_config: Optional model configuration that may include ``alias``.
+
+        Return:
+            str: Alias name trimmed and normalized; defaults to ``production``.
+        """
+        if not model_config:
+            return self._DEFAULT_MODEL_ALIAS
+        alias = str(model_config.get('alias', self._DEFAULT_MODEL_ALIAS)).strip()
+        return alias or self._DEFAULT_MODEL_ALIAS
+
+    @activity.defn(name='request_transform')
+    def request_transform(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
+        """
+        Pivot long-format sensor rows, load the production wrapper, and run ``wrapper.transform``.
+
+        Expected tabular shape after load: columns including ``variable``, ``timestamp``, ``value``,
+        and ``created_at`` for deduplication. Data are sorted by ``created_at``, de-duplicated per
+        ``(variable, timestamp)``, pivoted wide, then passed to the model. ``model_config`` may
+        include ``retention_minutes`` for wrapper cache TTL.
+
+        Args:
+         - input_data: Dict with ``metadata``, ``model_name``, ``data`` (``MinioDataFramePayload``
+           dict or inline dataframe dict), and optional ``model_config``.
+
+        Return:
+            ``MinioDataFramePayload`` with transformed frame and ``success: True``, or a payload
+            with ``success: False`` and exception details in ``status`` if transform fails.
+        """
+        metadata = input_data['metadata']
+        self.info('Transforming data...', metadata)
+
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+        model_alias = self._resolve_model_alias(model_config)
+
+        self._debug_dataframe('Raw input data:', data, metadata)
+
+        # Long → wide: keep newest row per (variable, timestamp), then pivot for the wrapper API.
+        data = data.sort_values('created_at', ascending=False).drop_duplicates(
+            subset=['variable', 'timestamp'], keep='first'
+        )
+
+        data = data.pivot(index='timestamp', columns='variable', values='value')
+        data.fillna(np.nan, inplace=True)
+
+        data.columns.name = None
+        data.index.name = None
+
+        data['timestamp'] = data.index
+
+        self._debug_dataframe('Processed input data:', data, metadata)
+
+        try:
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias=model_alias,
+                retention_minutes=model_config.get('retention_minutes', 0),
+                metadata=metadata,
+            )
+            transformed_df, transform_meta = wrapper.transform(data)
+            if transform_meta:
+                self.info(f'Wrapper transform metadata: {transform_meta}', metadata)
+
+            transformed_df = self._detect_and_parse_datetime_index(transformed_df, metadata)
+
+            response_data: dict[str, Any] = {'success': True, 'content': transformed_df}
+        except Exception as e:
+            response_data = {
+                'success': False,
+                'content': {'message': str(e), 'traceback': traceback.format_exc()},
+            }
+
+        self.debug(
+            f'Transform raw response data: \n {create_sample_dict(response_data, max_items=5, max_depth=5)}',
+            metadata,
+        )
+
+        self.debug(
+            f'Transform response data: \n {create_sample_dict(response_data, max_items=5, max_depth=5)}',
+            metadata,
+        )
+
+        self.info('Data transformed successfully', metadata)
+
+        if not response_data.get('success', False):
+            return MinioDataFramePayload.from_dataframe(
+                dataframe=None,
+                minio_repo=self.minio_repository,
+                model_name=model_name,
+                operation='transform',
+                status=response_data,
+                workflow_metadata=metadata,
+                last_timestamp=payload.last_timestamp,
+                logger=self.logger,
+            )
+
+        return MinioDataFramePayload.from_dataframe(
+            dataframe=response_data['content'],
+            minio_repo=self.minio_repository,
+            model_name=model_name,
+            operation='transform',
+            workflow_metadata=metadata,
+            status={
+                'success': True,
+            },
+            last_timestamp=payload.last_timestamp,
+            logger=self.logger,
+        )
+
+    @activity.defn(name='request_predict')
+    def request_predict(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
+        """
+        Load the production wrapper and call ``wrapper.predict`` on the prepared feature frame.
+
+        The activity normalizes ``NaN`` to ``None`` for JSON-friendly columns, sets the row index
+        the same way as ``retrain_model`` (UTC ``DatetimeIndex`` from ``DATETIME_FORMAT_WITH_TZ``),
+        restores that index on the prediction frame, normalizes the prediction index to
+        ``DATETIME_FORMAT_WITH_TZ`` strings like ``request_transform``, and records ``response_time``.
+        Non-DataFrame predictions are coerced to a single ``prediction`` column.
+
+        Args:
+         - input_data: Same envelope as ``request_transform`` (``metadata``, ``model_name``,
+           ``data``, optional ``model_config`` with ``retention_minutes``).
+
+        Return:
+            ``MinioDataFramePayload`` with predictions or error status mirroring transform behaviour.
+        """
+        metadata = input_data['metadata']
+        self.info('Predicting data...', metadata)
+
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+        model_alias = self._resolve_model_alias(model_config)
+
+        self._debug_dataframe('Input data for prediction:', data, metadata)
+
+        data.replace(np.nan, None, inplace=True)
+
+        data.index = pd.DatetimeIndex(
+            to_datetime(data.index, format=DATETIME_FORMAT_WITH_TZ, utc=True)
+        )
+        input_index = data.index
+
+        try:
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias=model_alias,
+                retention_minutes=model_config.get('retention_minutes', 0),
+                metadata=metadata,
+            )
+            start_time = datetime.now()
+            predict_data, pred_meta = wrapper.predict({}, data)
+            end_time = datetime.now()
+
+            if pred_meta:
+                self.info(f'Wrapper predict metadata: {pred_meta}', metadata)
+
+            if isinstance(predict_data, DataFrame):
+                self._debug_dataframe(
+                    'Data received from model prediction:', predict_data, metadata
+                )
+                predict_data.columns = pd.Index(['prediction'])
+            else:
+                self.debug(
+                    f'Data received from model prediction (not a DataFrame): {predict_data}',
+                    metadata,
+                )
+                predict_data = pd.DataFrame(predict_data, columns=['prediction'])
+
+            predict_data.index = input_index
+            predict_data['response_time'] = (end_time - start_time).total_seconds()
+            predict_data = self._detect_and_parse_datetime_index(predict_data, metadata)
+
+            response_data: dict[str, Any] = {'success': True, 'content': predict_data}
+        except Exception as e:
+            response_data = {
+                'success': False,
+                'content': {'message': str(e), 'traceback': traceback.format_exc()},
+            }
+
+        self.debug(
+            f'Prediction response data: \n {create_sample_dict(response_data, max_items=5, max_depth=5)}',
+            metadata,
+        )
+
+        self.info('Data predicted successfully', metadata)
+
+        if not response_data.get('success', False):
+            return MinioDataFramePayload.from_dataframe(
+                dataframe=None,
+                minio_repo=self.minio_repository,
+                model_name=model_name,
+                operation='predict',
+                status=response_data,
+                workflow_metadata=metadata,
+                last_timestamp=payload.last_timestamp,
+                logger=self.logger,
+            )
+
+        return MinioDataFramePayload.from_dataframe(
+            dataframe=response_data['content'],
+            minio_repo=self.minio_repository,
+            model_name=model_name,
+            operation='predict',
+            workflow_metadata=metadata,
+            status={
+                'success': True,
+            },
+            last_timestamp=payload.last_timestamp,
+            logger=self.logger,
+        )
+
+    @activity.defn(name='retrain_model')
+    def retrain_model(self, input_data: dict[str, Any]) -> dict[str, Any]:
+        """
+        Fit an updated wrapper from historical data, then log and register in MLflow.
+
+        Flow: load long-format data from MinIO → dedupe/pivot like inference prep → require
+        ``model_config['target']`` → read current ``production`` version for ``source_run_id`` tag →
+        run ``wrapper.retrain`` outside run timing → ``start_run`` with retrain tags → log input
+        CSV artifact → ``store_model`` and ``log_params``. Does not promote; the workflow calls
+        ``update_production_model`` after validation.
+
+        Args:
+         - input_data: Must include ``metadata``, ``model_name``, ``data`` (payload), and
+           ``model_config`` with at least ``target``.
+
+        Return:
+            On success: ``success``, ``experiment`` (``run_id``, ``experiment_id``, ``experiment_name``),
+            ``message``, ``timestamp``. On failure: ``success: False``, error fields, and optional trace.
+        """
+
+        if self.minio_repository is None:
+            raise ValueError('Minio repository not initialized')
+
+        metadata = input_data['metadata']
+
+        try:
+            payload = MinioDataFramePayload.from_dict(input_data['data'])
+            data = payload.retrieve(self.minio_repository, metadata)
+
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id='ERROR_LOADING_RETRAIN_DATA',
+                message=f'Error loading retrain data: {e}',
+                block='retrain_model',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+            self.error(trace, metadata)
+            return {
+                'success': False,
+                'message': f'Error loading retrain data: {e}',
+                'traceback': trace,
+                'timestamp': now().strftime(DATETIME_FORMAT_MS_WITH_TZ),
+            }
+
+        self.debug(f'Retrain data loaded successfully: shape {data.shape}', metadata)
+
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+
+        self.info(f'Retraining model {model_name}...', metadata)
+
+        timestamp = data['timestamp'].max()
+        self.debug(f'Timestamp: {timestamp}', metadata)
+
+        if 'created_at' in data.columns:
+            data = data.sort_values('created_at', ascending=False).drop_duplicates(
+                subset=['variable', 'timestamp'], keep='first'
+            )
+        else:
+            data = data.drop_duplicates(subset=['variable', 'timestamp'], keep='first')
+
+        data.drop(columns=['model_id'], inplace=True, errors='ignore')
+        data.drop(columns=['created_at'], inplace=True, errors='ignore')
+
+        data = data.pivot(index='timestamp', columns='variable', values='value')
+        data.fillna(np.nan, inplace=True)
+        data.columns.name = None
+        data.index.name = None
+
+        data.index = pd.DatetimeIndex(
+            to_datetime(data.index, format=DATETIME_FORMAT_WITH_TZ, utc=True)
+        )
+
+        target = model_config.get('target')
+        if target is None:
+            msg = 'model_config must include "target" for retraining'
+            self.info(msg, metadata)
+            return {
+                'success': False,
+                'experiment': None,
+                'message': msg,
+                'traceback': '',
+                'timestamp': str(timestamp),
+            }
+
+        try:
+            model_alias = self._resolve_model_alias(model_config)
+            mv_src = self.mlflow_repository._client.get_model_version_by_alias(
+                name=model_name,
+                alias=model_alias,
+            )
+            source_run_id = mv_src.run_id
+
+            wrapper = self.mlflow_repository.get_cached_model(
+                model_name=model_name,
+                alias=model_alias,
+                retention_minutes=0,
+                metadata=metadata,
+            )
+
+            # Keep heavy model fitting outside MLflow run timing.
+            prediction_data = wrapper.retrain(data)
+            prediction_data.rename(columns={target: 'prediction'}, inplace=True)
+
+            # Merge prediction data with retrain data
+            evaluation_data = pd.merge(
+                data, prediction_data, left_index=True, right_index=True, how='left'
+            )
+
+            # Rename target column to "target"
+            evaluation_data.rename(columns={target: 'target'}, inplace=True)
+
+            # Reset index and put as column "timestamp"
+            evaluation_data['timestamp'] = evaluation_data.index
+            evaluation_data.reset_index(drop=True, inplace=True)
+            evaluation_data.sort_values(by='timestamp', inplace=True, ascending=True)
+
+            run_name = f'{model_name}-retrain-{datetime.now().strftime("%Y%m%d%H%M%S")}'
+
+            with self.mlflow_repository.start_run(
+                model_name=model_name,
+                run_name=run_name,
+                experiment_name=model_name,
+                tags={'retrain': 'true', 'source_run_id': source_run_id},
+                metadata=metadata,
+            ) as run_info:
+                tmp_dir = tempfile.mkdtemp(prefix='laborious_retrain_')
+                try:
+                    raw_csv = Path(tmp_dir) / 'retrain_input.csv'
+                    evaluation_csv = Path(tmp_dir) / 'evaluation_data.csv'
+                    data.to_csv(raw_csv, index=False)
+                    evaluation_data.to_csv(evaluation_csv, index=False)
+
+                    mlflow.log_artifact(str(raw_csv))
+                    mlflow.log_artifact(str(evaluation_csv))
+                finally:
+                    rmtree(tmp_dir, ignore_errors=True)
+
+                wrapper.store_model(name=model_name)
+
+                self.mlflow_repository.log_params(
+                    {
+                        'retrain': 'true',
+                        'retrain_date': datetime.now().isoformat(),
+                        'source_run_id': source_run_id,
+                        'retrain_samples': str(data.shape),
+                    }
+                )
+
+                experiment_payload = {
+                    'run_id': run_info.run_id,
+                    'experiment_id': run_info.experiment_id,
+                    'experiment_name': model_name,
+                }
+
+            return {
+                'success': True,
+                'experiment': experiment_payload,
+                'message': 'Model retrained successfully.',
+                'timestamp': str(timestamp),
+            }
+
+        except Exception as e:
+            error_msg = f'Error retraining model {model_name}: {e}'
+            self.info(error_msg, metadata)
+            return {
+                'success': False,
+                'experiment': None,
+                'message': error_msg,
+                'traceback': traceback.format_exc(),
+                'timestamp': str(timestamp),
+            }
+
+    @activity.defn(name='update_production_model')
+    def update_production_model(self, input_data: dict[str, Any]) -> dict[Any, Any]:
+        """
+        Point the ``production`` alias at the model version registered for the retrain run.
+
+        Resolves the highest numeric registry version whose ``run_id`` matches
+        ``experiment['run_id']``, then calls ``promote_to_alias``. On failure, sends a notification
+        and re-raises so the workflow can surface the error.
+
+        Args:
+         - input_data: ``metadata``, ``model_name``, and ``experiment`` with ``run_id`` and
+           ``experiment_id`` (as returned from ``retrain_model``).
+
+        Return:
+            Dict with ``model_name``, promoted ``version``, ``mlflow_run_id``, ``mlflow_experiment_id``.
+        """
+        metadata = input_data['metadata']
+        model_name = input_data['model_name']
+        experiment = input_data['experiment']
+        self.info(
+            f'Updating production model {model_name} from experiment {experiment}...', metadata
+        )
+
+        try:
+            run_id = experiment['run_id']
+            experiment_id = experiment['experiment_id']
+
+            version = self._resolve_model_version_for_run(run_id)
+
+            promote_alias = self._resolve_model_alias(input_data.get('model_config'))
+            self.mlflow_repository.promote_to_alias(
+                model_name=model_name,
+                version=version,
+                alias=promote_alias,
+                metadata=metadata,
+            )
+
+            self.info(f'Production model {model_name} updated successfully', metadata)
+            return {
+                'model_name': model_name,
+                'version': version,
+                'mlflow_run_id': run_id,
+                'mlflow_experiment_id': experiment_id,
+            }
+
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id='UPDATE_PRODUCTION_MODEL_ERROR',
+                message=f'Error updating production model {model_name}: {e}',
+                block='update_production_model',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+            self.error(trace, metadata=metadata)
+            raise e
+
+    def _resolve_reference_artifact_name(self, run_id: str) -> str | None:
+        """
+        Pick the first available reference CSV artifact path from the MLflow run.
+
+        Candidates are checked in priority order: ``retrain_input.csv``, then ``train_data.csv``.
+        A path matches when it equals the candidate or ends with ``/<candidate>`` for nested layouts.
+
+        Args:
+         - run_id: MLflow run UUID linked to the production model version.
+
+        Return:
+            Artifact path string for ``download_artifacts``, or ``None`` if no candidate exists.
+        """
+        listed = self.mlflow_repository._client.list_artifacts(run_id)
+        paths = [file_info.path for file_info in listed]
+        for candidate in self._REFERENCE_ARTIFACT_CANDIDATES:
+            for path in paths:
+                if path == candidate or path.endswith(f'/{candidate}'):
+                    return path
+        return None
+
+    def _find_downloaded_csv(self, tmpdir: str, artifact_name: str) -> Path | None:
+        """
+        Locate a downloaded reference CSV in the temp directory.
+
+        Args:
+         - tmpdir: Directory where ``download_artifacts`` wrote files.
+         - artifact_name: Basename of the resolved artifact (e.g. ``retrain_input.csv``).
+
+        Return:
+            ``Path`` to the CSV file if found, else ``None``.
+        """
+        direct = Path(tmpdir) / artifact_name
+        if direct.exists():
+            return direct
+        matches = list(Path(tmpdir).rglob(artifact_name))
+        return matches[0] if matches else None
+
+    @activity.defn(name='get_reference_data')
+    def get_reference_data(self, input_data: dict[str, Any]) -> list[dict] | None:
+        """
+        Download reference training CSV from the MLflow run linked to the production alias.
+
+        Resolves ``retrain_input.csv`` or ``train_data.csv`` via artifact listing before download.
+        ``retrain_input.csv`` is preferred when both exist (most recent retrain snapshot). Used by
+        drift workflows to compare live data against the reference distribution logged with the model.
+        Timestamps are normalized to ``DATETIME_FORMAT`` string columns before returning records.
+
+        Args:
+         - input_data: ``metadata``, ``model_name``, and optional ``model_config`` with ``alias``.
+
+        Return:
+            List of row dicts with normalized timestamps, or ``None`` if resolution or load fails.
+        """
+        metadata = input_data['metadata']
+        model_name = input_data['model_name']
+
+        try:
+            model_alias = self._resolve_model_alias(input_data.get('model_config'))
+            mv = self.mlflow_repository._client.get_model_version_by_alias(
+                name=model_name,
+                alias=model_alias,
+            )
+            run_id = mv.run_id
+
+            artifact_path = self._resolve_reference_artifact_name(run_id)
+            if artifact_path is None:
+                self.warning(f'Reference data not found for model {model_name}', metadata)
+                return None
+
+            artifact_name = Path(artifact_path).name
+            tmpdir = tempfile.mkdtemp(prefix='laborious_eval_')
+            try:
+                self.mlflow_repository.download_artifacts(
+                    run_id=run_id,
+                    artifact_path=artifact_path,
+                    dst_path=tmpdir,
+                    metadata=metadata,
+                )
+                csv_path = self._find_downloaded_csv(tmpdir, artifact_name)
+                if csv_path is None:
+                    self.warning(f'Reference data not found for model {model_name}', metadata)
+                    return None
+                reference_data = pd.read_csv(csv_path)
+            finally:
+                rmtree(tmpdir, ignore_errors=True)
+
+            reference_data['timestamp'] = to_datetime(reference_data['timestamp'])
+            reference_data['timestamp'] = reference_data['timestamp'].dt.strftime(DATETIME_FORMAT)
+
+            return reference_data.to_dict(orient='records')
+
+        except Exception as e:
+            self.warning(f'Reference data not found for model {model_name}: {e}', metadata)
+            return None

