chore: bump version to 0.9.0-rc1

Feature rc/sientiapde 2164 - Prediction Metadata — Metadado rico no input_gate

laborious/activities/activities.py

@@ -17,18 +17,21 @@ with workflow.unsafe.imports_passed_through():
     from laborious.activities.mlflow import MLFlow
     from laborious.activities.model_metrics import ModelMetrics
     from laborious.activities.opc import OPC
+    from laborious.activities.prediction_metadata import PredictionMetadata
     from laborious.activities.storage import Storage
     from laborious.utils.connectors_config import build_mlflow_config
 
 
-class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
+class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API, PredictionMetadata):
     """
     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
+    drift/simple metrics, PI Web API writes, and PredictionMetadata (the single write point for
-    per process and registers its callables on multiple workers bound to different task queues.
+    gate metadata — see `laborious.activities.prediction_metadata`). 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
@@ -70,7 +73,7 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
          - 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()``.
-         - mongo_repository: Optional ``MongoDBRepository`` (gate metadata, best-effort) for
+         - mongo_repository: Optional ``MongoDBRepository`` (prediction metadata, best-effort) for
            unit/e2e tests; in production leave unset so the repository is built from environment
            via ``build_mongodb_config()``, same pattern as ``mlflow_repository`` above.
 
@@ -145,6 +148,13 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
         Gates.__init__(
             self,
             minio_repository=minio_repository,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=mc,
+        )
+
+        PredictionMetadata.__init__(
+            self,
             mongo_repository=mongo_repository,
             logger=logger,
             notification_handler=notification_handler,
@@ -192,3 +202,4 @@ class Activities(Storage, MLFlow, Gates, OPC, ModelMetrics, API):
         OPC.close(self)
         ModelMetrics.close(self)
         API.close(self)
+        PredictionMetadata.close(self)

laborious/activities/gates.py

@@ -5,7 +5,6 @@ with workflow.unsafe.imports_passed_through():
     from collections.abc import Callable, Mapping
     from typing import Any
 
-    from bson import ObjectId
     from pandas import DataFrame
     from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
     from sientia_do.notifications.models import NotificationLevel
@@ -13,7 +12,6 @@ with workflow.unsafe.imports_passed_through():
     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.repository.mongodb_repository_sync import MongoDBRepository
     from sientia_do.utils.formatters import create_sample_dict
 
     from laborious import metrics
@@ -29,6 +27,7 @@ with workflow.unsafe.imports_passed_through():
 InputFilterFunc = Callable[[DataFrame, dict[str, Any]], bool]
 ResponseFilterFunc = Callable[[dict[str, Any], dict[str, Any]], bool]
 ContentFilterFunc = Callable[[DataFrame, dict[str, Any]], bool]
+InputCounterFunc = Callable[[DataFrame, dict[str, Any]], int]
 
 # Input filter function mappings
 input_filter_functions: dict[str, InputFilterFunc] = {
@@ -36,6 +35,42 @@ input_filter_functions: dict[str, InputFilterFunc] = {
     'EMPTY_DATA': filter_empty_data,
 }
 
+
+def _count_specific_variables_null_values(data: DataFrame, config: dict[str, Any]) -> int:
+    """
+    Count rows with a null value in one of the configured variables.
+
+    Mirrors `filter_specific_variables_null_values`'s own trigger condition, so the count
+    reflects exactly what made that filter fire.
+
+    Return:
+        int: Number of rows with a null value in `config['variables']`.
+    """
+    if data.empty:
+        return 0
+    return int((data['variable'].isin(config['variables']) & data['value'].isna()).sum())
+
+
+def _count_empty_data(_data: DataFrame, _config: dict[str, Any]) -> int:
+    """
+    Count rows for the EMPTY_DATA filter.
+
+    `EMPTY_DATA` only triggers when there are no rows at all, so the count is always 0.
+
+    Return:
+        int: Always 0.
+    """
+    return 0
+
+
+# Per-filter row counts for the `counter` field in prediction metadata (gate-local: mirrors each
+# filter's own trigger condition instead of changing `conditional_filters.py`'s bool contract,
+# which `mlflow_content_gate` also depends on for EMPTY_DATA).
+input_filter_counters: dict[str, InputCounterFunc] = {
+    'SPECIFIC_VARIABLES_NULL_VALUES': _count_specific_variables_null_values,
+    'EMPTY_DATA': _count_empty_data,
+}
+
 # Confidence mappings kept separate from function maps to avoid Union types
 input_path_confidence: Mapping[str, int] = {
     'STOP': -1,
@@ -88,14 +123,11 @@ class Gates(SientiaMonitoring):
     """
 
     minio_repository: MinioRepository | None = None
-    mongo_repository: MongoDBRepository | None = None
     _MAX_DEBUG_DATAFRAME_ROWS = 100
-    _PREDICTION_METADATA_COLLECTION = 'prediction_metadata'
 
     def __init__(
         self,
         minio_repository: MinioRepository | None = None,
-        mongo_repository: MongoDBRepository | None = None,
         logger: Logger | None = None,
         notification_handler: NotificationHandler | None = None,
         metrics_controller: MetricsController | None = None,
@@ -106,14 +138,11 @@ class Gates(SientiaMonitoring):
         Args:
             logger: Logger instance for observability and debugging
             notification_handler: Notification handler for alerts and monitoring
-            mongo_repository: MongoDB, where gate metadata (triggered filter, policy, counter) is
-                written best-effort. `None` is a valid state (see `_write_prediction_metadata`).
 
         Raises:
             Exception: If BaseActivity initialization fails
         """
         self.minio_repository = minio_repository
-        self.mongo_repository = mongo_repository
         SientiaMonitoring.__init__(
             self,
             logger=logger,
@@ -131,11 +160,6 @@ class Gates(SientiaMonitoring):
                 self.minio_repository.close()
             finally:
                 self.minio_repository = None
-        if self.mongo_repository is not None:
-            try:
-                self.mongo_repository.close()
-            finally:
-                self.mongo_repository = None
         SientiaMonitoring.shutdown(self)
 
     def __del__(self):
@@ -159,56 +183,6 @@ class Gates(SientiaMonitoring):
             metadata,
         )
 
-    def _write_prediction_metadata(
-        self, document: dict[str, Any] | None, metadata: dict[str, Any]
-    ) -> str | None:
-        """
-        Best-effort write of gate metadata (triggered filter, policy, counter) to Mongo.
-
-        Implements both business rules for this feature in one place:
-        - The prediction write must never be blocked or failed by a metadata write. A failed
-          `insert(...)` is swallowed, never re-raised, logged here with gate-level context (which
-          `mongo_repository.insert(...)` itself can't know) — matching how `storage.py`'s
-          `cleanup_minio_objects_expired` also logs again at the call site even though the
-          repository already logs internally and emits its own Prometheus metric
-          (`MONGODB_WRITE_ERROR_COUNT`); the generic library log is for the metric/trace, this
-          one is for whoever is reading logs for this specific gate.
-        - Nothing is written when there's nothing relevant to report. A falsy `document` (e.g.
-          no filter triggered) skips Mongo entirely, so callers can call this unconditionally
-          and get `None` back for "no metadata this time" instead of deciding upfront whether
-          there's something worth saving.
-
-        `insert(...)` doesn't hand back the inserted document's `_id`, so it is generated here,
-        client-side, before the insert, as a plain string (not a BSON ObjectId) — matching this
-        codebase's convention of string `_id`s (e.g. the `models` collection) and the 24-hex-char
-        size of the `predictions.metadata_id` Postgres column that will reference it.
-
-        Args:
-            document: Metadata payload (filter, policy, counter, etc.); `None`/empty skips the
-                write entirely. No `_id` key expected.
-            metadata: Workflow metadata for contextual logging.
-
-        Return:
-            The generated id, or `None` when there's nothing to write, Mongo isn't configured, or
-            the insert failed.
-        """
-        if not document or self.mongo_repository is None:
-            return None
-
-        generated_id = str(ObjectId())
-
-        try:
-            self.mongo_repository.insert(
-                self._PREDICTION_METADATA_COLLECTION,
-                {**document, '_id': generated_id},
-                metadata,
-            )
-        except Exception as e:
-            self.error(f'Failed to write prediction metadata: {e}', metadata)
-            return None
-
-        return generated_id
-
     @staticmethod
     def _read_filter_entry(config: dict[str, Any]) -> tuple[str, dict[str, Any]]:
         """
@@ -226,7 +200,9 @@ class Gates(SientiaMonitoring):
         return policy, filter_config
 