laborious/activities/model_metrics.py

@@ -0,0 +1,432 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    import time
+    import traceback
+    import warnings
+    from typing import Any
+
+    import numpy as np
+    import pandas as pd
+    from pandas import DataFrame, Index, Series, to_datetime
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+    from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ
+    from sientia_model.analytics.drift_analysis import DriftAnalysis, DriftInsufficientDataError
+
+    from laborious import metrics
+    from laborious.utils.dataframe_debug import build_dataframe_debug_message
+
+warnings.filterwarnings('ignore', category=RuntimeWarning, message='Degrees of freedom <= 0')
+warnings.filterwarnings(
+    'ignore', category=RuntimeWarning, message='invalid value encountered in scalar divide'
+)
+
+
+class ModelMetrics(SientiaMonitoring):
+    """
+    Metrics and statistical analysis activities for the Laborious pipeline.
+
+    This class centralizes drift/statistical computations and model-quality
+    aggregates used by scheduled workflows. Besides producing tabular outputs
+    for persistence, it also emits operational metrics (count, lag, error)
+    through ``SientiaMonitoring`` so execution health is observable in runtime.
+    """
+
+    _MAX_DEBUG_DATAFRAME_ROWS = 100
+
+    def __init__(
+        self,
+        logger: Logger,
+        notification_handler: NotificationHandler,
+        metrics_controller: MetricsController,
+    ):
+        SientiaMonitoring.__init__(self, logger, notification_handler, metrics_controller)
+
+    def close(self) -> None:
+        """
+        Shutdown monitoring resources associated with model metrics activities.
+
+        This is invoked during worker teardown to flush/close metric controller
+        internals and prevent dangling telemetry tasks.
+        """
+        SientiaMonitoring.shutdown(self)
+
+    def __del__(self):
+        self.close()
+
+    def _debug_dataframe(self, message: str, data: Any, metadata: dict[str, Any]) -> None:
+        """
+        Log dataframe content only when row count is below the configured threshold
+
+        Args:
+         - message (str): Base log message to identify the dataframe in logs
+         - data (Any): Dataframe-like payload to be logged
+         - metadata (dict[str, Any]): Workflow metadata for contextual logging
+        """
+        self.debug(
+            build_dataframe_debug_message(
+                message=message,
+                data=data,
+                max_rows=self._MAX_DEBUG_DATAFRAME_ROWS,
+            ),
+            metadata,
+        )
+
+    def _drift_analyze_stage_error(
+        self,
+        exc: Exception,
+        context: str,
+        metadata: dict[str, Any],
+        core_labels: dict[str, Any],
+    ) -> None:
+        """
+        Log analyzer failure for a drift stage and increment the analyze error metric.
+
+        Args:
+         - exc (Exception): Failure raised by ``sientia_model``.
+         - context (str): Short label for the log line (e.g. univariate detection).
+         - metadata (dict[str, Any]): Workflow metadata for logging.
+         - core_labels (dict[str, Any]): Tags from ``get_core_labels`` for metrics.
+        """
+        self.error(f'{context}: {exc}', metadata)
+        self.emit_metric_sync(metric_object=metrics.MODEL_ANALYZE_ERROR_COUNT, tags=core_labels)
+
+    def get_drift_metrics(
+        self,
+        reference_data: DataFrame,
+        target_data: DataFrame,
+        target_name: str,
+        reference_columns: Index,
+        drift_metrics: list[str],
+        chunk_period: str,
+        metadata: dict[str, Any],
+    ) -> DataFrame:
+        """
+        Compute univariate and multivariate drift outputs and merge them into one dataframe.
+
+        The method orchestrates three analysis stages (univariate drift,
+        multivariate drift, and dataframe projection), emitting lag/count/error
+        metrics for each stage independently so failures are attributable.
+
+        Args:
+         - reference_data (DataFrame): Baseline dataset representing expected behavior.
+         - target_data (DataFrame): Current analysis dataset to compare against reference.
+         - target_name (str): Target column name used by ``DriftAnalysis`` config.
+         - reference_columns (Index): Feature columns evaluated for drift.
+         - drift_metrics (list[str]): Enabled univariate methods.
+         - chunk_period (str): Time bucket granularity used by analysis methods.
+         - metadata (dict[str, Any]): Workflow metadata for logs and notifications.
+
+        Return:
+         DataFrame: Consolidated drift dataframe from ``get_drift_metrics_dataframe`` using
+         ``method`` / ``value`` (and optional ``threshold``, ``drift_type``), ready for
+         activity-level formatting before Postgres export.
+        """
+        # ``DriftAnalysis`` uses truthiness checks on ``features`` (e.g. ``if not features``);
+        # a pandas ``Index`` is ambiguous in boolean context — normalize to a list.
+        feature_names: list[str] = list(reference_columns)
+
+        config = {
+            'target': target_name,
+            'prediction': 'prediction',
+            'timestamp': 'timestamp',
+            'features': feature_names,
+        }
+
+        drift_analysis = DriftAnalysis(config=config)
+
+        self._debug_dataframe(
+            f'Reference data: Size {reference_data.shape}', reference_data, metadata
+        )
+
+        self._debug_dataframe(f'Target data: Size {target_data.shape}', target_data, metadata)
+
+        core_labels = self.get_core_labels(metadata, operation_type='detect_univariate_drift')
+        start_time = time.time()
+        try:
+            univariate_drift = drift_analysis.detect_univariate_drift(
+                reference_df=reference_data,
+                analysis_df=target_data,
+                features=feature_names,
+                timestamp_col=config['timestamp'],
+                methods=drift_metrics,
+                chunk_period=chunk_period,
+            )
+        except Exception as e:
+            if isinstance(e, DriftInsufficientDataError):
+                raise
+            self._drift_analyze_stage_error(
+                e, 'Error detecting univariate drift', metadata, core_labels
+            )
+            raise
+        self.observe_lag_sync(start_time, metrics.MODEL_ANALYZE_LAG, core_labels)
+        self.emit_metric_sync(metric_object=metrics.MODEL_ANALYZE_COUNT, tags=core_labels)
+
+        core_labels = self.get_core_labels(metadata, operation_type='detect_multivariate_drift')
+        start_time = time.time()
+        try:
+            multivariate_drift = drift_analysis.detect_multivariate_drift(
+                reference_df=reference_data,
+                analysis_df=target_data,
+                features=feature_names,
+                timestamp_col=config['timestamp'],
+                chunk_period=chunk_period,
+            )
+        except Exception as e:
+            if isinstance(e, DriftInsufficientDataError):
+                raise
+            self._drift_analyze_stage_error(
+                e, 'Error detecting multivariate drift', metadata, core_labels
+            )
+            raise
+        self.observe_lag_sync(start_time, metrics.MODEL_ANALYZE_LAG, core_labels)
+        self.emit_metric_sync(metric_object=metrics.MODEL_ANALYZE_COUNT, tags=core_labels)
+
+        start_time = time.time()
+        core_labels = self.get_core_labels(metadata, operation_type='get_drift_metrics_dataframe')
+        try:
+            drift_df = drift_analysis.get_drift_metrics_dataframe(
+                univariate_drift=univariate_drift,
+                multivariate_drift=multivariate_drift,
+            )
+        except Exception as e:
+            if isinstance(e, DriftInsufficientDataError):
+                raise
+            self._drift_analyze_stage_error(
+                e, 'Error building drift metrics dataframe', metadata, core_labels
+            )
+            raise
+        self.observe_lag_sync(start_time, metrics.MODEL_ANALYZE_LAG, core_labels)
+        self.emit_metric_sync(metric_object=metrics.MODEL_ANALYZE_COUNT, tags=core_labels)
+
+        return drift_df
+
+    @staticmethod
+    def _to_naive_utc(series: Series) -> Series:
+        """
+        Parse ``series`` as datetime and return a TZ-naive UTC copy.
+
+        ``sientia_model.analytics.drift_analysis.DriftAnalysis`` preserves the
+        timezone of the input dataframe in its outputs, while target rows
+        loaded from PostgreSQL come in with ``+00:00``. Forcing both sides of
+        a comparison to TZ-naive UTC keeps ``isin`` / ``floor`` operations
+        deterministic regardless of how the analyzer constructs its
+        timestamps.
+
+        Args:
+         - series (Series): Input series containing datetime-parseable values.
+
+        Return:
+         Series: Datetime64 series with ``tz=None`` representing UTC instants.
+        """
+        parsed = to_datetime(series)
+        if getattr(parsed.dt, 'tz', None) is not None:
+            parsed = parsed.dt.tz_convert('UTC').dt.tz_localize(None)
+        return parsed
+
+    @activity.defn(name='calculate_drift')
+    def calculate_drift(self, input_data: dict[str, Any]) -> list[dict]:
+        """
+        Calculate drift metrics for a model.
+
+        Args:
+            input_data (dict[str, Any]): Input data containing:
+                - metadata (dict): Workflow execution metadata
+                - model_name (str): Name of the MLFlow model to calculate drift for
+                - reference_data (pd.DataFrame): Reference data for the model
+                - target_data (pd.DataFrame): Target data for calculating drift
+                - target_name (str): Name of the target column
+                - drift_metrics (list[str]): List of drift metrics to calculate
+        """
+        metadata = input_data['metadata']
+        model_name = input_data['model_name']
+        model_id = input_data['model_id']
+        reference_raw_data = input_data['reference_data']
+        target_data = DataFrame(input_data['target_data'])
+        target_name = input_data['target_name']
+        drift_metrics = input_data['drift_metrics']
+        chunk_period = input_data['chunk_period']
+
+        if chunk_period not in ['min', 's']:
+            self.error(f'Invalid chunk period: {chunk_period}', metadata)
+            raise ValueError(f'Invalid chunk period: {chunk_period}, must be "min" or "s"')
+
+        self.info(f'Calculating drift for model {model_name}', metadata)
+
+        target_data = target_data.pivot(index='timestamp', columns='variable', values='value')
+        target_data['timestamp'] = target_data.index
+        # Keep timestamps as datetime: DriftAnalysis._chunk_dataframe relies on
+        # ``pd.Grouper(freq=...)`` which rejects string timestamp columns.
+        target_data['timestamp'] = to_datetime(target_data['timestamp'])
+        target_data = target_data.reset_index(drop=True)
+        target_data.dropna(inplace=True)
+
+        if reference_raw_data is not None:
+            self.info('Using reference data', metadata)
+            reference_data = DataFrame(reference_raw_data)
+            if 'timestamp' in reference_data.columns:
+                reference_data['timestamp'] = to_datetime(reference_data['timestamp'])
+            accurate = True
+        else:
+            # Get 30% first rows of target_data
+            self.warning('Using 30% first rows of target data as reference data', metadata)
+            target_data.sort_values(by='timestamp', ascending=True, inplace=True)
+            reference_data = target_data.head(int(len(target_data) * 0.3))
+            accurate = False
+
+            self.send_notification(
+                metadata=metadata,
+                notification_id='MODEL_METRICS_REFERENCE_DATA_WARNING',
+                message='Using 30% first rows of target data as reference data',
+                block='model_metrics',
+                level=NotificationLevel.WARNING,
+                attachment_content=reference_data.to_csv(),
+            )
+
+        reference_columns = reference_data.drop(
+            columns=[target_name, 'timestamp', 'target', 'prediction'], errors='ignore'
+        ).columns
+
+        try:
+            drift_df = self.get_drift_metrics(
+                reference_data=reference_data,
+                target_data=target_data,
+                target_name=target_name,
+                reference_columns=reference_columns,
+                drift_metrics=drift_metrics,
+                chunk_period=chunk_period,
+                metadata=metadata,
+            )
+        except Exception as e:
+            if isinstance(e, DriftInsufficientDataError):
+                self.error(str(e), metadata)
+                notification_id = e.notification_id
+                notification_message = str(e)
+            else:
+                self.error(f'Error getting drift metrics: {e}', metadata)
+                notification_id = 'MODEL_METRICS_GET_DRIFT_METRICS_ERROR'
+                notification_message = f'Error getting drift metrics: {e}'
+            self.send_notification(
+                metadata=metadata,
+                notification_id=notification_id,
+                message=notification_message,
+                block='model_metrics',
+                level=NotificationLevel.ERROR,
+                attachment_content=traceback.format_exc(),
+            )
+            raise
+
+        # Drop chunks whose floored timestamp does not appear in the analysis window.
+        # ``DriftAnalysis`` chunks over ``analysis_df``; this only excludes rows that
+        # do not belong to the current target window (e.g. stray merged reference rows).
+        target_floor = self._to_naive_utc(target_data['timestamp']).dt.floor(chunk_period)
+        drift_floor = self._to_naive_utc(drift_df['timestamp']).dt.floor(chunk_period)
+        drift_df = drift_df[drift_floor.isin(target_floor)]
+
+        if drift_df.empty:
+            self.warning(
+                'No drift metrics found after dropping rows where timestamp is not in target data',
+                metadata,
+            )
+            return []
+
+        # Analyzer emits diagnostic columns that are not stored in ``sientia_data.drift_metrics``.
+        drift_df = drift_df.drop(columns=['threshold', 'drift_type'], errors='ignore')
+
+        drift_df['model_id'] = str(model_id)
+        drift_df['accurate'] = accurate
+
+        # ``timestamp`` is overridden with the most recent target instant so
+        # every persisted row shares a single business timestamp (the run's
+        # logical "now"), matching what downstream consumers expect.
+        latest_target_timestamp = self._to_naive_utc(target_data['timestamp']).max()
+        drift_df['timestamp'] = (
+            pd.Timestamp(latest_target_timestamp)
+            .tz_localize('UTC')
+            .strftime(DATETIME_FORMAT_WITH_TZ)
+        )
+
+        # ``chunk_start_date`` / ``chunk_end_date`` may carry nanosecond
+        # precision (beyond ``timestamptz`` microseconds), so serialize as ISO
+        # text for the ``text`` Postgres columns.
+        for column in ('chunk_start_date', 'chunk_end_date'):
+            drift_df[column] = drift_df[column].apply(
+                lambda value: pd.Timestamp(value).isoformat() if pd.notna(value) else None
+            )
+
+        self._debug_dataframe(f'Drift dataframe: Size {drift_df.shape}', drift_df, metadata)
+
+        return drift_df.to_dict(orient='records')
+
+    @activity.defn(name='calculate_simple_metrics')
+    def calculate_simple_metrics(self, input_data: dict[str, Any]) -> list[dict]:
+        """
+        Calculate simple metrics for a model. Metrics available are:
+            - rmse
+            - mse
+            - mae
+            - r2
+            - accuracy
+            - precision
+            - recall
+            - f1
+        Args:
+            input_data (dict[str, Any]): Input data containing:
+                - metadata (dict): Workflow execution metadata
+                - model_id (str): ID of the MLFlow model
+                - target_data (pd.DataFrame): Target data for calculating metrics, containing target and prediction columns
+                - metrics (list[str]): List of metrics to calculate
+        Returns:
+            dict[Hashable, Any]: Dictionary containing the calculated metrics
+        """
+
+        metadata = input_data['metadata']
+        model_id = input_data['model_id']
+        target_data = DataFrame(input_data['target_data'])
+        metric_names = input_data['metrics']
+        interval_minutes = input_data['interval_minutes']
+
+        data_size = target_data.shape[0]
+
+        output_data = []
+
+        diff = target_data['target'] - target_data['prediction']
+        diff_squared = diff**2
+
+        self.info(f'Calculating simple metrics for model {model_id}: {metric_names}', metadata)
+
+        for metric in metric_names:
+            if metric == 'rmse':
+                output_data.append({'metric': 'rmse', 'value': np.sqrt(np.mean(diff_squared))})
+            elif metric == 'mse':
+                output_data.append({'metric': 'mse', 'value': np.mean(diff_squared)})
+            elif metric == 'mae':
+                output_data.append({'metric': 'mae', 'value': np.mean(np.abs(diff))})
+            elif metric == 'r2':
+                y_true = target_data['target']
+                y_mean = np.mean(y_true)
+
+                ss_res = np.sum(diff_squared)
+                ss_tot = np.sum((y_true - y_mean) ** 2)
+
+                # Evita divisão por zero
+                if ss_tot == 0:
+                    r2_score = 0.0
+                else:
+                    r2_score = 1 - (ss_res / ss_tot)
+
+                output_data.append({'metric': 'r2', 'value': r2_score})
+
+        data = DataFrame(output_data)
+        data['model_id'] = model_id
+        data['timestamp'] = target_data['timestamp'].max()
+        data['data_size'] = data_size
+        data['interval_minutes'] = interval_minutes
+
+        self._debug_dataframe(f'Simple metrics dataframe: Size {data.shape}', data, metadata)
+
+        return data.to_dict(orient='records')

laborious/activities/opc.py

@@ -0,0 +1,514 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    import traceback
+    from collections.abc import Hashable
+    from typing import Any
+
+    from pandas import DataFrame
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+
+    from laborious.utils.repository.opc_repository import OpcRepository
+
+OPC_WRITTING_ERROR_CONFIDENCE = 12
+OPC_SESSION_BAD_CONFIDENCE = 14
+OPC_SESSION_BAD_COMMENT_PREFIX = 'OPC UA session/channel error:'
+OPC_WRITTING_ERROR_MESSAGE = 'Some data could not be written to OPC servers'
+OPC_RECONNECT_IN_PROGRESS_COMMENT = 'OPC UA reconnect in progress'
+OPC_COMMENT_SEPARATOR = ' | '
+
+
+def _opc_session_bad_comment(opc_status: str | None) -> str:
+    status = opc_status or 'Unknown'
+    return f'{OPC_SESSION_BAD_COMMENT_PREFIX} {status}'
+
+
+def _apply_opc_write_error(
+    error_info: dict[str, Any] | None,
+    session_bad_seen: bool,
+    session_bad_status: str | None,
+    reconnect_in_progress_seen: bool,
+) -> tuple[bool, str | None, bool]:
+    """
+    Update session/reconnect flags from an OPC write error payload.
+
+    Args:
+        error_info: Repository error details, or None when the write succeeded.
+        session_bad_seen: Whether a session_bad error was seen so far.
+        session_bad_status: Last known OPC status for session errors.
+        reconnect_in_progress_seen: Whether reconnect_in_progress was seen so far.
+
+    Return:
+        Updated (session_bad_seen, session_bad_status, reconnect_in_progress_seen).
+    """
+    if not error_info:
+        return session_bad_seen, session_bad_status, reconnect_in_progress_seen
+
+    kind = error_info.get('opc_error_kind')
+    if kind == 'session_bad':
+        return True, error_info.get('opc_status', session_bad_status), reconnect_in_progress_seen
+    if kind == 'reconnect_in_progress':
+        return session_bad_seen, session_bad_status, True
+    return session_bad_seen, session_bad_status, reconnect_in_progress_seen
+
+
+class OPC(SientiaMonitoring):
+    """
+    OPC server integration activities for real-time data export.
+
+    This class provides comprehensive OPC UA client functionality for connecting
+    to multiple OPC servers and writing prediction data in real-time. It implements
+    secure communication with certificate-based authentication and automatic
+    reconnection capabilities.
+
+    The class supports multiple OPC servers with individual configurations and
+    provides robust error handling and monitoring for production environments.
+
+    Attributes:
+        opc_servers (dict): Configuration for multiple OPC servers
+        opc_repository (dict): Active OPC repository connections
+        logger (Logger): Logging and observability instance
+        notification_handler (NotificationHandler): Notification management instance
+    """
+
+    def __init__(
+        self,
+        opc_servers: dict[str, dict[str, Any]],
+        logger: Logger,
+        notification_handler: NotificationHandler,
+        metrics_controller: MetricsController,
+    ):
+        self.opc_servers = opc_servers
+
+        SientiaMonitoring.__init__(self, logger, notification_handler, metrics_controller)
+
+        self.opc_repository: dict[str, OpcRepository] = {}
+
+    def init_opc(self):
+        """
+        Initialize OPC server connections and establish communication channels.
+
+        This method iterates through all configured OPC servers and attempts to
+        establish secure connections using certificate-based authentication.
+        Each server connection is managed independently, and connection failures
+        are reported through the notification system.
+
+        The method performs the following operations:
+        1. Creates OpcRepository instances for each configured server
+        2. Establishes secure connections with certificate validation
+        3. Reports connection success/failure through notifications
+        4. Logs connection status for operational visibility
+
+        Raises:
+            Exception: If OPC repository initialization fails or connection
+                establishment encounters critical errors
+
+        Note:
+            Connection failures are logged and reported but do not prevent
+            the initialization of other OPC servers. Each server is handled
+            independently to ensure maximum availability.
+        """
+        self.info('Initializing OPC servers...')
+        for opc_id, server in self.opc_servers.items():
+            self.opc_repository[opc_id] = OpcRepository(
+                opc_id=opc_id,
+                server_name=server['server_name'],
+                url=server['url'],
+                logger=self.logger,
+                server_uri=server['server_uri'],
+                cert_path=server['cert_path'],
+                private_key_path=server['private_key_path'],
+                server_cert_path=server['server_cert_path'],
+                notification_handler=self.notification_handler,
+                reconnection_interval=server.get('reconnection_interval', 60),
+                metrics_controller=self.metrics_controller,
+            )
+            is_connected, error_data = self.opc_repository[opc_id].connect()
+            if not is_connected:
+                self.send_notification(
+                    metadata={
+                        'model_id': '-',
+                        'model_name': '-',
+                        'workflow_name': '-',
+                        'schedule_name': 'INITIALIZATION',
+                    },
+                    notification_id=error_data['notification_id'],
+                    message=error_data['message'],
+                    block=error_data['block'],
+                    level=error_data.get('level', NotificationLevel.ERROR),
+                    attachment_content=error_data.get('attachment_content', None),
+                )
+            else:
+                self.info(f'OPC server {opc_id}:{server["server_name"]} connected successfully.')
+
+    def write_data(
+        self,
+        server_id: str,
+        tag: str,
+        data: Any,
+        data_type: str,
+        tag_type: str,
+        metadata: dict[str, Any],
+    ) -> tuple[float | None, dict[str, Any] | None]:
+        """
+        Write data to a specific OPC server tag with comprehensive error handling.
+
+        Return:
+            tuple[float | None, dict[str, Any] | None]: Response time on success, or
+                (None, error info_data) on repository failure.
+        """
+
+        try:
+            is_success, info_data = self.opc_repository[server_id].write_data(
+                tag, data, data_type, metadata
+            )
+            if not is_success:
+                self.send_notification(
+                    metadata=metadata,
+                    notification_id=info_data['notification_id'],
+                    message=info_data['message'],
+                    block=info_data['block'],
+                    level=info_data.get('level', NotificationLevel.ERROR),
+                    attachment_content=info_data.get('attachment_content', None),
+                )
+                return None, info_data
+            return info_data['response_time'], None
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id=f'WRITE_OPC_{tag_type.upper()}_ERROR',
+                message=f'Error writing data to OPC server: {e}',
+                block='write_opc_data',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+            raise
+
+    def validate_server(self, server_id: str, metadata: dict[str, Any]) -> bool:
+        """
+        Validate that an OPC server is available and configured for write operations.
+
+        This method checks if the specified OPC server exists in the active
+        repository and is available for data writing operations. It provides
+        immediate feedback for server availability and logs validation failures
+        for operational monitoring.
+
+        Args:
+            server_id (str): Unique identifier for the OPC server to validate
+            metadata (dict[str, Any]): Context metadata for logging and notifications
+
+        Returns:
+            bool: True if server is available, False otherwise
+
+        Note:
+            Server validation failures are automatically reported through the
+            notification system with detailed information about available servers.
+            This helps operators quickly identify configuration issues.
+        """
+        if self.opc_repository.get(server_id) is None:
+            message = f'OPC server {server_id} not found to perform write operation.'
+            self.send_notification(
+                metadata=metadata,
+                notification_id='OPC_SERVER_NOT_FOUND',
+                message=message,
+                block='write_opc_data',
+                level=NotificationLevel.ERROR,
+                attachment_content=f'OPC servers: {list(self.opc_repository.keys())}',
+            )
+            return False
+        return True
+
+    def _write_tags_from_config(
+        self,
+        server_id: str,
+        tags_config: dict[str, dict[str, Any]],
+        data: DataFrame,
+        data_column: str,
+        tag_type: str,
+        log_label: str,
+        metadata: dict[str, Any],
+    ) -> tuple[dict[str, float | None], bool, str | None, bool]:
+        """
+        Write a group of OPC tags and collect response times and error flags.
+
+        Args:
+            server_id: Target OPC server identifier.
+            tags_config: Tag name to configuration mapping.
+            data: DataFrame with prediction/confidence columns.
+            data_column: Column name whose first row value is written.
+            tag_type: Tag category passed to write_data ('prediction' or 'confidence').
+            log_label: Human-readable label for success logs.
+            metadata: Context metadata for logging and notifications.
+
+        Return:
+            (response_times, session_bad_seen, session_bad_status, reconnect_in_progress_seen)
+        """
+        response_times: dict[str, float | None] = {}
+        session_bad_seen = False
+        session_bad_status: str | None = None
+        reconnect_in_progress_seen = False
+
+        for tag, tag_config in tags_config.items():
+            response_time, error_info = self.write_data(
+                server_id=server_id,
+                tag=tag,
+                data=data.head(1)[data_column].values[0],
+                data_type=tag_config['data_type'],
+                tag_type=tag_type,
+                metadata=metadata,
+            )
+            session_bad_seen, session_bad_status, reconnect_in_progress_seen = (
+                _apply_opc_write_error(
+                    error_info,
+                    session_bad_seen,
+                    session_bad_status,
+                    reconnect_in_progress_seen,
+                )
+            )
+            if response_time is not None:
+                self.info(
+                    f'{log_label} written to OPC server {server_id} for tag {tag}.',
+                    metadata,
+                )
+            response_times[tag] = response_time
+
+        return response_times, session_bad_seen, session_bad_status, reconnect_in_progress_seen
+
+    def manage_output_tags(
+        self,
+        server_id: str,
+        config: dict[str, Any],
+        data: DataFrame,
+        metadata: dict[str, Any],
+    ) -> tuple[bool, dict[str, float | None], bool, str | None, bool]:
+        """
+        Manage the writing of prediction and confidence data to OPC server tags.
+
+        This method orchestrates the writing of multiple data types to OPC servers
+        based on configuration. It handles both prediction data and confidence
+        values independently, allowing for flexible tag configuration and
+        comprehensive error handling.
+
+        The method supports two main tag types:
+        1. Prediction tags: Write actual prediction values to configured OPC tags
+        2. Confidence tags: Write confidence scores to separate OPC tags
+
+        Args:
+            server_id (str): Unique identifier for the target OPC server
+            config (dict[str, Any]): OPC tag configuration containing:
+                - prediction_tags (dict, optional): Prediction tag configurations
+                - confidence_tags (dict, optional): Confidence tag configurations
+            data (DataFrame): DataFrame containing prediction and confidence data
+            metadata (dict[str, Any]): Context metadata for logging and notifications
+            success (bool): Current success status to maintain across operations
+
+        Returns:
+            tuple[bool, int]: (overall_success, total_tags_written)
+                - overall_success: True if all configured tags were written successfully
+                - total_tags_written: Count of successfully written tags
+        """
+        response_times: dict[str, float | None] = {}
+        session_bad_seen = False
+        session_bad_status: str | None = None
+        reconnect_in_progress_seen = False
+
+        tag_groups = (
+            ('prediction_tags', 'prediction', 'prediction', 'Prediction data'),
+            ('confidence_tags', 'prediction_confidence', 'confidence', 'Confidence data'),
+        )
+        for config_key, data_column, tag_type, log_label in tag_groups:
+            if config_key not in config:
+                continue
+            (
+                group_times,
+                group_session_bad,
+                group_status,
+                group_reconnect,
+            ) = self._write_tags_from_config(
+                server_id=server_id,
+                tags_config=config[config_key],
+                data=data,
+                data_column=data_column,
+                tag_type=tag_type,
+                log_label=log_label,
+                metadata=metadata,
+            )
+            response_times.update(group_times)
+            if group_session_bad:
+                session_bad_seen = True
+                session_bad_status = group_status or session_bad_status
+            if group_reconnect:
+                reconnect_in_progress_seen = True
+
+        success = None not in response_times.values()
+        return (
+            success,
+            response_times,
+            session_bad_seen,
+            session_bad_status,
+            reconnect_in_progress_seen,
+        )
+
+    @activity.defn(name='write_opc_data')
+    def write_opc_data(
+        self, input_data: dict[str, Any]
+    ) -> tuple[dict[Hashable, Any], dict[str, dict[str, float | None]]]:
+        """
+        Write prediction and confidence data to OPC servers. The two writing
+        operations are optional and independent of each other.
+
+        Args:
+            - input_data(dict[str, Any]): The input data. Contains the following keys:
+                - data(dict[str, Any]): The dataframe that contains the data to write
+                to the OPC servers.
+                - opc_output_config(dict[str, Any]): The OPC writing configuration.
+                The keys are the OPC server names and the values contain:
+                    - prediction_tags(dict[str, Any]): The tags to write to the OPC servers.
+                    - confidence_tags(dict[str, Any]): The tags to write to the OPC servers.
+
+        Returns:
+            - dict[Any, Any]: The data that was written to the OPC servers.
+
+        """
+        metadata = input_data['metadata']
+        self.info('Writing data to OPC servers...', metadata)
+        data = DataFrame(input_data['data'])
+        opc_output_config = input_data['opc_output_config']
+        self.info(f'Data to write: {data.size} rows', metadata)
+
+        success = True
+        session_bad_seen = False
+        session_bad_status: str | None = None
+        reconnect_in_progress_seen = False
+
+        opc_metrics: dict[str, dict[str, float | None]] = {}
+
+        for server_id, config in opc_output_config.items():
+            if not self.validate_server(server_id, metadata):
+                success = False
+                continue
+
+            (
+                local_success,
+                local_response_times,
+                local_session_bad,
+                local_status,
+                local_reconnect_in_progress,
+            ) = self.manage_output_tags(server_id, config, data, metadata)
+            opc_metrics[server_id] = local_response_times
+            local_count = len(local_response_times)
+            success = success and local_success
+            if local_session_bad:
+                session_bad_seen = True
+                session_bad_status = local_status or session_bad_status
+            if local_reconnect_in_progress:
+                reconnect_in_progress_seen = True
+
+            n_pred = len(config.get('prediction_tags') or {})
+            n_conf = len(config.get('confidence_tags') or {})
+            self.info(
+                f'Process completed for OPC server {server_id}: {local_count} of {n_pred} prediction tags and {n_conf} confidence tags',
+                metadata,
+            )
+
+        return (
+            self.process_confidence(
+                data,
+                success,
+                metadata,
+                session_bad=session_bad_seen,
+                opc_status=session_bad_status,
+                reconnect_in_progress=reconnect_in_progress_seen,
+            ),
+            opc_metrics,
+        )
+
+    def process_confidence(
+        self,
+        data: DataFrame,
+        success: bool,
+        metadata: dict[str, Any],
+        *,
+        session_bad: bool = False,
+        opc_status: str | None = None,
+        reconnect_in_progress: bool = False,
+    ) -> dict[Hashable, Any]:
+        """
+        Process prediction confidence based on OPC write operation success.
+
+        This method updates the prediction confidence values in the DataFrame
+        based on the success status of OPC server write operations. If any
+        write operations failed, it sets the confidence to a predefined error
+        value to indicate data quality issues.
+
+        The method implements a confidence degradation strategy:
+        - Success: Maintains original confidence values
+        - Failure: Sets confidence to error value for operational awareness
+
+        Args:
+            data (DataFrame): DataFrame containing prediction and confidence data
+            success (bool): Overall success status of OPC write operations
+            metadata (dict[str, Any]): Context metadata for logging and notifications
+
+        Returns:
+            dict[Any, Any]: Processed data as a dictionary with updated confidence values
+
+        Note:
+            The error confidence value (OPC_WRITTING_ERROR_CONFIDENCE = 12) is
+            used to indicate that data was not successfully exported to OPC servers.
+            This allows downstream systems to handle data quality appropriately.
+        """
+
+        if not success:
+            comment_parts: list[str] = []
+            confidence = OPC_WRITTING_ERROR_CONFIDENCE
+
+            if session_bad:
+                comment_parts.append(_opc_session_bad_comment(opc_status))
+                confidence = OPC_SESSION_BAD_CONFIDENCE
+            if reconnect_in_progress:
+                comment_parts.append(OPC_RECONNECT_IN_PROGRESS_COMMENT)
+                confidence = OPC_SESSION_BAD_CONFIDENCE
+            if not comment_parts:
+                comment_parts.append(OPC_WRITTING_ERROR_MESSAGE)
+
+            comments = OPC_COMMENT_SEPARATOR.join(comment_parts)
+            data['prediction_confidence'] = confidence
+            data['comments'] = comments
+            self.debug(
+                f'OPC write issues, confidence={confidence}, comments={comments}',
+                metadata,
+            )
+        else:
+            self.debug('Data written to OPC servers successfully.', metadata)
+
+        return data.to_dict()
+
+    def close(self):
+        """
+        Gracefully shutdown all OPC server connections and cleanup resources.
+
+        This method ensures proper cleanup of all active OPC server connections
+        by calling the disconnect method on each repository instance. It's
+        designed to be called during application shutdown to prevent resource
+        leaks and ensure clean termination.
+
+        The method performs the following cleanup operations:
+        1. Iterates through all active OPC repository connections
+        2. Calls disconnect() on each repository instance
+        3. Allows for graceful connection termination
+        4. Prevents resource leaks and connection hanging
+
+        Note:
+            This method should be called during application shutdown to ensure
+            proper cleanup. It handles all active connections regardless of
+            their current state and provides a clean shutdown experience.
+        """
+        for opc in self.opc_repository.values():
+            opc.disconnect()
+        self.opc_repository.clear()

laborious/activities/storage.py

@@ -0,0 +1,231 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    # Extend the Temporal Postgres activities for convenient query -> MinIO export
+    import traceback
+    from datetime import timedelta
+    from typing import Any
+
+    import pandas as pd
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.notifications.models import NotificationLevel
+    from sientia_do.observability.logger import Logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+    from sientia_do.repository.minio_repository_sync import MinioRepository
+    from sientia_do.temporal.activities.postgres_sync import Postgres
+    from sientia_do.temporal.constants import now
+
+    from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
+
+_LOAD_QUERY_OFFLOAD_SKIP_KEYS = frozenset({'model_name', 'key_prefix', 'size_threshold_bytes'})
+
+
+class Storage(Postgres, SientiaMonitoring):
+    """
+    Extensions for Postgres activities with a helper to export query results
+    directly to MinIO as Parquet and return the object name.
+    """
+
+    minio_repository: MinioRepository | None = None
+
+    def __init__(
+        self,
+        host: str,
+        port: int,
+        user: str,
+        password: str,
+        dbname: str,
+        min_connections: int,
+        max_connections: int,
+        retention_hours: int = 24,
+        minio_repository: MinioRepository | None = None,
+        logger: Logger | None = None,
+        notification_handler: NotificationHandler | None = None,
+        metrics_controller: MetricsController | None = None,
+    ):
+        self.retention_hours = retention_hours
+        Postgres.__init__(
+            self,
+            host=host,
+            port=port,
+            user=user,
+            password=password,
+            dbname=dbname,
+            min_connections=min_connections,
+            max_connections=max_connections,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+        )
+
+        self.minio_repository = minio_repository
+        SientiaMonitoring.__init__(
+            self,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+        )
+
+    @activity.defn(name='load_query_with_minio_offload')
+    def load_query_with_minio_offload(self, input_data: dict[str, Any]) -> MinioDataFramePayload:
+        """
+        Run the custom SQL load, then return a MinIO-aware dataframe wire dict.
+
+        Args (input_data):
+            metadata (dict): Workflow metadata (same as load_custom_query).
+            query (str): SQL query.
+            datetime_columns (list[str], optional): Datetime column names.
+            model_name (str): Model name for object key basename.
+            key_prefix (str, optional): Directory prefix inside the bucket.
+            size_threshold_bytes (int, optional): Override env offload threshold.
+
+        Returns:
+            dict[str, Any]: Flat ``MinioDataFramePayload`` dict or ``success: False`` on failure.
+        """
+        if self.minio_repository is None:
+            raise ValueError('Minio repository not initialized')
+
+        metadata: dict = input_data.get('metadata', {})
+        model_name = input_data['model_name']
+
+        rows = self.load_custom_query(
+            input_data,
+        )
+        if not rows:
+            self.error(
+                'load_query_with_minio_offload failed: No data returned from query', metadata
+            )
+            dataframe = None
+        else:
+            dataframe = pd.DataFrame(rows)
+
+        return MinioDataFramePayload.from_dataframe(
+            dataframe,
+            minio_repo=self.minio_repository,
+            workflow_metadata=metadata,
+            model_name=model_name,
+            operation='initial',
+            logger=self.logger,
+        )
+
+    @activity.defn(name='export_payload_to_postgres')
+    def export_payload_to_postgres(self, input_data: dict[str, Any]) -> dict:
+        """
+        Resolve a MinIO-aware payload into a DataFrame and persist it into PostgreSQL.
+
+        This activity accepts the serialized payload produced by previous steps
+        (inline dict or MinIO object reference), reconstructs the tabular data,
+        and delegates the final write to ``export_data_to_postgres`` using the
+        same input contract expected by the Postgres activity mixin.
+
+        Args:
+         - input_data (dict[str, Any]): Activity input containing ``data`` as a
+           ``MinioDataFramePayload``-compatible dict plus database write options
+           (schema/table/on_conflict/metadata and related fields).
+
+        Return:
+         dict: Result dictionary returned by ``export_data_to_postgres``, including
+         success status and optional write diagnostics.
+        """
+        metadata = input_data.get('metadata')
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        data = payload.retrieve(self.minio_repository, metadata)
+
+        return self.export_data_to_postgres(
+            {
+                **input_data,
+                'data': data,
+            }
+        )
+
+    @activity.defn(name='cleanup_minio_objects_expired')
+    def cleanup_minio_objects_expired(self, input_data: dict[str, Any]) -> dict[str, Any]:
+        """
+        Delete objects under the given prefixes that are older than the retention window.
+
+        Args (input_data):
+            metadata (dict): Workflow metadata for logging and metrics.
+            prefixes (list[str]): Key prefixes to scan (one level or subtree per prefix).
+
+        Returns:
+            dict[str, Any]: ``success``, ``deleted_count``, and optional ``message``.
+        """
+        if self.minio_repository is None:
+            raise ValueError('Minio repository not initialized')
+
+        metadata = input_data.get('metadata', {})
+        payload = MinioDataFramePayload.from_dict(input_data['data'])
+        prefix = payload.cleanup_prefix()
+        base = now()
+        cutoff = (base.replace(tzinfo=None) if base.tzinfo else base) - timedelta(
+            hours=self.retention_hours
+        )
+
+        report: dict[str, Any] = {
+            'failed': {},
+            'deleted': {},
+            'failed_count': 0,
+            'deleted_count': 0,
+        }
+        try:
+            keys = self.minio_repository.list_objects(
+                prefix=prefix,
+                recursive=True,
+                metadata=metadata,
+            )
+            for key in keys:
+                try:
+                    ts = MinioDataFramePayload.parse_object_timestamp(key)
+                    if ts is None:
+                        continue
+                    if ts >= cutoff:
+                        continue
+                    self.minio_repository.delete_file(
+                        object_name=key,
+                        metadata=metadata,
+                    )
+                except Exception as e:
+                    report['failed'][key] = {
+                        'success': False,
+                        'message': str(e),
+                    }
+                    report['failed_count'] += 1
+                    continue
+                report['deleted'][key] = {
+                    'success': True,
+                    'message': 'Deleted',
+                }
+                report['deleted_count'] += 1
+        except Exception as e:
+            trace = traceback.format_exc()
+            self.send_notification(
+                metadata=metadata,
+                notification_id='ERROR_CLEANUP_MINIO_OBJECTS_EXPIRED',
+                message=f'Error cleaning up MinIO objects: {e}',
+                block='cleanup_minio_objects_expired',
+                level=NotificationLevel.ERROR,
+                attachment_content=trace,
+            )
+            self.error(trace, metadata)
+        else:
+            # Cleanup success is expected in normal flow; avoid noisy INFO notifications
+            # that do not impact behavior and can flood observability in test runs.
+            self.info('MinIO objects cleaned up successfully', metadata)
+
+        return report
+
+    def close(self) -> None:
+        """
+        Shutdown Storage resources in deterministic order.
+
+        The method first closes Postgres resources via ``Postgres.close`` (engine,
+        sessions, and monitoring hooks), then closes the optional MinIO repository
+        and clears the local reference to avoid accidental reuse after shutdown.
+        """
+        Postgres.close(self)
+        if self.minio_repository is not None:
+            try:
+                self.minio_repository.close()
+            finally:
+                self.minio_repository = None