     @activity.defn(name='input_gate')
-    def input_gate(self, input_data: dict[str, Any]) -> tuple[str | None, int, str]:
+    def input_gate(
+        self, input_data: dict[str, Any]
+    ) -> tuple[str | None, int, str, dict[str, Any] | None]:
         """
         Apply input data quality filters and validation.
 
@@ -241,6 +217,9 @@ class Gates(SientiaMonitoring):
         3. Determines appropriate path decisions (STOP, CONTINUE, REPEAT)
         4. Provides confidence scores and detailed comments
         5. Handles errors gracefully with notification integration
+        6. Builds a metadata document describing which filter(s) triggered, with their policy
+           and a counter, whenever one does — this activity only computes it, it never writes it
+           (see `PredictionMetadata.write_prediction_metadata` for the actual write)
 
         Args:
             input_data: Configuration and data for input validation
@@ -251,10 +230,14 @@ class Gates(SientiaMonitoring):
                     - path_priority (list[str]): Priority order for path decisions
 
         Returns:
-            tuple: (path_flag, confidence, comment)
+            tuple: (path_flag, confidence, comment, document)
                 - 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
+                - document (dict | None): Metadata describing the filter(s) that produced
+                    `path_flag`, shaped `{'filters': [{'name', 'policy', 'counter'}, ...]}`;
+                    `None` when nothing was filtered. Not written anywhere by this activity —
+                    the caller decides whether and how to persist it.
 
         Raises:
             Exception: If filter execution fails or configuration is invalid
@@ -268,7 +251,8 @@ class Gates(SientiaMonitoring):
         data = payload.retrieve(self.minio_repository, metadata)
         path_priority = input_data['path_priority']
 
-        filter_output = []
+        # (filter_name, policy, filter_config) per filter that triggered, in configured order
+        triggered_filters: list[tuple[str, str, dict[str, Any]]] = []
 
         self._debug_dataframe('Input data:', data, metadata)
         self.debug(f'Filters: {filters}', metadata)
@@ -282,7 +266,7 @@ class Gates(SientiaMonitoring):
             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)
+                    triggered_filters.append((fil, policy, filter_config))
             except Exception as e:
                 trace = traceback.format_exc()
                 self.send_notification(
@@ -294,16 +278,46 @@ class Gates(SientiaMonitoring):
                     attachment_content=trace,
                 )
 
+        triggered_policies = [policy for _, policy, _ in triggered_filters]
+
         for path_flag in path_priority:
-            if path_flag in filter_output:
+            if path_flag in triggered_policies:
                 self.info(f'Input gate result: {path_flag}', metadata)
-                return path_flag, input_path_confidence[path_flag], 'Input data with bad quality'
+
+                matched = [
+                    (fil, filter_config)
+                    for fil, policy, filter_config in triggered_filters
+                    if policy == path_flag
+                ]
+                document = {
+                    'filters': [
+                        {
+                            'name': fil,
+                            'policy': path_flag,
+                            # `.get(..., lambda *_: 0)`, not `[fil]`: a filter added to
+                            # `input_filter_functions` without a matching counter here degrades to
+                            # counter=0 instead of a raw KeyError crashing this prediction (review
+                            # finding, 2026-09-30 — see test enforcing key parity between the two).
+                            'counter': input_filter_counters.get(fil, lambda *_: 0)(
+                                data, filter_config
+                            ),
+                        }
+                        for fil, filter_config in matched
+                    ],
+                }
+
+                return (
+                    path_flag,
+                    input_path_confidence[path_flag],
+                    'Input data with bad quality',
+                    document,
+                )
 
         self.info('Nothing was filtered by the input gate', metadata)
 
         del data
 
-        return None, 0, ''
+        return None, 0, '', None
 
     @activity.defn(name='mlflow_response_gate')
     def mlflow_response_gate(self, input_data: dict[str, Any]) -> tuple[str | None, int, str]:
@@ -699,6 +713,9 @@ class Gates(SientiaMonitoring):
                 - 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
+                - metadata_id (str | None, optional): Id of the `prediction_metadata` Mongo
+                    document to reference for this row; `None`/absent when there's nothing to
+                    reference
 
         Returns:
             dict: Formatted default prediction data with error indicators
@@ -716,6 +733,7 @@ class Gates(SientiaMonitoring):
                 'prediction_confidence': [input_data['prediction_confidence']],
                 'prediction_status': ['Bad'],
                 'comments': [input_data['comment']],
+                'metadata_id': [input_data.get('metadata_id')],
             }
         )
 