laborious/metrics.py

@@ -0,0 +1,200 @@
+"""
+Laborious Metrics Module
+
+This module defines all Prometheus metrics used by the Sientia DataOps Laborious system
+for monitoring and observability. The metrics provide insights into system performance,
+prediction quality, and operational health.
+
+The metrics are designed to be scraped by Prometheus and can be visualized in
+Grafana or other monitoring dashboards to provide real-time visibility into
+the system's operation.
+
+Key Metric Categories:
+- Application Health: Overall system status and availability
+- Prediction Operations: Count and performance of prediction operations
+- Data Quality: Confidence levels and validation results
+- Export Operations: Database and OPC export performance
+- Response Times: Performance monitoring for various operations
+
+Metric Labels:
+- pod_id: Kubernetes pod identifier for multi-instance deployments
+- runtime: Runtime / environment identifier (matches ``RUNTIME`` env, see ``SientiaMonitoring``)
+- model_name: Name of the ML model being used
+- workflow_name: Name of the prediction pipeline
+- opc_server_id: Identifier for OPC server operations
+"""
+
+from prometheus_client import Counter, Gauge, Histogram
+from sientia_do.observability.metrics import CORE_LABELS
+
+# Application health metric
+APP_UP = Gauge(
+    'app_up',
+    'Indicates if the application is running (1) or shutting down (0)',
+    ['pod_id'],
+)
+
+# Prediction operation metrics
+PREDICTIONS_WRITTEN_COUNT = Counter(
+    'laborious_predictions_written_count',
+    'Number of predictions written to the database table predictions',
+    CORE_LABELS,
+)
+
+# Prediction quality metrics
+PREDICTION_CONFIDENCE_MONITOR = Gauge(
+    'laborious_prediction_confidence_monitor',
+    'Current confidence of each prediction',
+    CORE_LABELS,
+)
+
+# Prediction total response time
+PREDICTION_RESPONSE_TIME_MONITOR = Histogram(
+    'laborious_prediction_response_time_monitor',
+    'Current response time of each prediction',
+    CORE_LABELS,
+    buckets=[0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0],
+)
+
+
+# ================== OPC metrics ==================
+
+
+PREDICTION_OPC_WRITING_COUNT = Counter(
+    'laborious_prediction_opc_writing_count',
+    'Number of predictions written to the OPC server',
+    [*CORE_LABELS, 'opc_server_id', 'tag'],
+)
+
+PREDICTION_OPC_WRITING_RESPONSE_TIME_MONITOR = Histogram(
+    'laborious_prediction_opc_writing_response_time_monitor',
+    'Current response time of each prediction written to the OPC server',
+    [*CORE_LABELS, 'opc_server_id', 'tag'],
+    buckets=[0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0],
+)
+
+OPC_CONNECTIONS_TOTAL = Counter(
+    'opc_connections_initiated_total',
+    'Total connection attempts to OPC servers',
+    ['pod_id', 'server_name'],
+)
+OPC_CONNECTIONS_FAILED = Counter(
+    'opc_connections_failed_total',
+    'Total failed connection attempts to OPC servers',
+    ['pod_id', 'server_name'],
+)
+OPC_CONNECTION_STATUS = Gauge(
+    'opc_connection_status',
+    'Connection status with the OPC server (1=connected, 0=disconnected)',
+    ['pod_id', 'server_name', 'server_url'],
+)
+
+_OPC_SESSION_DEBUG_LABELS = ['pod_id', 'server_name', 'runtime', 'opc_server_id', 'session_id']
+
+OPC_SESSION_CREATED_TOTAL = Counter(
+    'opc_session_created_total',
+    'OPC UA sessions established (after successful connect)',
+    _OPC_SESSION_DEBUG_LABELS,
+)
+
+OPC_SESSION_CLOSED_TOTAL = Counter(
+    'opc_session_closed_total',
+    'OPC UA client disconnects completed (session tear-down initiated)',
+    _OPC_SESSION_DEBUG_LABELS,
+)
+
+OPC_SESSION_REVISED_TIMEOUT_MS = Gauge(
+    'opc_session_revised_timeout_milliseconds',
+    'Server-revised OPC UA session timeout (RevisedSessionTimeout) in ms after connect',
+    _OPC_SESSION_DEBUG_LABELS,
+)
+
+OPC_WRITE_ATTEMPT_LABELS = [*_OPC_SESSION_DEBUG_LABELS, 'model_id', 'model_name', 'result']
+
+OPC_WRITE_ATTEMPTS_TOTAL = Counter(
+    'opc_write_attempts_total',
+    'OPC UA write attempts with session and outcome (result=OK or exception class name)',
+    OPC_WRITE_ATTEMPT_LABELS,
+)
+
+OPC_WRITE_INTER_ARRIVAL_OVER_SESSION_TIMEOUT_TOTAL = Counter(
+    'opc_write_inter_arrival_over_session_timeout_total',
+    'Successful writes where seconds since the previous successful write exceeded RevisedSessionTimeout (ms)',
+    _OPC_SESSION_DEBUG_LABELS,
+)
+
+# ================== Model metrics ==================
+
+MODEL_READ_LAG = Histogram(
+    'laborious_model_read_lag',
+    'Lag between the start and read of read operations',
+    CORE_LABELS,
+    buckets=[0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0],
+)
+
+MODEL_WRITE_LAG = Histogram(
+    'laborious_model_write_lag',
+    'Lag between the start and end of write operations',
+    CORE_LABELS,
+    buckets=[0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0],
+)
+
+MODEL_READ_COUNT = Counter(
+    'laborious_model_read_count',
+    'Number of reads from the model',
+    CORE_LABELS,
+)
+
+MODEL_WRITE_COUNT = Counter(
+    'laborious_model_write_count',
+    'Number of writes to the model',
+    CORE_LABELS,
+)
+
+MODEL_READ_ERROR_COUNT = Counter(
+    'laborious_model_read_error_count',
+    'Number of errors reading from the model',
+    CORE_LABELS,
+)
+
+MODEL_WRITE_ERROR_COUNT = Counter(
+    'laborious_model_write_error_count',
+    'Number of errors writing to the model',
+    CORE_LABELS,
+)
+
+MODEL_ANALYZE_LAG = Histogram(
+    'laborious_model_analyze_lag',
+    'Lag between the start and end of analyze operations',
+    CORE_LABELS,
+    buckets=[0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0],
+)
+
+MODEL_ANALYZE_COUNT = Counter(
+    'laborious_model_analyze_count',
+    'Number of analyze operations',
+    CORE_LABELS,
+)
+
+MODEL_ANALYZE_ERROR_COUNT = Counter(
+    'laborious_model_analyze_error_count',
+    'Number of errors during analyze operations',
+    CORE_LABELS,
+)
+
+
+# ================== PI Web API metrics ==================
+
+PI_WEB_API_LABELS = [*CORE_LABELS, 'tag_name']
+
+PI_WEB_API_PREDICTION_WRITTEN_COUNT = Counter(
+    'laborious_pi_web_api_prediction_written_count',
+    'Number of predictions written to the PI Web API',
+    PI_WEB_API_LABELS,
+)
+
+PI_WEB_API_PREDICTION_WRITTEN_ERROR_COUNT = Counter(
+    'laborious_pi_web_api_prediction_written_error_count',
+    'Number of errors writing predictions to the PI Web API',
+    PI_WEB_API_LABELS,
+)

laborious/utils/connectors_config.py

@@ -0,0 +1,127 @@
+import json
+from os import getenv
+from typing import Any
+
+
+def build_mlflow_config() -> dict[str, Any]:
+    """
+    Read MLflow tracking and registry credentials from the environment.
+
+    Used by ``Activities`` when constructing ``SientiaMLflowRepository``. The ``url`` value is the
+    same string workers and notebooks should use for ``MLFLOW_TRACKING_URI``-style clients.
+
+    Environment Variables:
+        MLFLOW_URL: Host with scheme
+        MLFLOW_USERNAME: Basic-auth or service user (default: aignosi)
+        MLFLOW_PASSWORD: Password or token (default: aignosi)
+
+    Return:
+        dict[str, Any]: ``url``, ``username``, ``password``.
+    """
+    return {
+        'url': getenv('MLFLOW_URL', 'http://localhost:5080'),
+        'username': getenv('MLFLOW_USERNAME', 'aignosi'),
+        'password': getenv('MLFLOW_PASSWORD', 'aignosi'),
+    }
+
+
+def build_plugin_store_config() -> dict[str, Any]:
+    """
+    Collect settings for ``PluginStore`` (Git-backed catalog + runtime install via pip).
+
+    Mirrors the model-manager service: the worker passes these kwargs into ``PluginStore`` after
+    ``install_runtime`` resolves wheels from the configured PyPI index. Missing optional env vars
+    become ``None`` so the store can run without auth in local dev.
+
+    Environment Variables:
+        STORE_BASE_URL: Git HTTP(S) server (e.g. Gitea) base URL (default: http://localhost:3000)
+        STORE_OWNER: Namespace or org owning the store repo (default: sientia)
+        STORE_REPO: Repository name (default: model-library-store)
+        STORE_BRANCH: Checkout branch; unset lets the client use default
+        STORE_USERNAME / STORE_PASSWORD: HTTP basic credentials for Git fetch
+        STORE_CACHE_TTL_SECONDS: Optional integer seconds for metadata cache TTL
+        PYPI_SERVER: Index URL for ``pip install`` during runtime install (default: http://localhost:5000)
+        PYPI_USERNAME / PYPI_PASSWORD: Optional index authentication
+
+    Return:
+        dict[str, Any]: Keys aligned with ``PluginStore`` constructor parameter names.
+    """
+
+    cache_ttl_seconds = getenv('STORE_CACHE_TTL_SECONDS')
+    return {
+        'base_url': getenv('STORE_BASE_URL', 'http://localhost:3000'),
+        'owner': getenv('STORE_OWNER', 'sientia'),
+        'repo': getenv('STORE_REPO', 'model-library-store'),
+        'username': getenv('STORE_USERNAME'),
+        'password': getenv('STORE_PASSWORD'),
+        'branch': getenv('STORE_BRANCH'),
+        'cache_ttl_seconds': int(cache_ttl_seconds) if cache_ttl_seconds else None,
+        'pypi_index_url': getenv('PYPI_SERVER', 'http://localhost:5000'),
+        'pypi_username': getenv('PYPI_USERNAME'),
+        'pypi_password': getenv('PYPI_PASSWORD'),
+    }
+
+
+def build_opc_config() -> dict[str, Any]:
+    """
+    Build OPC server configuration from environment variables.
+
+    This function constructs an OPC server configuration dictionary from
+    environment variables. It supports both single server and multi-server
+    configurations with flexible parameter handling.
+
+    Environment Variables:
+        OPC_CONFIG: JSON string containing multiple OPC server configurations
+        OPC_ID: OPC server ID (fallback, default: 1)
+        OPC_URL: Single OPC server URL (fallback, default: opc.tcp://localhost:4840)
+        OPC_SERVER_URI: Single OPC server URI (fallback, default: opc.tcp://localhost:4840)
+        OPC_CERT_PATH: Client certificate path (fallback, default: None)
+        OPC_PRIVATE_KEY_PATH: Client private key path (fallback, default: None)
+        OPC_SERVER_CERT_PATH: Server certificate path (fallback, default: None)
+        OPC_RECONNECTION_INTERVAL: Reconnection interval in seconds (fallback, default: 120)
+
+    Returns:
+        dict: OPC server configuration dictionary
+    """
+    opc_raw = getenv('OPC_CONFIG', None)
+
+    if opc_raw:
+        return json.loads(opc_raw)
+
+    return {
+        getenv('OPC_ID', '1'): {
+            'id': getenv('OPC_ID', '1'),
+            'server_name': getenv('OPC_SERVER_NAME', 'default_server'),
+            'url': getenv('OPC_URL', 'opc.tcp://localhost:4840'),
+            'server_uri': getenv('OPC_SERVER_URI', 'opc.tcp://localhost:4840'),
+            'cert_path': getenv('OPC_CERT_PATH', None),
+            'private_key_path': getenv('OPC_PRIVATE_KEY_PATH', None),
+            'server_cert_path': getenv('OPC_SERVER_CERT_PATH', None),
+            'reconnection_interval': int(getenv('OPC_RECONNECTION_INTERVAL', '120')),
+        }
+    }
+
+
+def build_minio_config() -> dict[str, Any]:
+    """
+    Build MinIO (S3-compatible) configuration from environment variables.
+
+    Environment Variables:
+        MINIO_ENDPOINT_URL: Host:port or URL for the S3 API (default: http://localhost:9000)
+        MINIO_ACCESS_KEY: Access key (default: minioadmin)
+        MINIO_SECRET_KEY: Secret key (default: minioadmin)
+        MINIO_DEFAULT_BUCKET: Default bucket for Laborious payloads (default: laborious)
+        MINIO_RETENTION_HOURS: Offloaded object retention window (default: 24)
+        MINIO_SECURE: If ``true``, use HTTPS (default: false)
+
+    Return:
+        dict[str, Any]: Keys consumed by ``Activities`` / ``MinioRepository``.
+    """
+    return {
+        'endpoint_url': getenv('MINIO_ENDPOINT_URL', 'http://localhost:9000'),
+        'access_key': getenv('MINIO_ACCESS_KEY', 'minioadmin'),
+        'secret_key': getenv('MINIO_SECRET_KEY', 'minioadmin'),
+        'default_bucket': getenv('MINIO_DEFAULT_BUCKET', 'laborious'),
+        'retention_hours': int(getenv('MINIO_RETENTION_HOURS', '24')),
+        'secure': getenv('MINIO_SECURE', 'false') == 'true',
+    }

laborious/utils/dataframe_debug.py

@@ -0,0 +1,34 @@
+from typing import Any
+
+from pandas import DataFrame
+
+DEFAULT_MAX_DEBUG_DATAFRAME_ROWS = 100
+
+
+def build_dataframe_debug_message(
+    message: str,
+    data: Any,
+    max_rows: int = DEFAULT_MAX_DEBUG_DATAFRAME_ROWS,
+) -> str:
+    """
+    Build a safe debug message for dataframe payloads
+
+    Args:
+     - message (str): Base message to identify the logged payload
+     - data (Any): Payload to evaluate for dataframe-aware logging
+     - max_rows (int): Maximum dataframe row count allowed for full payload logging
+
+    Return:
+     Formatted debug message with full dataframe content or compact summary
+    """
+    if not isinstance(data, DataFrame):
+        return f'{message} {data}'
+
+    rows = data.shape[0]
+    if rows <= max_rows:
+        return f'{message}\n{data.to_csv()}'
+
+    return (
+        f'{message} skipped because dataframe has {rows} rows '
+        f'(max: {max_rows}). Shape: {data.shape}'
+    )

laborious/utils/filters/conditional_filters.py

@@ -0,0 +1,48 @@
+from pandas import DataFrame
+
+
+def filter_specific_variables_null_values(data: DataFrame, config: dict) -> bool:
+    """
+    Filter to check if specific variables contain null values.
+
+    This function examines a DataFrame to determine if any of the specified variables
+    contain null (NaN) values. It returns True if null values are found for any of
+    the specified variables, False otherwise.
+
+    Args:
+        data (DataFrame): The pandas DataFrame to be examined. Must contain columns
+            named 'variable' and 'value'.
+        config (dict): Configuration dictionary containing the following key:
+            - variables (list): List of variable names to check for null values
+
+    Returns:
+        bool: True if any of the specified variables contain null values,
+              False if none of the specified variables contain null values.
+
+    """
+
+    if data.empty:
+        return False
+
+    return not data[data['variable'].isin(config['variables']) & data['value'].isna()].empty
+
+
+def filter_empty_data(data: DataFrame, _config: dict) -> bool:
+    """
+    Filter to check if the DataFrame is empty.
+
+    This function determines whether the provided DataFrame contains any data.
+    It's a simple utility function that can be used in conditional logic to
+    handle cases where no data is available.
+
+    Args:
+        data (DataFrame): The pandas DataFrame to be checked for emptiness.
+        _config (dict): Configuration dictionary (unused in this function).
+            The underscore prefix indicates this parameter is required for
+            interface consistency but not used in the implementation.
+
+    Returns:
+        bool: True if the DataFrame is empty (has no rows), False if it contains data.
+
+    """
+    return data.empty

laborious/utils/filters/mlflow_filters.py

@@ -0,0 +1,64 @@
+import numpy as np
+from pandas import DataFrame
+
+
+def api_error_filter(response: dict, _config: dict) -> bool:
+    """
+    Filter MLFlow API responses for error conditions.
+
+    This function analyzes MLFlow API responses to detect error conditions
+    and determine if the response should be filtered out due to quality
+    or reliability issues.
+
+
+    Args:
+        response: MLFlow API response data (dict)
+        _config: Filter configuration dictionary
+            Required keys:
+                - error_codes (list, optional): List of error codes to detect
+                - error_keywords (list, optional): List of error keywords to detect
+                - check_structure (bool, optional): Whether to validate response structure
+
+    Returns:
+        bool: True if data should be filtered (contains errors), False otherwise
+
+    """
+    if not response:
+        return True
+
+    if not response['success']:
+        return True
+
+    return False
+
+
+def nan_values_filter(predictions: DataFrame, _config: dict) -> bool:
+    """
+    Filter data for NaN (Not a Number) values.
+
+    This function detects NaN values in MLFlow prediction results and
+    determines if the data quality is sufficient for further processing
+    or export operations.
+
+    Args:
+        predictions: DataFrame containing prediction data to check for NaN values
+        _config: Filter configuration dictionary
+            Required keys:
+                - max_nan_ratio (float, optional): Maximum allowed NaN value ratio (0.0 to 1.0)
+                - max_nan_count (int, optional): Maximum allowed NaN value count
+                - check_nested (bool, optional): Whether to check nested data structures
+
+    Returns:
+        bool: True if data should be filtered (too many NaN values), False otherwise
+
+    """
+    data = (
+        predictions.replace({None: np.nan})
+        .drop(columns=['timestamp'], errors='ignore')
+        .infer_objects()
+    )
+
+    if data.isna().all().all():
+        return True
+
+    return False

laborious/utils/models/minio_dataframe_payload.py

@@ -0,0 +1,362 @@
+"""
+MinIO-backed DataFrame payload for Temporal workflows.
+
+Data is never stored as a pandas ``DataFrame`` field on the dataclass.
+Instead, the DataFrame is only provided as an input to:
+`from_dataframe` / `from_dataframe_to_dict`.
+
+At build time, the DataFrame is evaluated for its serialized size; if it exceeds
+the configured threshold, it is serialized to parquet bytes and uploaded to MinIO.
+Otherwise, it is inlined as a Temporal-friendly ``dict``.
+"""
+
+import pickle
+import re
+from collections.abc import Hashable
+from dataclasses import dataclass
+from datetime import datetime
+from io import BytesIO
+from os import getenv
+from typing import Any, Literal
+
+from pandas import DataFrame, read_parquet
+from sientia_do.observability.logger import Logger
+from sientia_do.repository.minio_repository_sync import MinioRepository
+from sientia_do.temporal.constants import DATETIME_FORMAT_FILENAME, DATETIME_FORMAT_WITH_TZ, now
+
+# Keys that are part of the serialized wire format (not arbitrary metadata).
+_SERIALIZED_FIELD_KEYS = frozenset({'data', 'bucket', 'object_key', 'object_prefix', 'uri'})
+
+_OBJECT_TIMESTAMP_PATTERN = re.compile(
+    r'-(?:initial|transform)-(\d{4}-\d{2}-\d{2}_\d{2}-\d{2}-\d{2})\.parquet$'
+)
+
+OFFLOAD_THRESHOLD_BYTES = int(
+    float(getenv('SIENTIA_MINIO_OFFLOAD_THRESHOLD_MEGABYTES', '1.5')) * 1024 * 1024
+)
+
+# Relative prefix used for storing offloaded prediction datasets in MinIO.
+# It is also the root directory for retention cleanup listing.
+PREDICTION_DATASETS_PREFIX = 'prediction_datasets'
+
+OperationKind = Literal['initial', 'transform', 'predict']
+
+
+def _build_object_key(
+    model_name: str, operation: OperationKind, timestamp: str
+) -> tuple[str, str | None]:
+    """
+    Build the MinIO object key and the directory prefix used for retention listing.
+
+    Args:
+        model_name: Registered model name used in the pipeline.
+        operation: Either initial (pre-transform load) or transform (post-MLFlow transform).
+        timestamp: Filename timestamp segment from DATETIME_FORMAT_FILENAME.
+
+    Return:
+        tuple[str, str | None]: Full object key and normalized prefix (or None if at bucket root).
+    """
+    # Naming convention:
+    #   - Directory is always `prediction_datasets/<model_name>`
+    #   - Filename follows the retention-parsing pattern
+    basename = f'{model_name}-{operation}-{timestamp}.parquet'
+    model_dir = model_name.strip().strip('/')
+    prefix = f'{PREDICTION_DATASETS_PREFIX}/{model_dir}'
+    return f'{prefix}/{basename}', prefix
+
+
+@dataclass
+class MinioDataFramePayload:
+    """
+    Serializable payload after a DataFrame was evaluated: inline tabular dict and/or MinIO keys.
+
+    Build from a live DataFrame only via `from_dataframe` / `from_dataframe_to_dict`.
+    Rehydrate from Temporal via `from_dict`. The DataFrame is not a field on this class.
+    """
+
+    last_timestamp: str
+    status: dict[str, Any] | None = None
+    data: dict[Hashable, Any] | None = None
+    bucket: str | None = None
+    object_key: str | None = None
+    object_prefix: str | None = None
+    uri: str | None = None
+
+    @staticmethod
+    def _debug(
+        logger: Logger | None,
+        message: str,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        """
+        Emit a debug message only when a logger instance is available.
+
+        Args:
+         - logger (Logger | None): Logger instance used for debug messages
+         - message (str): Message to be logged
+         - metadata (dict[str, Any] | None): Optional workflow metadata context
+        """
+        if logger is None:
+            return
+        logger.custom_debug(message, metadata)
+
+    @classmethod
+    def from_dict(cls, raw: 'dict[str, Any] | MinioDataFramePayload') -> 'MinioDataFramePayload':
+        """
+        Reconstruct a MinioDataFramePayload from a plain dict produced by Temporal serialization.
+
+        Temporal converts dataclass return values into plain dicts when crossing
+        workflow/activity boundaries. This method rebuilds the typed instance so
+        that methods like ``retrieve``, ``cleanup_prefix`` and ``has_data`` are
+        available on the receiving side.
+
+        If the argument is already a MinioDataFramePayload, it is returned as-is.
+
+        Args:
+            raw: Dict with keys matching the dataclass fields
+                (last_timestamp, status, data, bucket, object_key, object_prefix, uri),
+                or an existing MinioDataFramePayload instance.
+
+        Return:
+            MinioDataFramePayload: Reconstructed (or original) instance.
+        """
+        if isinstance(raw, MinioDataFramePayload):
+            return raw
+        return cls(
+            last_timestamp=raw['last_timestamp'],
+            status=raw.get('status'),
+            data=raw.get('data'),
+            bucket=raw.get('bucket'),
+            object_key=raw.get('object_key'),
+            object_prefix=raw.get('object_prefix'),
+            uri=raw.get('uri'),
+        )
+
+    @staticmethod
+    def estimate_size_bytes(
+        df: DataFrame,
+        metadata: dict[str, Any] | None = None,
+        logger: Logger | None = None,
+    ) -> int:
+        """
+        Approximate serialized size of the DataFrame as the default-orient dict.
+
+        Args:
+            df: DataFrame whose tabular content size is estimated.
+
+        Return:
+            int: Estimated size in bytes (pickle of dict representation).
+        """
+        try:
+            size = len(pickle.dumps(df.to_dict()))
+        except Exception:
+            size = len(pickle.dumps(df))
+
+        MinioDataFramePayload._debug(
+            logger,
+            f'DataFrame size: {size} bytes',
+            metadata,
+        )
+        return size
+
+    @staticmethod
+    def parse_object_timestamp(object_key: str) -> datetime | None:
+        """
+        Parse the timestamp embedded in the object key basename (before .parquet).
+
+        Args:
+            object_key: S3/MinIO object key whose basename follows
+                ``{model}-{initial|transform}-{DATETIME_FORMAT_FILENAME}.parquet``.
+
+        Return:
+            datetime | None: Parsed UTC-naive datetime from the key, or None if not matched.
+        """
+        basename = object_key.rsplit('/', 1)[-1]
+        match = _OBJECT_TIMESTAMP_PATTERN.search(basename)
+        if not match:
+            return None
+        try:
+            return datetime.strptime(match.group(1), DATETIME_FORMAT_FILENAME)
+        except ValueError:
+            return None
+
+    def cleanup_prefix(self) -> str | None:
+        """
+        Return the MinIO prefix eligible for retention cleanup.
+
+        Cleanup is only applicable when payload data was offloaded to MinIO
+        (``object_key`` present and inline ``data`` absent). Inline-only payloads
+        return ``None`` because there is no object tree to prune.
+
+        Return:
+         str | None: Prefix used by cleanup listing, or ``None`` when cleanup does not apply.
+        """
+        if self.object_key is not None and self.data is None:
+            return self.object_prefix
+        return None
+
+    def has_data(self) -> bool:
+        """
+        Indicate whether the payload contains retrievable tabular content.
+
+        A payload is considered non-empty when either inline ``data`` exists
+        (and is not an empty dict) or an ``object_key`` is available for MinIO
+        download.
+
+        Return:
+         bool: ``True`` when data can be retrieved, ``False`` otherwise.
+        """
+        return (self.data is not None and self.data != {}) or self.object_key is not None
+
+    @classmethod
+    def from_dataframe(
+        cls,
+        dataframe: DataFrame | None,
+        minio_repo: MinioRepository,
+        model_name: str,
+        operation: OperationKind,
+        status: dict[str, Any] | None = None,
+        workflow_metadata: dict | None = None,
+        last_timestamp: str | None = None,
+        logger: Logger | None = None,
+    ) -> 'MinioDataFramePayload':
+        """
+        Evaluate the DataFrame size, then either inline dict or upload parquet to MinIO.
+
+        The DataFrame is not stored on the returned instance.
+
+        Args:
+            dataframe: Tabular data to evaluate and persist (inline or MinIO).
+            metadata: Small metadata dict merged into the payload (e.g. success, message).
+            minio_repo: sientia_do MinioRepository (or compatible) with `upload_file()`.
+            workflow_metadata: Metadata passed to MinIO store for logging/metrics.
+            model_name: Registered model name used in the object basename.
+            operation: Either ``initial`` (query load) or ``transform`` (post-transform).
+            key_prefix: Backward-compatible parameter (currently ignored for object naming).
+            size_threshold_bytes: Byte limit before offload. When None, the module-level
+                environment-derived default is used.
+
+        Return:
+            MinioDataFramePayload: Instance with data and/or MinIO fields set.
+        """
+
+        if dataframe is None or dataframe.empty:
+            cls._debug(
+                logger,
+                'MinioDataFramePayload.from_dataframe received empty dataframe, returning empty payload',
+                workflow_metadata,
+            )
+            return cls(
+                data=None, last_timestamp=now().strftime(DATETIME_FORMAT_WITH_TZ), status=status
+            )
+
+        if last_timestamp is None:
+            last_timestamp = max(dataframe['timestamp'].values.tolist())
+
+        dataframe_size = cls.estimate_size_bytes(dataframe, workflow_metadata, logger)
+        cls._debug(
+            logger,
+            (
+                f'MinioDataFramePayload.from_dataframe estimated size: {dataframe_size} bytes '
+                f'(threshold: {OFFLOAD_THRESHOLD_BYTES} bytes)'
+            ),
+            workflow_metadata,
+        )
+
+        if dataframe_size <= OFFLOAD_THRESHOLD_BYTES:
+            cls._debug(
+                logger,
+                'MinioDataFramePayload.from_dataframe using inline payload',
+                workflow_metadata,
+            )
+            return cls(data=dataframe.to_dict(), last_timestamp=last_timestamp, status=status)
+
+        timestamp = now().strftime(DATETIME_FORMAT_FILENAME)
+        object_key, object_prefix = _build_object_key(model_name, operation, timestamp)
+        cls._debug(
+            logger,
+            (
+                'MinioDataFramePayload.from_dataframe offloading payload to MinIO '
+                f'with key {object_key}'
+            ),
+            workflow_metadata,
+        )
+
+        # Upload using the relative object key. The upstream repository will
+        # prefix it internally under its MinIO namespace.
+        parquet_buffer = BytesIO()
+        dataframe.to_parquet(parquet_buffer, engine='pyarrow', index=True)
+        file_bytes = parquet_buffer.getvalue()
+
+        upload_result = minio_repo.upload_file(
+            file_bytes=file_bytes,
+            relative_key=object_key,
+            metadata=workflow_metadata,
+        )
+
+        bucket = minio_repo.bucket
+        object_key_full = upload_result.get('minio_object_name', object_key)
+        uri = f's3://{bucket}/{object_key_full}' if bucket else None
+        cls._debug(
+            logger,
+            f'MinioDataFramePayload.from_dataframe upload completed: {uri}',
+            workflow_metadata,
+        )
+
+        return cls(
+            data=None,
+            bucket=bucket,
+            object_key=object_key_full,
+            object_prefix=object_prefix,
+            uri=uri,
+            last_timestamp=last_timestamp,
+            status=status,
+        )
+
+    def retrieve(
+        self,
+        minio_repo: MinioRepository,
+        workflow_metadata: dict[str, Any] | None = None,
+        logger: Logger | None = None,
+    ) -> DataFrame:
+        """
+        Load parquet from MinIO when object_key is set and populate inline data.
+
+        Args:
+            minio_repo: sientia_do MinioRepository (or compatible) with download_file().
+            workflow_metadata: Metadata passed to MinIO read for logging/metrics.
+
+        Return:
+            dict[str, Any]: Flat dict with data filled (same keys as to_dict after load).
+        """
+        if self.data is not None:
+            self._debug(
+                logger,
+                'MinioDataFramePayload.retrieve using inline payload data',
+                workflow_metadata,
+            )
+            return DataFrame(self.data)
+
+        if not self.has_data():
+            self._debug(
+                logger,
+                'MinioDataFramePayload.retrieve found no payload data, returning empty dataframe',
+                workflow_metadata,
+            )
+            return DataFrame()
+
+        self._debug(
+            logger,
+            f'MinioDataFramePayload.retrieve downloading object from MinIO: {self.object_key}',
+            workflow_metadata,
+        )
+        file_bytes = minio_repo.download_file(
+            object_name=self.object_key, metadata=workflow_metadata
+        )
+        df = read_parquet(BytesIO(file_bytes))
+        self._debug(
+            logger,
+            f'MinioDataFramePayload.retrieve loaded dataframe from MinIO with shape {df.shape}',
+            workflow_metadata,
+        )
+        return df