laborious/activities/prediction_metadata.py

@@ -0,0 +1,132 @@
+from temporalio import activity, workflow
+
+with workflow.unsafe.imports_passed_through():
+    from typing import Any
+
+    from bson import ObjectId
+    from sientia_do.notifications.handlers import CoreNotificationHandler as NotificationHandler
+    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.mongodb_repository_sync import MongoDBRepository
+
+
+class PredictionMetadata(SientiaMonitoring):
+    """
+    Centralized, best-effort writer for prediction gate metadata (triggered filter, policy,
+    counter) to MongoDB.
+
+    Gates only compute and return a metadata document; they never write it themselves. Keeping
+    the write here, instead of in each gate, puts the one real write point in the subworkflow
+    that's about to persist a `predictions` row — the only place that reliably knows whether one
+    is being written at all — and lets other metadata sources reuse the same activity instead of
+    duplicating write logic wherever they originate.
+    """
+
+    mongo_repository: MongoDBRepository | None = None
+    _PREDICTION_METADATA_COLLECTION = 'prediction_metadata'
+
+    def __init__(
+        self,
+        mongo_repository: MongoDBRepository | None = None,
+        logger: Logger | None = None,
+        notification_handler: NotificationHandler | None = None,
+        metrics_controller: MetricsController | None = None,
+    ):
+        """
+        Initialize the prediction metadata writer.
+
+        Args:
+            mongo_repository: MongoDB, where gate metadata (triggered filter, policy, counter) is
+                written best-effort. `None` is a valid state (see `write_prediction_metadata`).
+
+        Raises:
+            Exception: If BaseActivity initialization fails
+        """
+        self.mongo_repository = mongo_repository
+        SientiaMonitoring.__init__(
+            self,
+            logger=logger,
+            notification_handler=notification_handler,
+            metrics_controller=metrics_controller,
+        )
+
+    def close(self) -> None:
+        """
+        Close the prediction metadata writer and clean up resources.
+        """
+        if self.mongo_repository is not None:
+            try:
+                self.mongo_repository.close()
+            finally:
+                self.mongo_repository = None
+        SientiaMonitoring.shutdown(self)
+
+    def __del__(self):
+        self.close()
+
+    @activity.defn(name='write_prediction_metadata')
+    def write_prediction_metadata(self, input_data: dict[str, Any]) -> str | None:
+        """
+        Best-effort write of one or more gate metadata documents to Mongo, combined into a single
+        `prediction_metadata` document.
+
+        Implements both business rules for this feature in one place:
+        - The prediction write must never be blocked or failed by a metadata write. A failed
+          `insert(...)` is swallowed, never re-raised, logged here with call-site context (which
+          `mongo_repository.insert(...)` itself can't know) — matching how `storage.py`'s
+          `cleanup_minio_objects_expired` also logs again at the call site even though the
+          repository already logs internally and emits its own Prometheus metric
+          (`MONGODB_WRITE_ERROR_COUNT`); the generic library log is for the metric/trace, this
+          one is for whoever is reading logs for this specific call.
+        - Nothing is written when there's nothing relevant to report. An empty/missing
+          `documents` list (or one full of falsy/empty entries) skips Mongo entirely, so callers
+          can call this unconditionally and get `None` back for "no metadata this time" instead
+          of deciding upfront whether there's something worth saving.
+
+        Takes a *list* of documents, not a single one: each caller contributes its own document
+        (shaped like `{'filters': [...]}`), and this flattens every `filters` entry across all of
+        them into one `prediction_metadata` document before the single insert. Today only one
+        gate ever contributes a document per prediction cycle (the gate chain short-circuits on
+        the first non-`None` `path_flag`), but the list shape means this activity won't need to
+        change again once more than one source contributes metadata for the same cycle.
+
+        `insert(...)` doesn't hand back the inserted document's `_id`, so it is generated here,
+        client-side, before the insert, as a plain string (not a BSON ObjectId) — matching this
+        codebase's convention of string `_id`s (e.g. the `models` collection) and the 24-hex-char
+        size of the `predictions.metadata_id` Postgres column that will reference it.
+
+        Args:
+            input_data: Dict with:
+                - documents (list[dict[str, Any]]): Metadata payloads to combine, each shaped
+                  like `{'filters': [...]}`. Missing/empty, or entries with no `filters`, skip
+                  the write entirely.
+                - metadata (dict): Workflow metadata for contextual logging.
+
+        Return:
+            The generated id, or `None` when there's nothing to write, Mongo isn't configured, or
+            the insert failed.
+        """
+        metadata = input_data['metadata']
+        documents = input_data.get('documents') or []
+
+        filters = [
+            entry for document in documents if document for entry in document.get('filters', [])
+        ]
+
+        if not filters or self.mongo_repository is None:
+            return None
+
+        generated_id = str(ObjectId())
+
+        try:
+            self.mongo_repository.insert(
+                self._PREDICTION_METADATA_COLLECTION,
+                {'filters': filters, '_id': generated_id},
+                metadata,
+            )
+        except Exception as e:
+            self.error(f'Failed to write prediction metadata: {e}', metadata)
+            return None
+
+        return generated_id

laborious/worker/worker.py

@@ -249,6 +249,8 @@ async def main():
                 activities.format_transformed_data,
                 activities.format_prediction,
                 activities.format_default_prediction,
+                # Prediction metadata
+                activities.write_prediction_metadata,
                 # OPC
                 activities.write_opc_data,
                 # Postgres / MinIO offload

laborious/workflows/sub_workflows/format_and_export_prediction.py

@@ -75,6 +75,11 @@ class FormatAndExportPrediction:
                         Required when path_flag is None
                     - comment (str): Operational comment or error description
                         Required when path_flag is not None
+                    - metadata_id (str | None): Id of the `prediction_metadata` Mongo document to
+                        reference for this row, already written by the caller; `None` when there's
+                        nothing to reference. Only meaningful when path_flag is not None — the
+                        normal (path_flag is None) path never has metadata to reference, so this
+                        key is irrelevant there.
 
         Returns:
             None: The workflow completes successfully when all export operations finish
@@ -154,6 +159,7 @@ class FormatAndExportPrediction:
                     'model_id': input_data['model_id'],
                     'prediction_confidence': prediction_confidence,
                     'comment': input_data['comment'],
+                    'metadata_id': input_data.get('metadata_id'),
                 },
                 retry_policy=retry_policy,
                 start_to_close_timeout=timedelta(seconds=60),

laborious/workflows/sub_workflows/prediction_process.py

@@ -132,7 +132,9 @@ class PredictionProcess:
             'path_priority': input_data['path_priority'],
         }
 
-        path_flag, confidence, comment = await workflow.execute_local_activity_method(
+        # `document` describes which filter(s) produced `path_flag` (`None` when nothing was
+        # filtered); `path_flag_handler` is the one that decides whether and how to persist it.
+        path_flag, confidence, comment, document = await workflow.execute_local_activity_method(
             Activities.input_gate,
             gate_input,
             retry_policy=retry_policy,
@@ -141,7 +143,7 @@ class PredictionProcess:
 
         # Handle path decision based on filter results
         if await self.path_flag_handler(
-            data, path_flag, input_data, confidence, last_timestamp, comment
+            data, path_flag, input_data, confidence, last_timestamp, comment, document
         ):
             return
 
@@ -247,6 +249,51 @@ class PredictionProcess:
             },
         )
 
+    async def _write_prediction_metadata(
+        self, metadata: dict[str, Any], document: dict[str, Any] | None
+    ) -> str | None:
+        """
+        Write a gate's metadata document via the centralized write activity.
+
+        Shared by the CONTINUE and REPEAT branches of `path_flag_handler` — both persist a new
+        `predictions` row (`REPEAT` reuses the previous row's values but still inserts a new
+        one), so both need a `metadata_id` to reference it.
+
+        Args:
+            metadata: Workflow metadata for contextual logging.
+            document: Metadata produced by the gate that resolved this `path_flag`; `None` when
+                nothing was filtered, or when the gate that produced this `path_flag` doesn't
+                build one yet.
+
+        Return:
+            The generated id, or `None` when there's nothing to write, including when the
+            activity call itself fails (see `PredictionMetadata.write_prediction_metadata`).
+        """
+        if not document:
+            # Skip the remote activity call entirely rather than dispatching it just to have
+            # `write_prediction_metadata` decide there's nothing to write — `mlflow_response_gate`
+            # and `mlflow_content_gate` don't build a `document` yet, so this is the common case
+            # today, not a rare edge case (review finding, 2026-10-01).
+            return None
+
+        try:
+            return await workflow.execute_activity_method(
+                Activities.write_prediction_metadata,
+                {**metadata, 'documents': [document]},
+                retry_policy=retry_policy,
+                start_to_close_timeout=timedelta(minutes=1),
+            )
+        except Exception as e:
+            # Business rule 1: writing metadata must never block or fail the prediction write.
+            # `write_prediction_metadata` already swallows Mongo-specific failures internally and
+            # returns `None`, but that protection doesn't cover the activity call itself failing
+            # to run at all (its `retry_policy` has `maximum_attempts=1` — no retry — so a
+            # timeout, a momentarily unavailable worker, or any other dispatch failure raises
+            # straight into the workflow). Catching it here, instead of just at the Mongo client
+            # level, is what actually makes the rule hold (review finding, 2026-10-01).
+            workflow.logger.warning(f'Failed to write prediction metadata, continuing anyway: {e}')
+            return None
+
     async def path_flag_handler(
         self,
         data: dict[str, Any],
@@ -255,6 +302,7 @@ class PredictionProcess:
         confidence: int,
         last_timestamp: str,
         comment: str,
+        document: dict[str, Any] | None = None,
     ) -> bool:
         """
         Handle path decisions based on filter results and confidence levels.
@@ -280,6 +328,13 @@ class PredictionProcess:
             confidence: Confidence level from filter validation
             last_timestamp: Last processed timestamp
             comment: Additional information about the filter result
+            document: Metadata describing the filter(s) that produced `path_flag`, as returned by
+                the gate (`None` when nothing was filtered, or when the gate that produced this
+                `path_flag` doesn't build one yet). Written via `write_prediction_metadata` for
+                `CONTINUE` and `REPEAT` alike — both persist a `predictions` row (`REPEAT` reuses
+                the previous row's values but still inserts a new one), so both can reference the
+                resulting `metadata_id`; `STOP` never reaches a write at all, so its `document` is
+                simply never used here.
 