laborious/utils/repository/opc_repository.py

@@ -0,0 +1,893 @@
+"""
+Synchronous OPC UA client repository using asyncua ``sync`` API.
+
+``asyncua.sync.Client`` runs the asyncio stack on a background thread so Temporal
+activities and other callers stay blocking while preserving the same session
+lifecycle, security policy, reconnect semantics, and write error classification
+as the async ``main`` implementation at ``fcc8920a8be4`` (async → sync/thread conversion).
+"""
+
+import json
+import threading
+import time
+import traceback
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from asyncua import ua
+from asyncua.crypto import security_policies
+from asyncua.sync import Client
+from asyncua.ua.uaerrors import UaStatusCodeError
+from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+from sientia_do.notifications.models import NotificationLevel
+from sientia_do.observability.logger import Logger
+from sientia_do.observability.metrics_controller import MetricsController
+from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+
+from laborious import metrics
+
+# Requested session and secure channel lifetime (ms) before server revision; 10 minutes.
+OPC_UA_SESSION_AND_CHANNEL_TIMEOUT_MS = 10 * 60 * 1000
+
+
+class OpcClientAlreadyExistsError(RuntimeError):
+    """Raised when _create_client is called while self.client is already set."""
+
+
+class OpcSessionAlreadyConnectedError(RuntimeError):
+    """Raised when _open_session is called while a UA session is already open."""
+
+
+class OpcClientNotInitializedError(RuntimeError):
+    """Raised when _open_session is called before _create_client."""
+
+
+RECONNECTABLE_OPC_BAD_NAMES: frozenset[str] = frozenset(
+    {
+        'BadSessionIdInvalid',
+        'BadSessionClosed',
+        'BadSessionNotActivated',
+        'BadSecureChannelIdInvalid',
+        'BadSecureChannelClosed',
+        'BadSecureChannelTokenUnknown',
+        'BadTcpSecureChannelUnknown',
+        'BadServerNotConnected',
+        'BadConnectionClosed',
+        'BadDisconnect',
+        'BadConnectionRejected',
+        'BadCommunicationError',
+        'BadRequestInterrupted',
+        'BadUnknownResponse',
+        'BadTimeout',
+        'BadRequestTimeout',
+        'BadSequenceNumberInvalid',
+        'BadSequenceNumberUnknown',
+        'BadSecurityModeInsufficient',
+        'BadRequestHeaderInvalid',
+        'BadInvalidState',
+    }
+)
+
+
+def _opc_authentication_token_str(client: Client | None) -> str:
+    """
+    Serialize the current OPC UA authentication token (session handle) for logging and metrics.
+
+    Return:
+        str: Token string, or "unknown" if unavailable.
+    """
+    if client is None:
+        return 'unknown'
+    try:
+        proto = client.aio_obj.uaclient.protocol
+        if proto is None:
+            return 'unknown'
+        tok = getattr(proto, 'authentication_token', None)
+        if tok is None:
+            return 'unknown'
+        return str(tok)
+    except Exception:
+        return 'unknown'
+
+
+def _opc_status_from_exception(exc: BaseException) -> str:
+    """
+    Resolve OPC UA status name from an exception, including chained UaStatusCodeError causes.
+
+    Args:
+        exc (BaseException): Raised error from asyncua.
+
+    Return:
+        str: Status class name or generic Python exception name.
+    """
+    current: BaseException | None = exc
+    while current is not None:
+        if isinstance(current, UaStatusCodeError):
+            return type(current).__name__
+        current = current.__cause__
+    return type(exc).__name__
+
+
+def is_reconnectable_opcua_bad(exc: BaseException) -> bool:
+    """
+    Return whether the exception is a Tier-1 OPC UA Bad* that should trigger reconnect.
+
+    Args:
+        exc (BaseException): Raised error from get_node or write_value.
+
+    Return:
+        bool: True if reconnect should be scheduled.
+    """
+    return _opc_status_from_exception(exc) in RECONNECTABLE_OPC_BAD_NAMES
+
+
+def _model_labels_from_write_metadata(metadata: dict[str, Any] | None) -> dict[str, str]:
+    """
+    Extract model_id and model_name from write metadata for Prometheus labels.
+
+    Args:
+        metadata (dict[str, Any] | None): Context passed into write_data; may omit keys.
+
+    Return:
+        dict[str, str]: Labels model_id and model_name, defaulting to "unknown".
+    """
+    if not metadata:
+        return {'model_id': 'unknown', 'model_name': 'unknown'}
+    return {
+        'model_id': str(metadata.get('model_id', 'unknown')),
+        'model_name': str(metadata.get('model_name', 'unknown')),
+    }
+
+
+data_type_map = {
+    'float': {
+        'converter': float,
+        'opc_type': ua.VariantType.Float,
+    },
+    'double': {
+        'converter': float,
+        'opc_type': ua.VariantType.Double,
+    },
+    'int': {
+        'converter': int,
+        'opc_type': ua.VariantType.Int32,
+    },
+    'bool': {
+        'converter': bool,
+        'opc_type': ua.VariantType.Boolean,
+    },
+    'str': {
+        'converter': str,
+        'opc_type': ua.VariantType.String,
+    },
+}
+
+
+class OpcRepository(SientiaMonitoring):
+    """
+    Synchronous OPC UA repository for connect/disconnect and typed writes.
+
+    Uses ``asyncua.sync.Client`` with the same session metrics, Tier-1 Bad* reconnect,
+    and structured write error payloads as the async repository on ``main``.
+    """
+
+    def __init__(
+        self,
+        opc_id: str,
+        url: str,
+        server_name: str,
+        logger: Logger,
+        notification_handler: NotificationHandler,
+        metrics_controller: MetricsController,
+        reconnection_interval: int = 60,
+        server_uri: str | None = None,
+        cert_path: str | None = None,
+        private_key_path: str | None = None,
+        server_cert_path: str | None = None,
+    ):
+        self.url = url
+        self.id = opc_id
+        self.server_name = server_name
+        self.server_uri = server_uri
+        self.cert_path = cert_path
+        self.private_key_path = private_key_path
+        self.server_cert_path = server_cert_path
+        self.reconnection_interval = reconnection_interval
+        self.last_reconnection_time: None | datetime = None
+        self.disconnection_interval = 10.0
+        self.notification_handler = notification_handler
+        self.client: None | Client = None
+
+        SientiaMonitoring.__init__(self, logger, notification_handler, metrics_controller)
+
+        self.metadata = {
+            'model_name': '-',
+            'model_id': '-',
+            'workflow_name': 'opc_repository',
+            'schedule_name': '-',
+        }
+        self._last_write_mono: float | None = None
+        self._connection_lock = threading.Lock()
+        self._session_ready = threading.Event()
+        self._reconnect_thread: threading.Thread | None = None
+        self._allow_reconnect = True
+
+    def _opc_debug_tags(self, session_id: str) -> dict[str, str]:
+        """
+        Build Prometheus/log label tags for OPC session-scoped metrics.
+
+        Args:
+            session_id (str): OPC UA session token string.
+
+        Return:
+            dict[str, str]: Labels pod_id, server_name, runtime, opc_server_id, session_id.
+        """
+        return {
+            'pod_id': str(getattr(self, 'pod_id', 'unknown')),
+            'server_name': self.server_name,
+            'runtime': str(getattr(self, 'runtime', 'unknown')),
+            'opc_server_id': self.id,
+            'session_id': session_id,
+        }
+
+    def _is_session_open(self) -> bool:
+        """
+        Return whether the asyncua client has an open transport session.
+
+        Return:
+            bool: True when protocol exists and is not closed.
+        """
+        if self.client is None:
+            return False
+        try:
+            proto = self.client.aio_obj.uaclient.protocol
+            return proto is not None and proto.state != 'closed'
+        except Exception:
+            return False
+
+    def _reconnection_window_elapsed(self) -> bool:
+        """
+        Return whether enough time has passed since the last reconnect attempt.
+
+        Return:
+            bool: True if a new reconnect is allowed.
+        """
+        if self.last_reconnection_time is None:
+            return True
+        return (
+            datetime.now() - self.last_reconnection_time
+        ).total_seconds() > self.reconnection_interval
+
+    def _not_connected_error(self) -> dict[str, Any]:
+        """
+        Build the standard error payload when validate_connection finds no open protocol.
+
+        Return:
+            dict[str, Any]: Notification fields for OPC_CONNECTION_NOT_READY.
+        """
+        return {
+            'notification_id': f'OPC_CONNECTION_NOT_READY_{self.id}',
+            'message': f'OPC server {self.id} is not connected',
+            'block': 'opc_repository',
+            'level': NotificationLevel.WARNING,
+        }
+
+    def set_security(self) -> None:
+        """
+        Configure certificates and timeouts on the sync asyncua client.
+
+        Raises:
+            ValueError: If cert paths or client are missing.
+        """
+        if self.cert_path is None or self.private_key_path is None:
+            raise ValueError(
+                'Certificate and private key paths must be provided for secure connection.'
+            )
+
+        cert = Path(self.cert_path)
+        private_key = Path(self.private_key_path)
+        server_cert = Path(self.server_cert_path) if self.server_cert_path else None
+
+        if self.client is None:
+            raise ValueError('Client must be initialized before setting security')
+
+        self.client.application_uri = self.server_uri
+        self.info('Setting security...', self.metadata)
+        self.client.set_security(
+            security_policies.SecurityPolicyBasic256,
+            str(cert),
+            str(private_key),
+            None,
+            str(server_cert) if server_cert else None,
+        )
+        aio = self.client.aio_obj
+        aio.secure_channel_timeout = OPC_UA_SESSION_AND_CHANNEL_TIMEOUT_MS
+        aio.session_timeout = OPC_UA_SESSION_AND_CHANNEL_TIMEOUT_MS
+
+    def _create_client(self) -> None:
+        """
+        Instantiate the sync Client and apply security when configured.
+
+        Caller must hold _connection_lock. Does not open a UA session.
+
+        Raises:
+            OpcClientAlreadyExistsError: If self.client is already set.
+        """
+        if self.client is not None:
+            raise OpcClientAlreadyExistsError(
+                f'OPC client already exists for server {self.id}; '
+                'call disconnect() before creating a new client'
+            )
+
+        self.client = Client(self.url, timeout=10)
+        aio = self.client.aio_obj
+        if hasattr(aio, 'watchdog_intervall'):
+            aio.watchdog_intervall = 50
+        aio.name = self.pod_id
+        aio.description = self.pod_id
+        pod_uri = self.pod_id.replace('-', ':')
+        self.client.application_uri = pod_uri
+        aio.product_uri = pod_uri
+        if self.cert_path:
+            self.set_security()
+
+    def _open_session(self) -> tuple[bool, dict[str, Any]]:
+        """
+        Open the OPC UA session on the existing client.
+
+        Caller must hold _connection_lock.
+
+        Raises:
+            OpcClientNotInitializedError: If self.client is None.
+            OpcSessionAlreadyConnectedError: If a session is already open.
+
+        Return:
+            tuple[bool, dict[str, Any]]: Success flag and error payload on connect failure.
+        """
+        if self.client is None:
+            raise OpcClientNotInitializedError(
+                f'OPC client is not initialized for server {self.id}; '
+                'call _create_client() before opening a session'
+            )
+        if self._is_session_open():
+            raise OpcSessionAlreadyConnectedError(
+                f'OPC session already connected for server {self.id}; '
+                'call disconnect() before connecting again'
+            )
+
+        tags = {
+            'pod_id': self.pod_id,
+            'server_name': self.server_name,
+        }
+        self.emit_metric_sync(metrics.OPC_CONNECTIONS_TOTAL, tags)
+
+        try:
+            self.client.connect()
+
+            aio = self.client.aio_obj
+            session_id = _opc_authentication_token_str(self.client)
+            revised_session_timeout_ms = int(aio.session_timeout)
+            revised_secure_channel_timeout_ms = int(aio.secure_channel_timeout)
+            self.info(
+                f'OPC new session connected opc_server_id={self.id} session_id={session_id} '
+                f'revised_session_timeout_ms={revised_session_timeout_ms} '
+                f'revised_secure_channel_timeout_ms={revised_secure_channel_timeout_ms}',
+                self.metadata,
+            )
+            self.emit_metric_sync(
+                metrics.OPC_SESSION_CREATED_TOTAL, self._opc_debug_tags(session_id)
+            )
+            self.emit_metric_sync(
+                metric_object=metrics.OPC_SESSION_REVISED_TIMEOUT_MS,
+                method='set',
+                tags=self._opc_debug_tags(session_id),
+                value=revised_session_timeout_ms,
+            )
+            self.emit_metric_sync(
+                metric_object=metrics.OPC_CONNECTION_STATUS,
+                method='set',
+                tags={**tags, 'server_url': self.url},
+                value=1,
+            )
+
+            self._last_write_mono = None
+            self._session_ready.set()
+            return True, {}
+
+        except Exception as e:
+            self._disconnect_locked()
+            trace = traceback.format_exc()
+            self.error(trace, self.metadata)
+            self.emit_metric_sync(metrics.OPC_CONNECTIONS_FAILED, tags)
+            return False, {
+                'notification_id': f'OPC_CONNECTION_ERROR_{self.id}',
+                'message': f'Failed to connect to OPC server: {e}',
+                'block': 'opc_repository',
+                'level': NotificationLevel.ERROR,
+                'attachment_content': trace,
+            }
+
+    def _connect_locked(self) -> tuple[bool, dict[str, Any]]:
+        """
+        Create the client when absent, then open a UA session.
+
+        Caller must hold _connection_lock.
+
+        Raises:
+            OpcSessionAlreadyConnectedError: If a session is already open.
+
+        Return:
+            tuple[bool, dict[str, Any]]: Result from _open_session on connect failure.
+        """
+        if self._is_session_open():
+            raise OpcSessionAlreadyConnectedError(
+                f'OPC session already connected for server {self.id}; '
+                'call disconnect() before connecting again'
+            )
+        if self.client is None:
+            self._create_client()
+        return self._open_session()
+
+    def _disconnection_fallback(self) -> list[dict[str, Any]]:
+        """
+        Try up to five times to disconnect from the OPC UA server.
+        """
+        assert self.client is not None
+        error_stack: list[dict[str, Any]] = []
+        for i in range(5):
+            try:
+                self.info(
+                    f'Disconnecting from OPC UA server, attempt {i + 1} of 5',
+                    self.metadata,
+                )
+                self.client.disconnect()
+                return []
+            except Exception as e:
+                self.error(
+                    f'Failed to disconnect from OPC UA server in attempt {i + 1} of 5: {e}',
+                    self.metadata,
+                )
+                error_stack.append(
+                    {
+                        'attempt': i + 1,
+                        'error': str(e),
+                        'traceback': traceback.format_exc(),
+                    }
+                )
+                time.sleep(self.disconnection_interval * i)
+        return error_stack
+
+    def _disconnect_locked(self) -> None:
+        """
+        Tear down the current session and client.
+
+        Caller must hold _connection_lock.
+        """
+        self._last_write_mono = None
+        self._session_ready.clear()
+
+        if self.client is None:
+            return
+
+        session_id = _opc_authentication_token_str(self.client)
+        self.info(
+            f'OPC disconnecting opc_server_id={self.id} session_id={session_id}',
+            self.metadata,
+        )
+        self.emit_metric_sync(metrics.OPC_SESSION_CLOSED_TOTAL, self._opc_debug_tags(session_id))
+
+        errors = self._disconnection_fallback()
+        if errors:
+            self.send_notification(
+                metadata=self.metadata,
+                notification_id=f'OPC_DISCONNECTION_ERROR_{self.id}',
+                message='Failed to disconnect from OPC server in 5 attempts.',
+                block='opc_repository',
+                level=NotificationLevel.ERROR,
+                attachment_content=json.dumps(errors, indent=4),
+            )
+        else:
+            self.warning(f'Disconnected from OPC server {self.id} successfully', self.metadata)
+
+        self.emit_metric_sync(
+            metric_object=metrics.OPC_CONNECTION_STATUS,
+            method='set',
+            tags={
+                'pod_id': self.pod_id,
+                'server_name': self.server_name,
+                'server_url': self.url,
+            },
+            value=0,
+        )
+        self.client = None
+
+    def _reconnect_locked(self) -> tuple[bool, dict[str, Any]]:
+        """
+        Close the current session and open a new one.
+
+        Caller must hold _connection_lock. Records last_reconnection_time for interval gating.
+
+        Return:
+            tuple[bool, dict[str, Any]]: Result from _connect_locked after teardown.
+        """
+        self.last_reconnection_time = datetime.now()
+        self._disconnect_locked()
+        return self._connect_locked()
+
+    def connect(self) -> tuple[bool, dict[str, Any]]:
+        """
+        Open an OPC UA session under the connection lock (worker initialization).
+        """
+        with self._connection_lock:
+            self.info(
+                f'Starting connection to OPC server {self.id}:{self.server_name}...',
+                self.metadata,
+            )
+            return self._connect_locked()
+
+    def disconnect(self) -> None:
+        """
+        Gracefully disconnect from the OPC server under the connection lock.
+
+        Disables background reconnect so late writes during worker shutdown do not
+        respawn sessions.
+        """
+        with self._connection_lock:
+            self._allow_reconnect = False
+            self._disconnect_locked()
+
+    def validate_connection(self) -> tuple[bool, dict[str, Any]]:
+        """
+        Read-only check that the asyncua protocol is open.
+
+        Caller must ensure _session_ready before writing. Does not connect or reconnect.
+
+        Return:
+            tuple[bool, dict[str, Any]]: (True, {}) when open, otherwise (False, error).
+        """
+        if self._is_session_open():
+            return True, {}
+        self.error(f'OPC server {self.id} is not connected', self.metadata)
+        return False, self._not_connected_error()
+
+    def _reconnect_thread_in_progress(self) -> bool:
+        """
+        Return whether a background reconnect thread is currently running.
+
+        Return:
+            bool: True when a reconnect thread exists and is alive.
+        """
+        return self._reconnect_thread is not None and self._reconnect_thread.is_alive()
+
+    def _start_reconnect(self, reason: str, session_id: str) -> None:
+        """
+        Schedule a background reconnect when allowed by interval and thread state.
+
+        Clears _session_ready before starting the thread. No-op when _allow_reconnect is
+        False, the reconnection window has not elapsed, or a reconnect is already running.
+
+        Args:
+            reason (str): Trigger for reconnect (OPC status name or synthetic reason).
+            session_id (str): Session token before failure.
+        """
+        if not self._allow_reconnect:
+            return
+        if not self._reconnection_window_elapsed():
+            self.warning(
+                f'OPC reconnect skipped reason=reconnection_window opc_server_id={self.id} '
+                f'reconnect_reason={reason}',
+                self.metadata,
+            )
+            return
+        if self._reconnect_thread_in_progress():
+            self.warning(
+                f'OPC reconnect skipped reason=in_progress opc_server_id={self.id} '
+                f'reconnect_reason={reason}',
+                self.metadata,
+            )
+            return
+
+        self._session_ready.clear()
+        self.info(
+            f'OPC reconnect scheduled reconnect_reason={reason} opc_server_id={self.id} '
+            f'old_session_id={session_id}',
+            self.metadata,
+        )
+        self._reconnect_thread = threading.Thread(
+            target=self._run_reconnect,
+            args=(reason, session_id),
+            daemon=True,
+        )
+        self._reconnect_thread.start()
+
+    def _run_reconnect(self, reason: str, session_id: str) -> None:
+        """
+        Tear down and re-establish the OPC UA session under the connection lock.
+
+        Args:
+            reason (str): Trigger for reconnect (OPC status or ProtocolClosed).
+            session_id (str): Previous session token string.
+        """
+        try:
+            with self._connection_lock:
+                self.info(
+                    f'OPC reconnect started reconnect_reason={reason} opc_server_id={self.id} '
+                    f'old_session_id={session_id}',
+                    self.metadata,
+                )
+                success, error = self._reconnect_locked()
+            if not success:
+                self.error(
+                    f'OPC reconnect failed reconnect_reason={reason} opc_server_id={self.id}',
+                    self.metadata,
+                )
+                if error:
+                    self.error(error.get('message', ''), self.metadata)
+        except Exception:
+            self.error(
+                f'OPC reconnect task failed opc_server_id={self.id} reconnect_reason={reason}',
+                self.metadata,
+            )
+            self.error(traceback.format_exc(), self.metadata)
+
+    def _log_write_inter_arrival(self, session_id: str, node: str) -> None:
+        """
+        Log elapsed wall time since the previous successful OPC write on this repository.
+
+        Args:
+            session_id (str): Current OPC UA session token string.
+            node (str): Node id written in this operation.
+        """
+        now = time.monotonic()
+        if self._last_write_mono is not None:
+            delta_s = now - self._last_write_mono
+            self.info(
+                f'OPC write inter-arrival_s={delta_s:.6f} opc_server_id={self.id} '
+                f'session_id={session_id} node={node}',
+                self.metadata,
+            )
+            if self.client is not None:
+                session_timeout_ms = float(self.client.aio_obj.session_timeout)
+                if session_timeout_ms > 0 and delta_s > (session_timeout_ms / 1000.0):
+                    self.emit_metric_sync(
+                        metrics.OPC_WRITE_INTER_ARRIVAL_OVER_SESSION_TIMEOUT_TOTAL,
+                        self._opc_debug_tags(session_id),
+                    )
+        self._last_write_mono = now
+
+    def _emit_opc_write_metric(
+        self, session_id: str, result: str, metadata: dict[str, Any] | None
+    ) -> None:
+        self.emit_metric_sync(
+            metrics.OPC_WRITE_ATTEMPTS_TOTAL,
+            {
+                **self._opc_debug_tags(session_id),
+                **_model_labels_from_write_metadata(metadata),
+                'result': result,
+            },
+        )
+
+    def _write_failure_payload(
+        self,
+        notification_id: str,
+        message: str,
+        level: NotificationLevel = NotificationLevel.ERROR,
+        attachment_content: str | None = None,
+        opc_error_kind: str | None = None,
+        opc_status: str | None = None,
+    ) -> dict[str, Any]:
+        payload: dict[str, Any] = {
+            'notification_id': notification_id,
+            'message': message,
+            'block': 'opc_repository',
+            'level': level,
+        }
+        if attachment_content is not None:
+            payload['attachment_content'] = attachment_content
+        if opc_error_kind is not None:
+            payload['opc_error_kind'] = opc_error_kind
+        if opc_status is not None:
+            payload['opc_status'] = opc_status
+        return payload
+
+    def _handle_tier1_bad(
+        self,
+        exc: BaseException,
+        session_id: str,
+        node: str,
+        metadata: dict[str, Any],
+        phase: str,
+    ) -> tuple[bool, dict[str, Any]]:
+        """
+        Record metrics/logs and schedule reconnect after a Tier-1 Bad* error.
+
+        Args:
+            exc (BaseException): Tier-1 OPC UA error.
+            session_id (str): Session token at failure time.
+            node (str): Node id being written.
+            metadata (dict[str, Any]): Write context.
+            phase (str): get_node or write_value.
+
+        Return:
+            tuple[bool, dict[str, Any]]: Always (False, error payload).
+        """
+        opc_status = _opc_status_from_exception(exc)
+        trace = traceback.format_exc()
+        self.error(trace, metadata)
+        self._emit_opc_write_metric(session_id, opc_status, metadata)
+        self.error(
+            f'OPC write failed opc_status={opc_status} opc_server_id={self.id} '
+            f'session_id={session_id} model_id={metadata.get("model_id", "unknown")} '
+            f'model_name={metadata.get("model_name", "unknown")} node={node} phase={phase}',
+            metadata,
+        )
+        self._start_reconnect(opc_status, session_id)
+        return False, self._write_failure_payload(
+            notification_id=f'OPC_WRITE_DATA_ERROR_{self.id}',
+            message=f'Failed to {phase} on OPC server: {exc} | metadata: {metadata}',
+            attachment_content=trace,
+            opc_error_kind='session_bad',
+            opc_status=opc_status,
+        )
+
+    @staticmethod
+    def _write_node_value(
+        node_obj: Any,
+        ua_data: ua.DataValue,
+        data: Any,
+        variant_type: ua.VariantType,
+    ) -> None:
+        """
+        Write a DataValue to a node, falling back to set_value when write_value is unavailable.
+
+        Args:
+            node_obj: Sync or async node wrapper from asyncua.
+            ua_data (ua.DataValue): Encoded value for write_value.
+            data: Scalar converted value for set_value fallback.
+            variant_type (ua.VariantType): OPC UA type for set_value fallback.
+        """
+        if hasattr(node_obj, 'write_value'):
+            try:
+                node_obj.write_value(ua_data)
+                return
+            except (AttributeError, TypeError):
+                pass
+        node_obj.set_value(data, variant_type)
+
+    def _write_reconnect_in_progress(self, metadata: dict[str, Any]) -> tuple[bool, dict[str, Any]]:
+        """
+        Fail a write because a background reconnect thread is already running.
+
+        Args:
+            metadata (dict[str, Any]): Write context passed through to the activity.
+
+        Return:
+            tuple[bool, dict[str, Any]]: (False, error info with opc_error_kind reconnect_in_progress).
+        """
+        self._emit_opc_write_metric('unknown', 'ReconnectInProgress', metadata)
+        self.warning(
+            f'OPC write rejected reconnect_in_progress opc_server_id={self.id} '
+            f'model_id={metadata.get("model_id", "unknown")} '
+            f'model_name={metadata.get("model_name", "unknown")}',
+            metadata,
+        )
+        return False, {
+            'notification_id': f'OPC_WRITE_RECONNECT_IN_PROGRESS_{self.id}',
+            'message': f'OPC write skipped: reconnect in progress | metadata: {metadata}',
+            'block': 'opc_repository',
+            'level': NotificationLevel.WARNING,
+            'opc_error_kind': 'reconnect_in_progress',
+        }
+
+    def _write_connection_lost(
+        self, metadata: dict[str, Any], opc_status: str
+    ) -> tuple[bool, dict[str, Any]]:
+        """
+        Fail a write after scheduling reconnect for a closed or stale session.
+
+        Args:
+            metadata (dict[str, Any]): Write context passed through to the activity.
+            opc_status (str): Synthetic reason (ProtocolClosed, SessionNotReady).
+
+        Return:
+            tuple[bool, dict[str, Any]]: (False, error info with opc_error_kind connection_lost).
+        """
+        self._emit_opc_write_metric('unknown', opc_status, metadata)
+        return False, self._write_failure_payload(
+            notification_id=f'OPC_WRITE_CONNECTION_LOST_{self.id}',
+            message=f'OPC write skipped: connection lost ({opc_status}) | metadata: {metadata}',
+            level=NotificationLevel.WARNING,
+            opc_error_kind='connection_lost',
+            opc_status=opc_status,
+        )
+
+    def write_data(
+        self, node: str, value: Any, data_type: str, metadata: dict[str, Any]
+    ) -> tuple[bool, dict[str, Any]]:
+        """
+        Write data to OPC server with a single attempt and background reconnect scheduling.
+
+        Reconnect is scheduled on Tier-1 Bad*, closed protocol, or stale session readiness.
+        There is no retry within the same call.
+
+        Args:
+            node (str): OPC UA node id to write.
+            value (Any): Value to convert and send.
+            data_type (str): Logical type key (float, int, bool, str, double).
+            metadata (dict[str, Any]): Activity context (model_id, model_name, etc.).
+
+        Return:
+            tuple[bool, dict[str, Any]]: (True, {response_time}) on success, or
+                (False, structured error info) on failure.
+        """
+        if self._reconnect_thread_in_progress():
+            return self._write_reconnect_in_progress(metadata)
+
+        if not self._session_ready.is_set():
+            session_id = _opc_authentication_token_str(self.client)
+            self._start_reconnect('SessionNotReady', session_id)
+            if self._reconnect_thread_in_progress():
+                return self._write_reconnect_in_progress(metadata)
+            return self._write_connection_lost(metadata, 'SessionNotReady')
+
+        is_connected, _error = self.validate_connection()
+        if not is_connected:
+            session_id = _opc_authentication_token_str(self.client)
+            self._start_reconnect('ProtocolClosed', session_id)
+            return self._write_connection_lost(metadata, 'ProtocolClosed')
+
+        start_time = time.time()
+        session_id = _opc_authentication_token_str(self.client)
+
+        try:
+            assert self.client is not None
+            node_obj = self.client.get_node(node)
+        except Exception as e:
+            if is_reconnectable_opcua_bad(e):
+                return self._handle_tier1_bad(e, session_id, node, metadata, 'get_node')
+            trace = traceback.format_exc()
+            self.error(trace, metadata)
+            self._emit_opc_write_metric(session_id, f'GetNodeError:{type(e).__name__}', metadata)
+            return False, self._write_failure_payload(
+                notification_id=f'OPC_WRITE_GET_NODE_ERROR_{self.id}',
+                message=f'Failed to get node from OPC server: {e} | metadata: {metadata}',
+                attachment_content=trace,
+            )
+
+        if data_type not in data_type_map:
+            self._emit_opc_write_metric(session_id, 'UnsupportedDataType', metadata)
+            return False, self._write_failure_payload(
+                notification_id=f'OPC_WRITE_DATA_TYPE_ERROR_{self.id}',
+                message=f'Unsupported data type: {data_type} | metadata: {metadata}',
+            )
+
+        data = data_type_map[data_type]['converter'](value)
+        self.info(f'Writing {data} - {type(data)} to {node}', metadata)
+        variant_type = data_type_map[data_type]['opc_type']
+        ua_data = ua.DataValue(
+            ua.Variant(data, variant_type),
+        )
+
+        try:
+            self._write_node_value(node_obj, ua_data, data, variant_type)
+            end_time = time.time()
+            response_time = end_time - start_time
+        except Exception as e:
+            if is_reconnectable_opcua_bad(e):
+                return self._handle_tier1_bad(e, session_id, node, metadata, 'write_value')
+            trace = traceback.format_exc()
+            self.error(trace, metadata)
+            self._emit_opc_write_metric(session_id, type(e).__name__, metadata)
+            return False, self._write_failure_payload(
+                notification_id=f'OPC_WRITE_DATA_ERROR_{self.id}',
+                message=f'Failed to write data to OPC server: {e} | metadata: {metadata}',
+                attachment_content=trace,
+            )
+
+        self._emit_opc_write_metric(session_id, 'OK', metadata)
+        self._log_write_inter_arrival(session_id, node)
+
+        return True, {
+            'response_time': response_time,
+        }