         Returns:
             bool: True if processing should stop, False to continue
@@ -304,7 +359,11 @@ class PredictionProcess:
             # Stop processing and exit workflow
             return True
         elif path_flag == 'REPEAT':
-            # Repeat last prediction if available
+            metadata_id = await self._write_prediction_metadata(metadata, document)
+            # Repeat last prediction if available. `metadata_id` isn't consumed by
+            # `repeat_last_prediction` yet (its INSERT...SELECT has no `metadata_id` column —
+            # pending a change in `sientia-dataops-library`); passed through already so no further
+            # change is needed here once that lands.
             await workflow.execute_activity_method(
                 Activities.repeat_last_prediction,
                 {
@@ -313,12 +372,14 @@ class PredictionProcess:
                     'table_name': table_name,
                     'model': model_id,
                     'last_timestamp': last_timestamp,
+                    'metadata_id': metadata_id,
                 },
                 retry_policy=retry_policy,
                 start_to_close_timeout=timedelta(minutes=1),
             )
             return True
         elif path_flag == 'CONTINUE':
+            metadata_id = await self._write_prediction_metadata(metadata, document)
             # call write workflow
             await workflow.execute_child_workflow(
                 'subworkflow.format_and_export_prediction',
@@ -335,6 +396,7 @@ class PredictionProcess:
                     'table_name': table_name,
                     'transform_table_name': transform_table_name,
                     'comment': comment,
+                    'metadata_id': metadata_id,
                     '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'],

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "laborious"
-version = "0.9.0"
+version = "0.9.0-rc1"
 description = "Sientia DataOps Laborious - ML Model Orchestration System"
 readme = "README.md"
 requires-python = ">=3.11"

tests/laborious/activities/test_activities.py

@@ -6,6 +6,7 @@ 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.prediction_metadata import PredictionMetadata
 from laborious.activities.storage import Storage
 
 
@@ -13,6 +14,7 @@ from laborious.activities.storage import Storage
 @patch('laborious.activities.activities.MLFlow.__init__')
 @patch('laborious.activities.activities.OPC.__init__')
 @patch('laborious.activities.activities.Gates.__init__')
+@patch('laborious.activities.activities.PredictionMetadata.__init__')
 @patch('laborious.activities.activities.ModelMetrics.__init__')
 @patch('laborious.activities.activities.API.__init__')
 @patch('laborious.activities.activities.MinioRepository')
@@ -22,6 +24,7 @@ def test___init__(
     mock_minio_repository,
     mock_api_init,
     mock_model_metrics_init,
+    mock_prediction_metadata_init,
     mock_gates_init,
     mock_opc_init,
     mock_mlflow_init,
@@ -84,6 +87,7 @@ def test___init__(
     assert isinstance(activities, Gates)
     assert isinstance(activities, ModelMetrics)
     assert isinstance(activities, API)
+    assert isinstance(activities, PredictionMetadata)
 
     mock_storage_init.assert_called_once_with(
         ANY,
@@ -122,6 +126,13 @@ def test___init__(
     mock_gates_init.assert_called_once_with(
         ANY,
         minio_repository=mock_minio_repository.return_value,
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=mock_metrics_controller.return_value,
+    )
+
+    mock_prediction_metadata_init.assert_called_once_with(
+        ANY,
         mongo_repository=mongo_repository,
         logger=logger,
         notification_handler=notification_handler,
@@ -161,6 +172,7 @@ def test___init__(
 @patch('laborious.activities.activities.MLFlow')
 @patch('laborious.activities.activities.OPC')
 @patch('laborious.activities.activities.Gates')
+@patch('laborious.activities.activities.PredictionMetadata')
 @patch('laborious.activities.activities.ModelMetrics')
 @patch('laborious.activities.activities.API')
 @patch('laborious.activities.activities.MinioRepository')
@@ -168,6 +180,7 @@ def test_shutdown(
     _mock_minio_repository,
     mock_api_init,
     mock_model_metrics_init,
+    mock_prediction_metadata_init,
     mock_gates_init,
     mock_opc_init,
     mock_mlflow_init,
@@ -230,6 +243,7 @@ def test_shutdown(
     mock_gates_init.close.assert_called_once()
     mock_model_metrics_init.close.assert_called_once()
     mock_api_init.close.assert_called_once()
+    mock_prediction_metadata_init.close.assert_called_once()
 
 
 @patch('laborious.activities.activities.SientiaMLflowRepository')
@@ -238,6 +252,7 @@ def test_shutdown(
 @patch('laborious.activities.activities.MLFlow.__init__')
 @patch('laborious.activities.activities.OPC.__init__')
 @patch('laborious.activities.activities.Gates.__init__')
+@patch('laborious.activities.activities.PredictionMetadata.__init__')
 @patch('laborious.activities.activities.ModelMetrics.__init__')
 @patch('laborious.activities.activities.API.__init__')
 @patch('laborious.activities.activities.MinioRepository')
@@ -247,6 +262,7 @@ def test___init___builds_mlflow_repository_when_not_provided(
     mock_minio_repository,
     _mock_api_init,
     _mock_model_metrics_init,
+    _mock_prediction_metadata_init,
     _mock_gates_init,
     _mock_opc_init,
     _mock_mlflow_init,
@@ -310,6 +326,7 @@ def test___init___builds_mlflow_repository_when_not_provided(
 @patch('laborious.activities.activities.MLFlow.__init__')
 @patch('laborious.activities.activities.OPC.__init__')
 @patch('laborious.activities.activities.Gates.__init__')
+@patch('laborious.activities.activities.PredictionMetadata.__init__')
 @patch('laborious.activities.activities.ModelMetrics.__init__')
 @patch('laborious.activities.activities.API.__init__')
 @patch('laborious.activities.activities.MinioRepository')
@@ -319,6 +336,7 @@ def test___init___builds_mongo_repository_when_not_provided(
     mock_minio_repository,
     _mock_api_init,
     _mock_model_metrics_init,
+    mock_prediction_metadata_init,
     mock_gates_init,
     _mock_opc_init,
     _mock_mlflow_init,
@@ -373,10 +391,17 @@ def test___init___builds_mongo_repository_when_not_provided(
         notification_handler=notification_handler,
         metrics_controller=mock_metrics_controller.return_value,
     )
-    # The same repository built here is the one handed to Gates, not a second instance.
     mock_gates_init.assert_called_once_with(
         ANY,
         minio_repository=mock_minio_repository.return_value,
+        logger=logger,
+        notification_handler=notification_handler,
+        metrics_controller=mock_metrics_controller.return_value,
+    )
+    # The same repository built here is the one handed to PredictionMetadata, not a second
+    # instance.
+    mock_prediction_metadata_init.assert_called_once_with(
+        ANY,
         mongo_repository=mock_mongo_repository_cls.return_value,
         logger=logger,
         notification_handler=notification_handler,

tests/laborious/activities/test_gates.py

@@ -6,7 +6,7 @@ from sientia_do.notifications.models import NotificationLevel
 from sientia_do.observability.metrics_controller import MetricsController
 from sientia_do.observability.sientia_monitoring import SientiaMonitoring
 
-from laborious.activities.gates import Gates
+from laborious.activities.gates import Gates, input_filter_counters, input_filter_functions
 