laborious/worker/worker.py

@@ -0,0 +1,313 @@
+"""
+Laborious Worker Module
+
+Entry process that connects to Temporal, registers Laborious activities, and runs four workers in
+parallel. Each worker shares the same ``Activities`` instance (single Postgres pool, single MLflow
+repository, single PluginStore handle) but polls a different task queue.
+
+Task queues (see ``sientia_do.temporal.worker.prepare_worker``):
+- ``predictions_batch-{runtime}-queue`` + sub-workflows on the same queue (ML-heavy path).
+- ``minimal_retrain-{runtime}-queue`` (retrain + promote + export).
+- ``drift-queue`` and ``simple_metrics-queue`` without a runtime suffix so existing schedulers
+  keep stable queue names.
+
+Bootstrap order:
+1. Prometheus app metrics and Mongo-backed notification handler.
+2. ``RUNTIME`` validation and ``PluginStore.install_runtime`` so ``SientiaModel`` code is importable.
+3. ``Activities`` construction (builds ``SientiaMLflowRepository`` internally from env).
+4. OPC client initialization inside activities.
+5. Temporal ``Runtime`` with SDK Prometheus bind, client connect, then ``prepare_worker`` per workflow.
+
+Shutdown closes workers, notifications, activities (pools + OPC), and clears ``app_up``.
+
+Environment Variables:
+- RUNTIME: Required non-empty string passed to ``install_runtime``.
+- STORE_* / PYPI_*: Plugin store and private index (see ``build_plugin_store_config``).
+- TEMPORAL_HOST, TEMPORAL_NAMESPACE: Cluster connection.
+- POD_ID, HTTP_METRICS_PORT, HTTP_SDK_METRICS_PORT: Observability.
+- PROJECT_NAME, MONGODB_*: Notifications (via ``build_mongodb_config`` in handler).
+- POSTGRES_*, MINIO_*, OPC_*, PI_WEB_API_*, MLFLOW_*: Passed through ``Activities`` helpers.
+"""
+
+from temporalio import client, workflow
+from temporalio.runtime import PrometheusConfig, Runtime, TelemetryConfig
+
+with workflow.unsafe.imports_passed_through():
+    import asyncio
+    import os
+    import sys
+
+    from prometheus_client import start_http_server
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    from sientia_do.observability.logger import get_logger
+    from sientia_do.observability.metrics_controller import MetricsController
+    from sientia_do.temporal.worker.prepare_worker import prepare_worker
+    from sientia_do.utils.connectors_config import (
+        build_api_config,
+        build_mongodb_config,
+        build_postgres_config,
+    )
+    from sientia_model.model_repository.plugin_store import PluginStore
+
+    from laborious import metrics
+    from laborious.activities.activities import Activities
+    from laborious.utils.connectors_config import (
+        build_minio_config,
+        build_opc_config,
+        build_plugin_store_config,
+    )
+    from laborious.workflows.drift import Drift
+    from laborious.workflows.minimal_retrain import MinimalRetrain
+    from laborious.workflows.predictions_batch import PredictionsBatch
+    from laborious.workflows.simple_metrics import SimpleMetrics
+    from laborious.workflows.sub_workflows.format_and_export_prediction import (
+        FormatAndExportPrediction,
+    )
+    from laborious.workflows.sub_workflows.prediction_process import PredictionProcess
+
+POD_ID = os.getenv('HOSTNAME')
+SDK_METRICS_PORT = int(os.getenv('HTTP_SDK_METRICS_PORT', '9091'))
+
+
+async def main():
+    """
+    Run the full worker lifecycle: metrics, notifications, runtime install, workers, gather.
+
+    Exits the process with code 0 on normal completion of all worker tasks, or 1 after logging
+    if any worker raises. ``finally`` always shuts down notifications and activities and sets
+    ``app_up`` to 0 before ``sys.exit``.
+
+    Raises:
+        Exception: Propagated from ``asyncio.gather`` only before ``finally`` handling; typically
+            workers run until cancelled.
+
+    Return:
+        None (process terminates via ``sys.exit`` from the ``finally`` block).
+    """
+    host = os.getenv('TEMPORAL_HOST', 'localhost:7233')
+    logger = get_logger(__name__)
+
+    metadata = {
+        'pod_id': POD_ID,
+        'model_name': '-',
+        'model_id': '-',
+        'workflow_name': '-',
+        'schedule_name': '-',
+    }
+
+    logger.custom_info(f'Starting Worker with POD_ID: {POD_ID}', metadata)
+
+    logger.custom_info('Starting prometheus client...', metadata)
+    start_prometheus_server()
+
+    logger.custom_info('Starting Notification Handler...', metadata)
+
+    mongo_config = build_mongodb_config()
+    notification_handler = NotificationHandler(
+        connection_string=mongo_config['connection_string'],
+        database=mongo_config['database_name'],
+        logger=logger,
+        project_name=os.getenv('PROJECT_NAME', 'laborious'),
+    )
+
+    metrics_controller = MetricsController(logger=logger)
+
+    runtime = os.getenv('RUNTIME', '').strip()
+    if not runtime:
+        logger.custom_critical(
+            'RUNTIME environment variable is required and must be non-empty',
+            metadata,
+        )
+        metrics.APP_UP.labels(pod_id=POD_ID).set(0)
+        sys.exit(1)
+
+    metadata_runtime = {**metadata, 'runtime': runtime}
+    logger.custom_info(f'Installing PluginStore runtime: {runtime}', metadata_runtime)
+
+    ps_cfg = build_plugin_store_config()
+    plugin_store = PluginStore(
+        base_url=ps_cfg['base_url'],
+        owner=ps_cfg['owner'],
+        repo=ps_cfg['repo'],
+        username=ps_cfg['username'],
+        password=ps_cfg['password'],
+        branch=ps_cfg['branch'],
+        cache_ttl_seconds=ps_cfg['cache_ttl_seconds'],
+        pypi_index_url=ps_cfg['pypi_index_url'],
+        pypi_username=ps_cfg['pypi_username'],
+        pypi_password=ps_cfg['pypi_password'],
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=metrics_controller,
+    )
+
+    if runtime == 'legacy':
+        to_install_runtime = 'single'
+    else:
+        to_install_runtime = runtime
+
+    try:
+        await plugin_store.install_runtime(
+            runtime_name=to_install_runtime, metadata=metadata_runtime
+        )
+    except Exception as exc:
+        logger.custom_critical(
+            f'Failed to install runtime {to_install_runtime}: {exc}', metadata_runtime
+        )
+        metrics.APP_UP.labels(pod_id=POD_ID).set(0)
+        sys.exit(1)
+
+    logger.custom_info('Starting Activities...', metadata)
+
+    activities = Activities(
+        postgres_config=build_postgres_config(),
+        plugin_store=plugin_store,
+        minio_config=build_minio_config(),
+        opc_config=build_opc_config(),
+        pi_web_api_config=build_api_config(),
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=metrics_controller,
+    )
+
+    logger.custom_info('Initializing OPC...', metadata)
+    activities.init_opc()
+
+    logger.custom_info(f'Starting SDK Metrics Server on port {SDK_METRICS_PORT}...', metadata)
+
+    new_runtime = Runtime(
+        telemetry=TelemetryConfig(
+            metrics=PrometheusConfig(bind_address=f'0.0.0.0:{SDK_METRICS_PORT}')
+        )
+    )
+
+    logger.custom_info(f'Starting Temporal Client at {host}...', metadata)
+
+    temporal_client = await client.Client.connect(
+        target_host=host,
+        namespace=os.getenv('TEMPORAL_NAMESPACE', 'laborious'),
+        runtime=new_runtime,
+    )
+
+    logger.custom_info('Starting Workers...', metadata)
+
+    workers = [
+        prepare_worker(
+            temporal_client=temporal_client,
+            main_workflow=MinimalRetrain,
+            other_workflows=[],
+            activities=[
+                activities.load_query_with_minio_offload,
+                activities.retrain_model,
+                activities.update_production_model,
+                activities.format_retrain_report,
+                activities.export_data_to_postgres,
+            ],
+            logger=logger,
+            runtime=runtime,
+        ),
+        prepare_worker(
+            temporal_client=temporal_client,
+            main_workflow=SimpleMetrics,
+            other_workflows=[],
+            activities=[
+                activities.load_custom_query,
+                activities.calculate_simple_metrics,
+                activities.export_data_to_postgres,
+            ],
+            logger=logger,
+            runtime='core',
+        ),
+        prepare_worker(
+            temporal_client=temporal_client,
+            main_workflow=Drift,
+            other_workflows=[],
+            activities=[
+                activities.load_custom_query,
+                activities.get_reference_data,
+                activities.calculate_drift,
+                activities.export_data_to_postgres,
+            ],
+            logger=logger,
+            runtime='core',
+        ),
+        prepare_worker(
+            temporal_client=temporal_client,
+            main_workflow=PredictionsBatch,
+            other_workflows=[PredictionProcess, FormatAndExportPrediction],
+            activities=[
+                # MLFlow
+                activities.request_predict,
+                activities.request_transform,
+                # Gates
+                activities.input_gate,
+                activities.mlflow_response_gate,
+                activities.mlflow_content_gate,
+                activities.format_transformed_data,
+                activities.format_prediction,
+                activities.format_default_prediction,
+                # OPC
+                activities.write_opc_data,
+                # Postgres / MinIO offload
+                activities.load_query_with_minio_offload,
+                activities.cleanup_minio_objects_expired,
+                activities.repeat_last_prediction,
+                activities.export_data_to_postgres,
+                activities.export_payload_to_postgres,
+                activities.write_metrics,
+                # Pi Web API
+                activities.write_pi_web_api_data,
+            ],
+            logger=logger,
+            runtime=runtime,
+        ),
+    ]
+
+    handlers = []
+    for w in workers:
+        handlers.append(w.run())
+
+    logger.custom_info('Workers started successfully', metadata)
+
+    exit_code = 0
+    try:
+        await asyncio.gather(*handlers)
+    except BaseException as e:  # NOSONAR
+        logger.custom_error(f'An unhandled exception occurred: {e}', metadata)
+        exit_code = 1
+    finally:
+        notification_handler.shutdown()
+        activities.shutdown()
+        metrics.APP_UP.labels(pod_id=POD_ID).set(0)  # Mark app as DOWN
+        sys.exit(exit_code)
+
+
+def start_prometheus_server():
+    """
+    Starts the Prometheus metrics server for monitoring and observability.
+
+    This function initializes the Prometheus HTTP server on the configured port
+    and sets the application health metric to indicate the service is running.
+
+    The server exposes metrics that can be scraped by Prometheus for monitoring
+    the health and performance of the Laborious worker.
+
+    Environment Variables:
+        HTTP_METRICS_PORT: Port for the metrics server (default: 9090)
+        POD_ID: Pod identifier for metrics labeling
+
+    Raises:
+        SystemExit: If the metrics server fails to start
+    """
+    try:
+        port = int(os.getenv('HTTP_METRICS_PORT', 9090))
+        start_http_server(port)
+        print(f'Prometheus server started on port {port}.')
+        metrics.APP_UP.labels(pod_id=POD_ID).set(1)  # Mark app as UP
+    except Exception as e:
+        print(f'Failed to start Prometheus server: {e}')
+        os._exit(1)
+
+
+if __name__ == '__main__':
+    asyncio.run(main())

laborious/workflows/drift.py

@@ -0,0 +1,107 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+
+
+@workflow.defn(name='drift')
+class Drift:
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the drift workflow.
+
+        This method orchestrates the complete drift process by:
+        1. Loading data using the provided custom SQL query
+        2. Preparing prediction configuration and filters
+        3. Delegating to the PredictionProcess workflow for ML operations
+        """
+
+        metadata = {
+            'metadata': {
+                'schedule_name': input_data['schedule_name'],
+                'model_name': input_data['model_name'],
+                'model_id': input_data['model_id'],
+                'workflow_name': 'drift',
+            }
+        }
+
+        print(f'Input data: {input_data}', metadata)
+
+        model_config = input_data['model_config']
+        target_name = model_config['target']
+
+        gathering_query = f"""
+            SELECT *
+            FROM "{input_data['schema']}"."{input_data['source_table_name']}"
+            WHERE
+                model_id = '{input_data['model_id']}' AND
+                timestamp > NOW() - INTERVAL '{input_data['interval']} minutes'
+            ORDER BY timestamp ASC
+        """  # nosec B608 - values come from internal Temporal workflow config, not user input
+
+        target_data_handler = workflow.start_activity_method(
+            Activities.load_custom_query,
+            {
+                **metadata,
+                'query': gathering_query,
+                'datetime_columns': ['timestamp', 'created_at'],
+                'orient': 'records',
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        reference_data_handler = workflow.start_activity_method(
+            Activities.get_reference_data,
+            {**metadata, 'model_name': input_data['model_name']},
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        target_data = await target_data_handler
+        reference_data = await reference_data_handler
+
+        if not target_data:
+            return
+
+        drift_data = await workflow.execute_local_activity_method(
+            Activities.calculate_drift,
+            {
+                **metadata,
+                'target_data': target_data,
+                'reference_data': reference_data,
+                'model_name': input_data['model_name'],
+                'model_id': input_data['model_id'],
+                'target_name': target_name,
+                'drift_metrics': input_data.get(
+                    'drift_metrics', ['kolmogorov_smirnov', 'jensen_shannon', 'wasserstein']
+                ),
+                'chunk_period': input_data.get('chunk_period', 'min'),
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        if drift_data:
+            await workflow.execute_activity_method(
+                Activities.export_data_to_postgres,
+                {
+                    **metadata,
+                    'data': drift_data,
+                    'schema': input_data['schema'],
+                    'table_name': input_data['target_table_name'],
+                    'timestamp_conversion': {
+                        'column': 'timestamp',
+                        'format': DATETIME_FORMAT_WITH_TZ,
+                    },
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=300),
+            )

laborious/workflows/minimal_retrain.py

@@ -0,0 +1,137 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+    from laborious.utils.models.minio_dataframe_payload import MinioDataFramePayload
+
+
+@workflow.defn(name='minimal_retrain')
+class MinimalRetrain:
+    """
+    Automated model retraining workflow for the Laborious system.
+
+    This workflow implements a complete model retraining pipeline that loads
+    training data, executes model retraining, updates production models,
+    and maintains comprehensive audit trails. It's designed for automated
+    model lifecycle management with minimal manual intervention.
+
+    The workflow provides a robust retraining process with:
+    - Automated data loading from configured data sources
+    - MLFlow model retraining with quality validation
+    - Production model updates with version control
+    - Comprehensive reporting and audit trail maintenance
+    - Error handling and notification integration
+    """
+
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the automated model retraining workflow.
+
+        This method orchestrates the complete model retraining process by:
+        1. Loading training data using the provided custom SQL query
+        2. Executing MLFlow model retraining with the loaded data
+        3. Updating production models with newly trained versions
+        4. Persisting comprehensive retraining reports to database
+
+        The method implements comprehensive error handling and ensures all
+        required parameters are properly configured before proceeding.
+
+        Args:
+            input_data: Complete configuration for the retraining workflow
+                Required keys:
+                    - schedule_name (str): Schedule identifier for the retraining
+                    - model_name (str): Name of the ML model to retrain
+                    - model_id (int): Unique identifier for the model version
+                    - query (str): SQL query for training data loading
+                    - schema (str, optional): Database schema for report storage
+                    - table_name (str, optional): Target table for retraining reports
+                    - datetime_columns (list[str], optional): Columns to treat as datetime
+
+        Returns:
+            None: The workflow completes successfully when all steps finish
+
+        Raises:
+            Exception: If any required parameters are missing or if the workflow fails
+                during data loading, retraining, or model update operations
+        """
+
+        metadata = {
+            'metadata': {
+                'schedule_name': input_data['schedule_name'],
+                'model_name': input_data['model_name'],
+                'model_id': input_data['model_id'],
+                'workflow_name': 'minimal_retrain',
+            }
+        }
+
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+
+        storage_result = await workflow.execute_activity_method(
+            Activities.load_query_with_minio_offload,
+            {
+                **metadata,
+                'query': input_data['query'],
+                'datetime_columns': input_data.get('datetime_columns', []),
+                'model_name': model_name,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=600),
+        )
+
+        storage_payload = MinioDataFramePayload.from_dict(storage_result)
+        if not storage_payload.has_data():
+            raise ValueError('No data returned from query')
+
+        experiment_response = await workflow.execute_activity_method(
+            Activities.retrain_model,
+            {
+                **metadata,
+                'data': storage_result,
+                'model_name': model_name,
+                'model_config': model_config,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(hours=1),
+        )
+
+        if experiment_response['success']:
+            update_report = await workflow.execute_activity_method(
+                Activities.update_production_model,
+                {**metadata, 'model_name': model_name, **experiment_response},
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=60),
+            )
+        else:
+            update_report = {}
+
+        report = await workflow.execute_local_activity_method(
+            Activities.format_retrain_report,
+            {
+                **metadata,
+                'experiment_response': experiment_response,
+                'model_name': model_name,
+                'model_id': input_data['model_id'],
+                'update_report': update_report,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=60),
+        )
+
+        await workflow.execute_activity_method(
+            Activities.export_data_to_postgres,
+            {
+                **metadata,
+                'data': report,
+                'schema': input_data['schema'],
+                'table_name': input_data['table_name'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=600),
+        )

laborious/workflows/predictions_batch.py

@@ -0,0 +1,127 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+
+
+@workflow.defn(name='predictions_batch')
+class PredictionsBatch:
+    """
+    Main batch prediction workflow for the Laborious system.
+
+    This workflow orchestrates the complete batch prediction process, handling
+    data loading, configuration management, and workflow delegation. It serves
+    as the primary entry point for batch prediction operations and ensures
+    proper data preparation before ML model inference.
+
+    The workflow implements a robust data processing pipeline with:
+    - Custom SQL query execution for data loading
+    - Comprehensive configuration management
+    - Data quality filter application
+    - MLFlow model integration
+    - Workflow delegation to specialized sub-workflows
+
+    Workflow Execution:
+        1. Data Loading: Executes custom SQL query to load prediction data
+        2. Configuration Preparation: Sets up prediction parameters and filters
+        3. Workflow Delegation: Spawns PredictionProcess child workflow
+        4. Error Handling: Implements comprehensive error handling and retry policies
+    """
+
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the batch prediction workflow.
+
+        This method orchestrates the complete batch prediction process by:
+        1. Loading data using the provided custom SQL query
+        2. Preparing prediction configuration and filters
+        3. Delegating to the PredictionProcess workflow for ML operations
+
+        The method implements comprehensive error handling and ensures all
+        required parameters are properly configured before proceeding.
+
+        Args:
+            input_data: Complete configuration for the batch prediction
+                Required keys:
+                    - schedule_name (str): Schedule identifier for the prediction
+                    - model_name (str): Name of the ML model to use
+                    - model_id (int): Unique identifier for the model
+                    - query (str): SQL query for data loading
+                    - schema (dict, optional): Data schema definition
+                    - table_name (str, optional): Target table for predictions
+                    - input_filters (dict, optional): Data quality filters
+                    - mlflow_transform_filters (dict, optional): MLFlow transform filters
+                    - mlflow_predict_filters (dict, optional): MLFlow prediction filters
+                    - model_retention (int, optional): Model retention period in minutes
+                    - path_priority (list[str]): Decision path priority configuration
+                    - opc_output_config (dict, optional): OPC server export configuration
+                    - pi_web_api_output_config (dict, optional): PI Web API export configuration
+                    - datetime_columns (list[str], optional): Columns to treat as datetime
+                    - save_transform (bool, optional): Whether to save transformed data (default: True)
+                    - prediction_store_policy (str, optional): Data retention policy (default: 'lts:1')
+
+        Returns:
+            None: The workflow completes successfully when the child workflow finishes
+
+        Raises:
+            Exception: If any required parameters are missing or if the workflow fails
+                during data loading or workflow delegation
+        """
+
+        metadata = {
+            'metadata': {
+                'schedule_name': input_data['schedule_name'],
+                'model_name': input_data['model_name'],
+                'model_id': input_data['model_id'],
+                'workflow_name': 'predictions_batch',
+            }
+        }
+
+        # Load data using custom query with optional MinIO offload for large frames
+        data = await workflow.execute_activity_method(
+            Activities.load_query_with_minio_offload,
+            {
+                **metadata,
+                'query': input_data['query'],
+                'datetime_columns': input_data.get('datetime_columns', []),
+                'model_name': input_data['model_name'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        # Prepare input for prediction_process workflow
+        prediction_input = {
+            'metadata': metadata,
+            'data': data,
+            'schema': input_data['schema'],
+            'table_name': input_data['table_name'],
+            'transform_table_name': input_data['transform_table_name'],
+            'model_id': input_data['model_id'],
+            'model_name': input_data['model_name'],
+            'input_filters': input_data.get(
+                'input_filters', {'EMPTY_DATA': {'POLICY': 'STOP', 'CONFIG': {}}}
+            ),
+            'mlflow_transform_filters': input_data.get(
+                'mlflow_transform_filters', {'API_ERROR': {'POLICY': 'STOP', 'CONFIG': {}}}
+            ),
+            'mlflow_predict_filters': input_data.get(
+                'mlflow_predict_filters', {'API_ERROR': {'POLICY': 'STOP', 'CONFIG': {}}}
+            ),
+            'model_config': input_data.get('model_config', {}),
+            'path_priority': input_data.get('path_priority', ['STOP', 'CONTINUE', 'REPEAT']),
+            'opc_output_config': input_data.get('opc_output_config', {}),
+            'on_conflict': input_data.get('on_conflict', 'error'),
+            'pi_web_api_output_config': input_data.get('pi_web_api_output_config', {}),
+            'prediction_store_policy': input_data.get('prediction_store_policy', 'lts:1'),
+            'save_transform': input_data.get('save_transform', True),
+        }
+
+        # Execute prediction process workflow
+        await workflow.execute_child_workflow('subworkflow.prediction_process', prediction_input)