 
 @fixture(autouse=True)
@@ -35,14 +35,8 @@ def _minio_payload(retrieve_return, status=None):
 
 
 @fixture
-def mongo() -> MagicMock:
+def gates_activity():
-    return MagicMock()
-
-
-@fixture
-def gates_activity(mongo: MagicMock):
     gates = Gates(
-        mongo_repository=mongo,
         logger=MagicMock(),
         notification_handler=MagicMock(),
         metrics_controller=MagicMock(spec=MetricsController),
@@ -67,73 +61,11 @@ metadata = {
 }
 
 
-def test_write_prediction_metadata_no_document_skips_mongo(gates_activity, mongo: MagicMock):
+def test_input_filter_counters_covers_every_input_filter_function():
-    # Act
+    # A filter added to input_filter_functions without a matching entry here would degrade to
-    result = gates_activity._write_prediction_metadata(None, metadata)
+    # counter=0 (see the `.get(..., lambda *_: 0)` fallback in input_gate) instead of crashing —
-
+    # but this test should still catch the gap before it ships (review finding, 2026-09-30).
-    # Assert
+    assert set(input_filter_counters) == set(input_filter_functions)
-    assert result is None
-    mongo.insert.assert_not_called()
-
-
-def test_write_prediction_metadata_empty_document_skips_mongo(gates_activity, mongo: MagicMock):
-    # Act
-    result = gates_activity._write_prediction_metadata({}, metadata)
-
-    # Assert
-    assert result is None
-    mongo.insert.assert_not_called()
-
-
-def test_write_prediction_metadata_not_configured_returns_none():
-    # Arrange
-    gates = Gates(
-        mongo_repository=None,
-        logger=MagicMock(),
-        notification_handler=MagicMock(),
-        metrics_controller=MagicMock(spec=MetricsController),
-    )
-    gates.error = MagicMock()
-
-    # Act
-    result = gates._write_prediction_metadata({'filter_triggered': 'EMPTY_DATA'}, metadata)
-
-    # Assert
-    assert result is None
-    gates.error.assert_not_called()
-
-
-def test_write_prediction_metadata_success_generates_id_and_inserts(
-    gates_activity, mongo: MagicMock
-):
-    # Arrange
-    document = {'filter_triggered': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 3}
-
-    # Act
-    result = gates_activity._write_prediction_metadata(document, metadata)
-
-    # Assert: a plain 24-hex-char string id (matches predictions.metadata_id's varchar(24)),
-    # not a BSON ObjectId — `insert(...)` never hands one back, so it's generated client-side.
-    assert result is not None
-    assert len(result) == 24
-    int(result, 16)  # raises ValueError if not hex
-
-    collection, inserted_document, inserted_metadata = mongo.insert.call_args.args
-    assert collection == Gates._PREDICTION_METADATA_COLLECTION
-    assert inserted_document == {**document, '_id': result}
-    assert inserted_metadata == metadata
-
-
-def test_write_prediction_metadata_insert_failure_swallowed(gates_activity, mongo: MagicMock):
-    # Arrange
-    mongo.insert.side_effect = RuntimeError('connection reset')
-
-    # Act
-    result = gates_activity._write_prediction_metadata({'filter_triggered': 'EMPTY_DATA'}, metadata)
-
-    # Assert: the prediction write must never be blocked or failed by this
-    assert result is None
-    gates_activity.error.assert_called_once()
 
 
 def test_input_gate_invalid_filter(gates_activity):
@@ -149,7 +81,7 @@ def test_input_gate_invalid_filter(gates_activity):
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.error.assert_called_once_with(
         'Filter INVALID_FILTER not found', metadata['metadata']
     )
@@ -173,7 +105,7 @@ def test_input_gate_filter_exception(mock_input_filter_functions, gates_activity
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.send_notification.assert_called_once_with(
         metadata=metadata['metadata'],
         notification_id='INTPUT_GATE_ERROR__EMPTY_DATA',
@@ -197,7 +129,7 @@ def test_input_gate_no_filters(gates_activity):
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
 
 
@@ -211,10 +143,14 @@ def test_input_gate_with_filter(gates_activity):
     }
 
     # Act
-    result = gates_activity.input_gate(input_data)
+    path_flag, confidence, comment, document = gates_activity.input_gate(input_data)
 
-    # Assert
+    # Assert: the gate only computes the document now — it never writes anywhere, and never
-    assert result == ('STOP', -1, 'Input data with bad quality')
+    # checks path_flag to decide whether to build one (that decision moved to the caller, see
+    # `PredictionMetadata.write_prediction_metadata` and `prediction_process.py`), so even a STOP
+    # resolution still gets its document built.
+    assert (path_flag, confidence, comment) == ('STOP', -1, 'Input data with bad quality')
+    assert document == {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]}
     gates_activity.debug.assert_called()
 
 
@@ -231,7 +167,12 @@ def test_input_gate_with_filter_lowercase_keys(gates_activity):
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == ('STOP', -1, 'Input data with bad quality')
+    assert result == (
+        'STOP',
+        -1,
+        'Input data with bad quality',
+        {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]},
+    )
 
 
 def test_input_gate_with_filter_capitalized_keys(gates_activity):
@@ -247,7 +188,12 @@ def test_input_gate_with_filter_capitalized_keys(gates_activity):
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == ('STOP', -1, 'Input data with bad quality')
+    assert result == (
+        'STOP',
+        -1,
+        'Input data with bad quality',
+        {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]},
+    )
 
 
 def test_input_gate_with_filter_not_caught(gates_activity):
@@ -263,10 +209,94 @@ def test_input_gate_with_filter_not_caught(gates_activity):
     result = gates_activity.input_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
 
 
+def test_input_gate_repeat_also_builds_document(gates_activity):
+    # Arrange: REPEAT used to be treated like STOP (no document at all) — it isn't anymore, since
+    # REPEAT does insert a new `predictions` row (reusing the previous row's values), so it needs
+    # a document/metadata_id just like CONTINUE (2026-10-01 redesign).
+    input_data = {
+        **metadata,
+        'filters': {'EMPTY_DATA': {'POLICY': 'REPEAT', 'CONFIG': {}}},
+        'data': _minio_payload(DataFrame({'value': []})),
+        'path_priority': ['STOP', 'CONTINUE', 'REPEAT'],
+    }
+
+    # Act
+    result = gates_activity.input_gate(input_data)
+
+    # Assert
+    assert result == (
+        'REPEAT',
+        -1,
+        'Input data with bad quality',
+        {'filters': [{'name': 'EMPTY_DATA', 'policy': 'REPEAT', 'counter': 0}]},
+    )
+
+
+def test_input_gate_continue_builds_document_with_counter(gates_activity):
+    # Arrange: two rows with a null in one of the configured variables ('x'); a third row is
+    # clean, to confirm the counter reflects only the rows that actually matched the condition.
+    input_data = {
+        **metadata,
+        'filters': {
+            'SPECIFIC_VARIABLES_NULL_VALUES': {
+                'POLICY': 'CONTINUE',
+                'CONFIG': {'variables': ['x']},
+            }
+        },
+        'data': _minio_payload(
+            DataFrame({'variable': ['x', 'x', 'y'], 'value': [None, None, None]})
+        ),
+        'path_priority': ['STOP', 'CONTINUE', 'REPEAT'],
+    }
+
+    # Act
+    path_flag, confidence, comment, document = gates_activity.input_gate(input_data)
+
+    # Assert
+    assert (path_flag, confidence, comment) == ('CONTINUE', 2, 'Input data with bad quality')
+    assert document == {
+        'filters': [{'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2}]
+    }
+
+
+def test_input_gate_multiple_filters_same_policy_lists_each(gates_activity):
+    # Arrange: force both input filters to trigger (not reachable with today's real filter
+    # conditions — EMPTY_DATA and SPECIFIC_VARIABLES_NULL_VALUES are mutually exclusive) to lock
+    # in the list-building behavior approved by Vitor (2026-09-30): one entry per filter sharing
+    # the winning policy, instead of a single joined/aggregated value.
+    with patch('laborious.activities.gates.input_filter_functions') as mock_filters:
+        mock_filters.__contains__.return_value = True
+        mock_filters.__getitem__.return_value = MagicMock(return_value=True)
+
+        input_data = {
+            **metadata,
+            'filters': {
+                'SPECIFIC_VARIABLES_NULL_VALUES': {
+                    'POLICY': 'CONTINUE',
+                    'CONFIG': {'variables': ['x']},
+                },
+                'EMPTY_DATA': {'POLICY': 'CONTINUE', 'CONFIG': {}},
+            },
+            'data': _minio_payload(DataFrame({'variable': ['x'], 'value': [None]})),
+            'path_priority': ['STOP', 'CONTINUE', 'REPEAT'],
+        }
+
+        # Act
+        path_flag, _, _, document = gates_activity.input_gate(input_data)
+
+    # Assert
+    assert path_flag == 'CONTINUE'
+    assert {f['name'] for f in document['filters']} == {
+        'SPECIFIC_VARIABLES_NULL_VALUES',
+        'EMPTY_DATA',
+    }
+    assert all(f['policy'] == 'CONTINUE' for f in document['filters'])
+
+
 def test_mlflow_response_gate_invalid_filter(gates_activity):
     # Arrange
     input_data = {

tests/laborious/activities/test_prediction_metadata.py

@@ -0,0 +1,179 @@
+from unittest.mock import MagicMock, call, patch
+
+from pytest import fixture
+from sientia_do.observability.metrics_controller import MetricsController
+from sientia_do.observability.sientia_monitoring import SientiaMonitoring
+
+from laborious.activities.prediction_metadata import PredictionMetadata
+
+
+@fixture
+def mongo() -> MagicMock:
+    return MagicMock()
+
+
+@fixture
+def prediction_metadata(mongo: MagicMock):
+    activity = PredictionMetadata(
+        mongo_repository=mongo,
+        logger=MagicMock(),
+        notification_handler=MagicMock(),
+        metrics_controller=MagicMock(spec=MetricsController),
+    )
+    activity.error = MagicMock()
+    return activity
+
+
+metadata = {
+    'metadata': {
+        'model_id': 'test_model',
+        'model_name': 'test_model',
+        'workflow_name': 'test_workflow',
+        'schema_name': 'test_schedule',
+    },
+}
+
+
+def test_write_prediction_metadata_no_documents_skips_mongo(prediction_metadata, mongo: MagicMock):
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': []})
+
+    # Assert
+    assert result is None
+    mongo.insert.assert_not_called()
+
+
+def test_write_prediction_metadata_missing_documents_key_skips_mongo(
+    prediction_metadata, mongo: MagicMock
+):
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata})
+
+    # Assert
+    assert result is None
+    mongo.insert.assert_not_called()
+
+
+def test_write_prediction_metadata_documents_with_no_filters_skips_mongo(
+    prediction_metadata, mongo: MagicMock
+):
+    # Arrange: a falsy document (None) and one with an empty `filters` list — neither contributes
+    # anything to write.
+    documents = [None, {'filters': []}]
+
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents})
+
+    # Assert
+    assert result is None
+    mongo.insert.assert_not_called()
+
+
+def test_write_prediction_metadata_not_configured_returns_none():
+    # Arrange
+    activity = PredictionMetadata(
+        mongo_repository=None,
+        logger=MagicMock(),
+        notification_handler=MagicMock(),
+        metrics_controller=MagicMock(spec=MetricsController),
+    )
+    activity.error = MagicMock()
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]}
+
+    # Act
+    result = activity.write_prediction_metadata({**metadata, 'documents': [document]})
+
+    # Assert
+    assert result is None
+    activity.error.assert_not_called()
+
+
+def test_write_prediction_metadata_success_generates_id_and_inserts(
+    prediction_metadata, mongo: MagicMock
+):
+    # Arrange
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 3}]}
+
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]})
+
+    # Assert: a plain 24-hex-char string id (matches predictions.metadata_id's varchar(24)),
+    # not a BSON ObjectId — `insert(...)` never hands one back, so it's generated client-side.
+    assert result is not None
+    assert len(result) == 24
+    int(result, 16)  # raises ValueError if not hex
+
+    collection, inserted_document, inserted_metadata = mongo.insert.call_args.args
+    assert collection == PredictionMetadata._PREDICTION_METADATA_COLLECTION
+    assert inserted_document == {'filters': document['filters'], '_id': result}
+    assert inserted_metadata == metadata['metadata']
+
+
+def test_write_prediction_metadata_combines_multiple_documents(
+    prediction_metadata, mongo: MagicMock
+):
+    # Arrange: two documents (e.g. from two different sources contributing metadata for the same
+    # prediction cycle) — their `filters` entries are flattened into one combined document.
+    documents = [
+        {'filters': [{'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}]},
+        {
+            'filters': [
+                {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2}
+            ]
+        },
+    ]
+
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': documents})
+
+    # Assert
+    assert result is not None
+    _, inserted_document, _ = mongo.insert.call_args.args
+    assert inserted_document['filters'] == [
+        {'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0},
+        {'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2},
+    ]
+
+
+def test_write_prediction_metadata_insert_failure_swallowed(prediction_metadata, mongo: MagicMock):
+    # Arrange
+    mongo.insert.side_effect = RuntimeError('connection reset')
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]}
+
+    # Act
+    result = prediction_metadata.write_prediction_metadata({**metadata, 'documents': [document]})
+
+    # Assert: the prediction write must never be blocked or failed by this
+    assert result is None
+    prediction_metadata.error.assert_called_once()
+
+
+@patch.object(SientiaMonitoring, 'shutdown')
+def test_close_disposes_mongo_repository(mock_shutdown):
+    """``PredictionMetadata.close`` should close the Mongo client and clear the reference."""
+    mongo = MagicMock()
+    activity = PredictionMetadata(
+        mongo_repository=mongo,
+        logger=MagicMock(),
+        notification_handler=MagicMock(),
+        metrics_controller=MagicMock(spec=MetricsController),
+    )
+    activity.close()
+    mongo.close.assert_called_once()
+    assert activity.mongo_repository is None
+    # Other instances collected by GC during the test also hit the patched shutdown.
+    assert mock_shutdown.call_args_list.count(call(activity)) == 1
+
+
+@patch.object(SientiaMonitoring, 'shutdown')
+def test_close_without_mongo_repository(mock_shutdown):
+    """When no Mongo repository is configured, ``close`` only shuts down monitoring."""
+    activity = PredictionMetadata(
+        mongo_repository=None,
+        logger=MagicMock(),
+        notification_handler=MagicMock(),
+        metrics_controller=MagicMock(spec=MetricsController),
+    )
+    activity.close()
+    # Other instances collected by GC during the test also hit the patched shutdown.
+    assert mock_shutdown.call_args_list.count(call(activity)) == 1

tests/laborious/workflows/subworkflows/test_format_and_export_prediction.py

@@ -326,6 +326,7 @@ async def test_run_default_path_flag(workflow_mock, format_and_export_prediction
                     'model_id': input_data['model_id'],
                     'prediction_confidence': input_data['prediction_confidence'],
                     'comment': input_data['comment'],
+                    'metadata_id': None,
                 },
                 retry_policy=ANY,
                 start_to_close_timeout=ANY,
@@ -389,6 +390,56 @@ async def test_run_default_path_flag(workflow_mock, format_and_export_prediction
     assert workflow_mock.execute_local_activity_method.call_count == 1
 