laborious/workflows/simple_metrics.py

@@ -0,0 +1,95 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+
+
+@workflow.defn(name='simple_metrics')
+class SimpleMetrics:
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the simple metrics workflow.
+        """
+        metadata = {
+            'metadata': {
+                'model_id': input_data['model_id'],
+                'model_name': input_data['model_name'],
+                'workflow_name': 'simple_metrics',
+                'schedule_name': input_data['schedule_name'],
+            }
+        }
+
+        model_id = input_data['model_id']
+        interval_minutes = input_data['interval_minutes']
+
+        model_config = input_data['model_config']
+        target_name = model_config['target']
+
+        query = f"""
+            select p."timestamp", p.prediction, ld.value as "target"
+            from "{input_data['schema']}"."{input_data['predictions_table_name']}" p
+            inner join "{input_data['schema']}"."{input_data['data_table_name']}" ld
+                on p."timestamp" = ld."timestamp"
+            where
+                p.model_id = '{model_id}' and
+                p.prediction is not null and
+                ld.variable = '{target_name}' and
+                ld.value is not null and
+                p."timestamp" >= NOW() - INTERVAL '{interval_minutes} minutes'
+            order by
+                p."timestamp" desc;
+        """  # nosec B608 - values come from internal Temporal workflow config, not user input
+
+        target_data = await workflow.execute_activity_method(
+            Activities.load_custom_query,
+            {
+                **metadata,
+                'query': query,
+                'datetime_columns': ['timestamp'],
+                'orient': 'records',
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        if not target_data:
+            return
+
+        simple_metrics = await workflow.execute_local_activity_method(
+            Activities.calculate_simple_metrics,
+            {
+                **metadata,
+                'model_id': model_id,
+                'target_data': target_data,
+                'metrics': input_data.get('metrics', ['rmse', 'mse', 'mae', 'r2']),
+                'interval_minutes': interval_minutes,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )
+
+        if not simple_metrics:
+            return
+
+        await workflow.execute_activity_method(
+            Activities.export_data_to_postgres,
+            {
+                **metadata,
+                'data': simple_metrics,
+                'schema': input_data['schema'],
+                'table_name': input_data['target_table_name'],
+                'timestamp_conversion': {
+                    'column': 'timestamp',
+                    'format': DATETIME_FORMAT_WITH_TZ,
+                },
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=300),
+        )

laborious/workflows/sub_workflows/format_and_export_prediction.py

@@ -0,0 +1,220 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+
+
+@workflow.defn(name='subworkflow.format_and_export_prediction')
+class FormatAndExportPrediction:
+    """
+    Data formatting and export workflow for prediction results.
+
+    This workflow handles the final stages of the prediction pipeline, including
+    data formatting, database persistence, OPC server export, and metrics recording.
+    It implements flexible formatting based on prediction quality and provides
+    comprehensive export capabilities to multiple destinations.
+
+    The workflow supports two main prediction paths:
+    1. Normal Prediction: Formats and exports successful prediction results
+    2. Default Prediction: Creates fallback predictions for error conditions
+
+    Export Destinations:
+    - PostgreSQL Database: Persistent storage with timestamp conversion
+    - PI Web API: Real-time industrial system integration for prediction and confidence values
+    - OPC Servers: Real-time industrial system integration
+    - Prometheus Metrics: Performance monitoring and operational visibility
+    """
+
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the prediction formatting and export workflow.
+
+        This method orchestrates the complete data export process by:
+        1. Determining the appropriate formatting strategy based on path_flag
+        2. Formatting prediction data according to quality and requirements
+        3. Exporting data to PI Web API for real-time industrial access (if configured)
+        4. Exporting data to OPC servers for real-time industrial access (if configured)
+        5. Persisting data to PostgreSQL database with comprehensive metadata
+        6. Recording performance metrics for operational monitoring
+
+        The method implements flexible formatting strategies:
+        - Normal predictions: Full data formatting with confidence scores
+        - Error predictions: Default formatting with error indicators
+        - Comprehensive export: Multi-destination data distribution
+
+        Args:
+            input_data: Complete configuration for the export workflow
+                Required keys:
+                    - metadata (dict): Workflow execution metadata
+                    - path_flag (str | None): Decision path flag for formatting strategy
+                        - None: Normal prediction path with full formatting
+                        - Any other value: Default prediction path for error conditions
+                    - data (dict[str, Any]): Prediction data to format and export
+                    - prediction_confidence (float): Confidence score for the prediction
+                    - timestamp (str): ISO-formatted timestamp for the prediction
+                    - model_id (int): Unique identifier for the ML model
+                    - model_name (str): Name of the ML model
+                    - schema (str): Database schema for data storage
+                    - table_name (str): Target table for data persistence
+                Optional keys:
+                    - opc_output_config (dict[str, Any]): OPC server export configuration
+                    - pi_web_api_output_config (dict[str, Any]): PI Web API export configuration
+                        Contains endpoint, prediction_tags, and confidence_tags mappings
+                    - transformed_data (dict[str, Any]): Transformed data to export separately
+                        Only processed when path_flag is None
+                    - transform_table_name (str): Target table for transformed data export
+                        Required if transformed_data is provided
+                    - prediction_store_policy (str): Data retention policy (e.g., 'lts:1', 'erl:2')
+                        Required when path_flag is None
+                    - comment (str): Operational comment or error description
+                        Required when path_flag is not None
+
+        Returns:
+            None: The workflow completes successfully when all export operations finish
+
+        Note:
+            When transformed_data is provided and path_flag is None, the workflow will:
+            1. Format the transformed data using format_transformed_data
+            2. Export it to a separate table (transform_table_name) asynchronously
+            3. Wait for both prediction and transformed data exports to complete
+        """
+        metadata = input_data['metadata']
+        path_flag = input_data['path_flag']
+        data = input_data['data']
+        transformed_data = input_data.get('transformed_data', None)
+        prediction_confidence = input_data['prediction_confidence']
+
+        opc_output_config = input_data.get('opc_output_config', None)
+        pi_web_api_output_config = input_data.get('pi_web_api_output_config', None)
+
+        if path_flag is None:
+            # Normal prediction path: format prediction data with full metadata
+            prediction = await workflow.execute_local_activity_method(
+                Activities.format_prediction,
+                {
+                    **metadata,
+                    'data': data,
+                    'timestamp': input_data['timestamp'],
+                    'model_id': input_data['model_id'],
+                    'prediction_confidence': prediction_confidence,
+                    'prediction_store_policy': input_data['prediction_store_policy'],
+                    'model_name': input_data['model_name'],
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=60),
+            )
+
+            # Optionally format and export transformed data to separate table
+            if transformed_data is not None:
+                transformed = await workflow.execute_local_activity_method(
+                    Activities.format_transformed_data,
+                    {
+                        **metadata,
+                        'data': transformed_data,
+                        'model_id': input_data['model_id'],
+                        'model_name': input_data['model_name'],
+                    },
+                    retry_policy=retry_policy,
+                    start_to_close_timeout=timedelta(seconds=60),
+                )
+
+                write_transformed_handler = workflow.start_activity_method(
+                    Activities.export_payload_to_postgres,
+                    {
+                        **metadata,
+                        'schema': input_data['schema'],
+                        'table_name': input_data['transform_table_name'],
+                        'data': transformed,
+                        'timestamp_conversion': {
+                            'column': 'timestamp',
+                            'format': DATETIME_FORMAT_WITH_TZ,
+                        },
+                    },
+                    retry_policy=retry_policy,
+                    start_to_close_timeout=timedelta(seconds=60),
+                )
+
+            else:
+                write_transformed_handler = None
+
+        else:
+            # Error path: create default prediction with error indicators
+            prediction = await workflow.execute_local_activity_method(
+                Activities.format_default_prediction,
+                {
+                    **metadata,
+                    'timestamp': input_data['timestamp'],
+                    'model_id': input_data['model_id'],
+                    'prediction_confidence': prediction_confidence,
+                    'comment': input_data['comment'],
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=60),
+            )
+
+            write_transformed_handler = None
+
+        opc_metrics: dict[str, dict[str, float | None]] = {}
+
+        # write to pi web api
+        if pi_web_api_output_config:
+            prediction = await workflow.execute_activity_method(
+                Activities.write_pi_web_api_data,
+                {
+                    'pi_web_api_output_config': pi_web_api_output_config,
+                    'data': prediction,
+                    **metadata,
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=60),
+            )
+
+        # write to opc
+        if opc_output_config:
+            prediction, opc_metrics = await workflow.execute_activity_method(
+                Activities.write_opc_data,
+                {
+                    'opc_output_config': opc_output_config,
+                    'data': prediction,
+                    **metadata,
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(seconds=60),
+            )
+
+        # write to postgres
+        await workflow.execute_activity_method(
+            Activities.export_data_to_postgres,
+            {
+                **metadata,
+                'schema': input_data['schema'],
+                'table_name': input_data['table_name'],
+                'data': prediction,
+                'timestamp_conversion': {'column': 'timestamp', 'format': DATETIME_FORMAT_WITH_TZ},
+                'on_conflict': input_data.get('on_conflict', 'error'),
+                'unique_columns': ['model_id', 'timestamp'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=180),
+        )
+
+        if write_transformed_handler is not None:
+            await write_transformed_handler
+
+        await workflow.execute_activity_method(
+            Activities.write_metrics,
+            {
+                **metadata,
+                'prediction': prediction,
+                'opc_metrics': opc_metrics,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(seconds=60),
+        )

laborious/workflows/sub_workflows/prediction_process.py

@@ -0,0 +1,346 @@
+from temporalio import workflow
+
+with workflow.unsafe.imports_passed_through():
+    from datetime import timedelta
+    from typing import Any
+
+    from sientia_do.temporal.policies import retry_policy
+
+    from laborious.activities.activities import Activities
+
+
+@workflow.defn(name='subworkflow.prediction_process')
+class PredictionProcess:
+    """
+    Core prediction processing workflow for the Laborious system.
+
+    This workflow implements the complete ML model inference pipeline, handling
+    data quality validation, MLFlow model interactions, and prediction processing.
+    It serves as the central orchestrator for all prediction operations and ensures
+    data quality throughout the entire process.
+
+    The workflow implements a robust data processing pipeline with:
+    - Data quality validation using configurable filters
+    - MLFlow model transformation and prediction
+    - Response validation and quality assurance
+    - Flexible decision path handling
+    - Comprehensive error handling and retry policies
+
+    Workflow Execution:
+        1. Timestamp Retrieval: Gets last processed timestamp for incremental processing
+        2. Input Data Gate: Applies data quality filters
+        3. Path Decision: Determines processing path based on filter results
+        4. MLFlow Transform: Requests data transformation using MLFlow models
+        5. Response Validation: Filters transform responses for quality assurance
+        6. MLFlow Prediction: Executes prediction using transformed data
+        7. Content Validation: Filters prediction responses for final quality check
+        8. Export Delegation: Delegates to FormatAndExportPrediction workflow
+    """
+
+    @workflow.run
+    async def run(self, input_data: dict[str, Any]):
+        """
+        Execute the prediction process workflow.
+
+        This method orchestrates the complete prediction processing pipeline by:
+        1. Retrieving the last processed timestamp for incremental processing
+        2. Applying data quality filters to validate input data
+        3. Executing MLFlow model transformation and prediction
+        4. Validating all responses for quality assurance
+        5. Delegating to export workflow for data persistence
+
+        The method implements comprehensive error handling and ensures all
+        data quality requirements are met before proceeding with ML operations.
+
+        Args:
+            input_data: Complete configuration for the prediction process
+                Required keys:
+                    - metadata (dict): Workflow execution metadata
+                    - data (dict): Input data for prediction processing
+                    - schema (dict): Data schema definition
+                    - table_name (str): Target table for predictions
+                    - model_id (str): ML model identifier
+                    - model_name (str): ML model name
+                    - input_filters (dict): Data quality filters
+                    - mlflow_transform_filters (dict): MLFlow transform filters
+                    - mlflow_predict_filters (dict): MLFlow prediction filters
+                    - model_retention (int): Model retention period in minutes
+                    - path_priority (list[str]): Decision path priority configuration
+                    - opc_output_config (dict, optional): OPC server export configuration
+                    - pi_web_api_output_config (dict, optional): PI Web API export configuration
+                    - save_transform (bool, optional): Whether to save transformed data (default: True)
+                    - prediction_store_policy (str, optional): Data retention policy (default: 'lts:1')
+
+        Returns:
+            None: The workflow completes successfully when export workflow finishes
+
+        Raises:
+            Exception: If any required parameters are missing or if the workflow fails
+                during data processing, MLFlow operations, or workflow delegation
+
+        """
+
+        metadata = input_data['metadata']
+        data = input_data['data']
+        model_id = input_data['model_id']
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+        save_transform = input_data.get('save_transform', True)
+
+        try:
+            await self._run_prediction_pipeline(
+                input_data,
+                metadata,
+                data,
+                model_id,
+                model_name,
+                model_config,
+                save_transform,
+            )
+            await workflow.execute_activity_method(
+                Activities.cleanup_minio_objects_expired,
+                {**metadata, 'data': data},
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(minutes=5),
+            )
+        except Exception as e:
+            await workflow.execute_activity_method(
+                Activities.cleanup_minio_objects_expired,
+                {**metadata, 'data': data},
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(minutes=5),
+            )
+            raise e
+
+    async def _run_prediction_pipeline(
+        self,
+        input_data: dict[str, Any],
+        metadata: dict[str, Any],
+        data: dict[str, Any],
+        model_id: str,
+        model_name: str,
+        model_config: dict[str, Any],
+        save_transform: bool,
+    ) -> None:
+        last_timestamp = data['last_timestamp']
+
+        # Apply input data quality gates
+        gate_input = {
+            **metadata,
+            'filters': input_data['input_filters'],
+            'data': data,
+            'path_priority': input_data['path_priority'],
+        }
+
+        path_flag, confidence, comment = await workflow.execute_local_activity_method(
+            Activities.input_gate,
+            gate_input,
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=1),
+        )
+
+        # Handle path decision based on filter results
+        if await self.path_flag_handler(
+            data, path_flag, input_data, confidence, last_timestamp, comment
+        ):
+            return
+
+        # Request MLFlow model transformation
+        transformed_data = await workflow.execute_activity_method(
+            Activities.request_transform,
+            {**metadata, 'data': data, 'model_name': model_name, 'model_config': model_config},
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=5),
+        )
+
+        # Validate MLFlow transform response
+        path_flag, confidence, comment = await workflow.execute_local_activity_method(
+            Activities.mlflow_response_gate,
+            {
+                **metadata,
+                'filters': input_data['mlflow_transform_filters'],
+                'data': transformed_data,
+                'type': 'transform',
+                'path_priority': input_data['path_priority'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=1),
+        )
+
+        # Handle path decision based on transform validation
+        if await self.path_flag_handler(
+            data, path_flag, input_data, confidence, last_timestamp, comment
+        ):
+            return
+
+        path_flag, confidence, comment = await workflow.execute_local_activity_method(
+            Activities.mlflow_content_gate,
+            {
+                **metadata,
+                'filters': input_data['mlflow_transform_filters'],
+                'data': transformed_data,
+                'type': 'transform',
+                'path_priority': input_data['path_priority'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=1),
+        )
+
+        if await self.path_flag_handler(
+            data, path_flag, input_data, confidence, last_timestamp, comment
+        ):
+            return
+
+        predicted_data = await workflow.execute_activity_method(
+            Activities.request_predict,
+            {
+                **metadata,
+                'data': transformed_data,
+                'model_name': model_name,
+                'model_config': model_config,
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=5),
+        )
+
+        # Validate MLFlow prediction response
+        path_flag, confidence, comment = await workflow.execute_local_activity_method(
+            Activities.mlflow_response_gate,
+            {
+                **metadata,
+                'filters': input_data['mlflow_predict_filters'],
+                'data': predicted_data,
+                'type': 'predict',
+                'path_priority': input_data['path_priority'],
+            },
+            retry_policy=retry_policy,
+            start_to_close_timeout=timedelta(minutes=1),
+        )
+
+        # Handle path decision based on prediction validation
+        if await self.path_flag_handler(
+            data, path_flag, input_data, confidence, last_timestamp, comment
+        ):
+            return
+
+        # Delegate to export workflow for data persistence
+        await workflow.execute_child_workflow(
+            'subworkflow.format_and_export_prediction',
+            {
+                'metadata': metadata,
+                'on_conflict': input_data.get('on_conflict', 'error'),
+                'path_flag': path_flag,
+                'data': predicted_data,
+                'transformed_data': transformed_data if save_transform else None,
+                'prediction_confidence': confidence,
+                'timestamp': last_timestamp,
+                'model_id': model_id,
+                'model_name': model_name,
+                'model_config': model_config,
+                'opc_output_config': input_data['opc_output_config'],
+                'pi_web_api_output_config': input_data['pi_web_api_output_config'],
+                'schema': input_data['schema'],
+                'table_name': input_data['table_name'],
+                'transform_table_name': input_data['transform_table_name'],
+                'comment': comment,
+                'prediction_store_policy': input_data['prediction_store_policy'],
+            },
+        )
+
+    async def path_flag_handler(
+        self,
+        data: dict[str, Any],
+        path_flag: str | None,
+        input_data: dict,
+        confidence: int,
+        last_timestamp: str,
+        comment: str,
+    ) -> bool:
+        """
+        Handle path decisions based on filter results and confidence levels.
+
+        This method determines the appropriate action based on the path flag
+        returned by data quality filters. It can stop processing, continue,
+        or repeat operations based on the configured path priority.
+
+        Args:
+            data: Input data for processing
+            path_flag: Path decision from filter (STOP, CONTINUE, REPEAT)
+            input_data: Complete workflow input configuration including:
+                - metadata (dict): Workflow execution metadata
+                - schema (str): Database schema
+                - table_name (str): Target table for predictions
+                - transform_table_name (str): Target table for transformed data
+                - model_id (str): ML model identifier
+                - model_name (str): ML model name
+                - model_config (dict, optional): Model configuration
+                - opc_output_config (dict, optional): OPC server export configuration
+                - pi_web_api_output_config (dict, optional): PI Web API export configuration
+                - prediction_store_policy (str, optional): Data retention policy
+            confidence: Confidence level from filter validation
+            last_timestamp: Last processed timestamp
+            comment: Additional information about the filter result
+
+        Returns:
+            bool: True if processing should stop, False to continue
+
+        Path Handling:
+            - STOP: Terminates workflow execution
+            - CONTINUE: Delegates to FormatAndExportPrediction workflow with current data
+            - REPEAT: Repeats last prediction if available
+        """
+        metadata = input_data['metadata']
+
+        schema = input_data['schema']
+        table_name = input_data['table_name']
+        transform_table_name = input_data['transform_table_name']
+        model_id = input_data['model_id']
+        model_name = input_data['model_name']
+        model_config = input_data.get('model_config', {})
+
+        path_flag = path_flag.upper() if path_flag else ''
+
+        if path_flag == 'STOP':
+            # Stop processing and exit workflow
+            return True
+        elif path_flag == 'REPEAT':
+            # Repeat last prediction if available
+            await workflow.execute_activity_method(
+                Activities.repeat_last_prediction,
+                {
+                    **metadata,
+                    'schema': schema,
+                    'table_name': table_name,
+                    'model': model_id,
+                    'last_timestamp': last_timestamp,
+                },
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(minutes=1),
+            )
+            return True
+        elif path_flag == 'CONTINUE':
+            # call write workflow
+            await workflow.execute_child_workflow(
+                'subworkflow.format_and_export_prediction',
+                {
+                    'metadata': metadata,
+                    'path_flag': path_flag,
+                    'data': data,
+                    'prediction_confidence': confidence,
+                    'timestamp': last_timestamp,
+                    'model_id': model_id,
+                    'model_name': model_name,
+                    'model_config': model_config,
+                    'schema': schema,
+                    'table_name': table_name,
+                    'transform_table_name': transform_table_name,
+                    'comment': comment,
+                    'opc_output_config': input_data['opc_output_config'],
+                    'pi_web_api_output_config': input_data['pi_web_api_output_config'],
+                    'prediction_store_policy': input_data['prediction_store_policy'],
+                    'on_conflict': input_data.get('on_conflict', 'error'),
+                },
+            )
+            return True
+
+        return False