 
+@mark.asyncio
+@patch(
+    'laborious.workflows.sub_workflows.format_and_export_prediction.workflow',
+    new_callable=AsyncMock,
+)
+async def test_run_default_path_flag_passes_metadata_id(
+    workflow_mock, format_and_export_prediction
+):
+    # Arrange: metadata_id reaches format_default_prediction, which is the row that ends up in
+    # Postgres via export_data_to_postgres (CONTINUE/REPEAT path).
+    input_data = {
+        'metadata': metadata,
+        'path_flag': 'default',
+        'data': {'test': 'data'},
+        'timestamp': '2021-01-01',
+        'model_id': 1,
+        'model_name': metadata['metadata']['model_name'],
+        'prediction_confidence': 0,
+        'schema': 'test_schema',
+        'table_name': 'test_table',
+        'opc_output_config': {'test': 'config'},
+        'comment': 'test_comment',
+        'metadata_id': 'a' * 24,
+    }
+
+    workflow_mock.execute_activity_method.side_effect = [
+        (MagicMock(), MagicMock()),
+        MagicMock(),
+        MagicMock(),
+    ]
+
+    # Act
+    await format_and_export_prediction.run(input_data)
+
+    # Assert
+    workflow_mock.execute_local_activity_method.assert_any_call(
+        Activities.format_default_prediction,
+        {
+            **metadata,
+            'timestamp': input_data['timestamp'],
+            'model_id': input_data['model_id'],
+            'prediction_confidence': input_data['prediction_confidence'],
+            'comment': input_data['comment'],
+            'metadata_id': 'a' * 24,
+        },
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
+
+
 @mark.asyncio
 @patch(
     'laborious.workflows.sub_workflows.format_and_export_prediction.workflow',
@@ -638,6 +689,20 @@ async def test_run_default_path_flag_with_pi_web_api(workflow_mock, format_and_e
 
     await format_and_export_prediction.run(input_data)
 
+    workflow_mock.execute_local_activity_method.assert_any_call(
+        Activities.format_default_prediction,
+        {
+            **metadata,
+            'timestamp': input_data['timestamp'],
+            'model_id': input_data['model_id'],
+            'prediction_confidence': input_data['prediction_confidence'],
+            'comment': input_data['comment'],
+            'metadata_id': None,
+        },
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
+
     workflow_mock.execute_activity_method.assert_has_calls(
         [
             call(

tests/laborious/workflows/subworkflows/test_prediction_process.py

@@ -57,7 +57,7 @@ async def test_run(workflow_mock, prediction_process):
         MagicMock(),
     ]
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('continue', 0.95, 'Input data with bad quality'),  # input_gate
+        ('continue', 0.95, 'Input data with bad quality', None),  # input_gate
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error'),
         # mlflow_content_gate (transform)
@@ -217,7 +217,7 @@ async def test_run_stop_at_input_gate(workflow_mock, prediction_process):
 
     # Mock the activity responses
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('stop', 0.95, 'Input data with bad quality'),  # input_gate
+        ('stop', 0.95, 'Input data with bad quality', None),  # input_gate
     ]
 
     # Act
@@ -275,7 +275,7 @@ async def test_run_stop_at_first_mlflow_response_gate(workflow_mock, prediction_
         MagicMock(),
     ]
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('repeat', 0.95, 'Input data with bad quality'),  # input_gate
+        ('repeat', 0.95, 'Input data with bad quality', None),  # input_gate
         ('continue', 0.95, 'Error'),  # mlflow_response_gate (transform)
     ]
 
@@ -365,7 +365,7 @@ async def test_run_stop_at_mlflow_content_gate(workflow_mock, prediction_process
         MagicMock(),
     ]
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('continue', 0.95, 'Input data with bad quality'),  # input_gate
+        ('continue', 0.95, 'Input data with bad quality', None),  # input_gate
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error'),
         # mlflow_content_gate (transform)
@@ -476,7 +476,7 @@ async def test_run_stop_at_mlflow_last_response_gate(workflow_mock, prediction_p
         MagicMock(),
     ]
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('continue', 0.95, 'Input data with bad quality'),  # input_gate
+        ('continue', 0.95, 'Input data with bad quality', None),  # input_gate
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error'),
         # mlflow_content_gate (transform)
@@ -598,6 +598,7 @@ async def test_path_flag_handler_stop(workflow_mock, prediction_process):
     last_timestamp = '2024-01-01'
     model_name = 'test_model_name'
     model_config = {'retention': '30'}
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'STOP', 'counter': 0}]}
 
     # Act
     result = await prediction_process.path_flag_handler(
@@ -616,18 +617,22 @@ async def test_path_flag_handler_stop(workflow_mock, prediction_process):
         confidence,
         last_timestamp,
         '',
+        document,
     )
 
-    # Assert
+    # Assert: STOP never reaches a write, even though it now gets a `document` same as CONTINUE
+    # and REPEAT do — it just never has anywhere to use it.
     assert result is True
     workflow_mock.execute_local_activity_method.assert_not_called()
+    workflow_mock.execute_activity_method.assert_not_called()
     workflow_mock.execute_child_workflow.assert_not_called()
 
 
 @mark.asyncio
 @patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
 async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
-    # Arrange
+    # Arrange: REPEAT now writes metadata just like CONTINUE — it does insert a new `predictions`
+    # row (reusing the previous row's values), so it needs a `metadata_id` too.
     data = {'test': 'data'}
     path_flag = 'repeat'
     confidence = 0.95
@@ -637,6 +642,7 @@ async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
     last_timestamp = '2024-01-01'
     model_name = 'test_model_name'
     model_config = {'retention': '30'}
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'REPEAT', 'counter': 0}]}
 
     # Act
     result = await prediction_process.path_flag_handler(
@@ -655,11 +661,20 @@ async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
         confidence,
         last_timestamp,
         '',
+        document,
     )
 
     # Assert
     assert result is True
-    workflow_mock.execute_activity_method.assert_called_once_with(
+    metadata_id = workflow_mock.execute_activity_method.return_value
+    assert workflow_mock.execute_activity_method.call_count == 2
+    workflow_mock.execute_activity_method.assert_any_call(
+        Activities.write_prediction_metadata,
+        {**metadata, 'documents': [document]},
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
+    workflow_mock.execute_activity_method.assert_any_call(
         Activities.repeat_last_prediction,
         {
             **metadata,
@@ -667,6 +682,7 @@ async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
             'table_name': table_name,
             'model': model,
             'last_timestamp': last_timestamp,
+            'metadata_id': metadata_id,
         },
         retry_policy=ANY,
         start_to_close_timeout=ANY,
@@ -674,6 +690,56 @@ async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
     workflow_mock.execute_child_workflow.assert_not_called()
 
 
+@mark.asyncio
+@patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
+async def test_path_flag_handler_repeat_no_document_skips_write_activity_call(
+    workflow_mock, prediction_process
+):
+    # Arrange: no document (e.g. REPEAT triggered by a gate that doesn't build one yet — today,
+    # `mlflow_response_gate`/`mlflow_content_gate`) must not even dispatch the remote
+    # write_prediction_metadata activity — that would be a wasted Temporal round-trip on every
+    # REPEAT caused by those two gates, not a rare edge case (review finding, 2026-10-01).
+    data = {'test': 'data'}
+
+    # Act
+    result = await prediction_process.path_flag_handler(
+        data,
+        'repeat',
+        {
+            'metadata': metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'transform_table_name': 'test_transform_table',
+            'model_id': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'model_name': 'test_model_name',
+            'model_config': {'retention': '30'},
+        },
+        0.95,
+        '2024-01-01',
+        '',
+    )
+
+    # Assert
+    assert result is True
+    called_activities = [c.args[0] for c in workflow_mock.execute_activity_method.call_args_list]
+    assert Activities.write_prediction_metadata not in called_activities
+    # repeat_last_prediction still happens, just with metadata_id=None.
+    workflow_mock.execute_activity_method.assert_any_call(
+        Activities.repeat_last_prediction,
+        {
+            **metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'model': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'metadata_id': None,
+        },
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
+
+
 @mark.asyncio
 @patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
 async def test_path_flag_handler_continue(workflow_mock, prediction_process):
@@ -688,6 +754,9 @@ async def test_path_flag_handler_continue(workflow_mock, prediction_process):
     model_name = 'test_model_name'
     model_config = {'retention': '30'}
     prediction_store_policy = 'erl:1'
+    document = {
+        'filters': [{'name': 'SPECIFIC_VARIABLES_NULL_VALUES', 'policy': 'CONTINUE', 'counter': 2}]
+    }
 
     # Act
     result = await prediction_process.path_flag_handler(
@@ -713,11 +782,18 @@ async def test_path_flag_handler_continue(workflow_mock, prediction_process):
         confidence,
         last_timestamp,
         'Prediction Process',
+        document,
     )
 
     # Assert
+    metadata_id = workflow_mock.execute_activity_method.return_value
     assert result is True
-    workflow_mock.execute_activity_method.assert_not_called()
+    workflow_mock.execute_activity_method.assert_called_once_with(
+        Activities.write_prediction_metadata,
+        {**metadata, 'documents': [document]},
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
     workflow_mock.execute_child_workflow.assert_called_once_with(
         'subworkflow.format_and_export_prediction',
         {
@@ -733,6 +809,7 @@ async def test_path_flag_handler_continue(workflow_mock, prediction_process):
             'table_name': table_name,
             'transform_table_name': 'test_transform_table',
             'comment': 'Prediction Process',
+            'metadata_id': metadata_id,
             'opc_output_config': {'test': 'config'},
             'pi_web_api_output_config': {
                 'endpoint': 'https://test-pi-server.com',
@@ -745,6 +822,153 @@ async def test_path_flag_handler_continue(workflow_mock, prediction_process):
     )
 
 
+@mark.asyncio
+@patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
+async def test_path_flag_handler_continue_no_document_skips_write_activity_call(
+    workflow_mock, prediction_process
+):
+    # Arrange: same gap as the REPEAT case above (review finding, 2026-10-01) — no document must
+    # not dispatch the remote write_prediction_metadata activity.
+    data = {'test': 'data'}
+
+    # Act
+    result = await prediction_process.path_flag_handler(
+        data,
+        'continue',
+        {
+            'metadata': metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'transform_table_name': 'test_transform_table',
+            'model_id': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'model_name': 'test_model_name',
+            'model_config': {'retention': '30'},
+            'opc_output_config': {'test': 'config'},
+            'pi_web_api_output_config': {
+                'endpoint': 'https://test-pi-server.com',
+                'prediction_tags': {},
+                'confidence_tags': {},
+            },
+            'prediction_store_policy': 'erl:1',
+        },
+        0.95,
+        '2024-01-01',
+        '',
+    )
+
+    # Assert
+    assert result is True
+    workflow_mock.execute_activity_method.assert_not_called()
+    workflow_mock.execute_child_workflow.assert_called_once_with(
+        'subworkflow.format_and_export_prediction',
+        ANY,
+    )
+    _, export_input = workflow_mock.execute_child_workflow.call_args.args
+    assert export_input['metadata_id'] is None
+
+
+@mark.asyncio
+@patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
+async def test_path_flag_handler_continue_write_metadata_failure_does_not_block_prediction(
+    workflow_mock, prediction_process
+):
+    # Arrange: write_prediction_metadata's retry_policy has maximum_attempts=1 — a timeout, an
+    # unavailable worker, or any dispatch failure raises straight into the workflow with no
+    # retry. Business rule 1 says this must never block the prediction write (review finding,
+    # 2026-10-01) — confirm the CONTINUE branch still proceeds instead of letting it propagate.
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'CONTINUE', 'counter': 0}]}
+    workflow_mock.execute_activity_method.side_effect = RuntimeError('activity dispatch failed')
+
+    # Act
+    result = await prediction_process.path_flag_handler(
+        {'test': 'data'},
+        'continue',
+        {
+            'metadata': metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'transform_table_name': 'test_transform_table',
+            'model_id': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'model_name': 'test_model_name',
+            'model_config': {'retention': '30'},
+            'opc_output_config': {'test': 'config'},
+            'pi_web_api_output_config': {
+                'endpoint': 'https://test-pi-server.com',
+                'prediction_tags': {},
+                'confidence_tags': {},
+            },
+            'prediction_store_policy': 'erl:1',
+        },
+        0.95,
+        '2024-01-01',
+        '',
+        document,
+    )
+
+    # Assert: the prediction write still happens, with metadata_id=None, instead of raising.
+    assert result is True
+    workflow_mock.execute_child_workflow.assert_called_once_with(
+        'subworkflow.format_and_export_prediction',
+        ANY,
+    )
+    _, export_input = workflow_mock.execute_child_workflow.call_args.args
+    assert export_input['metadata_id'] is None
+
+
+@mark.asyncio
+@patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
+async def test_path_flag_handler_repeat_write_metadata_failure_does_not_block_prediction(
+    workflow_mock, prediction_process
+):
+    # Arrange: only the write_prediction_metadata call (first) fails — repeat_last_prediction
+    # (second) must still run normally; its own failure is a separate, intentional
+    # non-best-effort path, not what this test is about.
+    document = {'filters': [{'name': 'EMPTY_DATA', 'policy': 'REPEAT', 'counter': 0}]}
+    workflow_mock.execute_activity_method.side_effect = [
+        RuntimeError('activity dispatch failed'),
+        MagicMock(),
+    ]
+
+    # Act
+    result = await prediction_process.path_flag_handler(
+        {'test': 'data'},
+        'repeat',
+        {
+            'metadata': metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'transform_table_name': 'test_transform_table',
+            'model_id': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'model_name': 'test_model_name',
+            'model_config': {'retention': '30'},
+        },
+        0.95,
+        '2024-01-01',
+        '',
+        document,
+    )
+
+    # Assert: repeat_last_prediction still runs, with metadata_id=None, instead of raising.
+    assert result is True
+    assert workflow_mock.execute_activity_method.call_count == 2
+    workflow_mock.execute_activity_method.assert_any_call(
+        Activities.repeat_last_prediction,
+        {
+            **metadata,
+            'schema': 'test_schema',
+            'table_name': 'test_table',
+            'model': 'test_model',
+            'last_timestamp': '2024-01-01',
+            'metadata_id': None,
+        },
+        retry_policy=ANY,
+        start_to_close_timeout=ANY,
+    )
+
+
 @mark.asyncio
 @patch('laborious.workflows.sub_workflows.prediction_process.workflow', new_callable=AsyncMock)
 async def test_path_flag_handler_unknown(workflow_mock, prediction_process):
@@ -850,7 +1074,7 @@ async def test_run_with_cleanup_prefixes(workflow_mock, prediction_process):
         MagicMock(),
     ]
     workflow_mock.execute_local_activity_method.side_effect = [
-        ('continue', 0.95, 'ok'),
+        ('continue', 0.95, 'ok', None),  # input_gate
         ('continue', 0.95, ''),
         ('continue', 0.95, ''),
         ('continue', 0.95, ''),