classifyre-cli 0.4.29__tar.gz → 0.4.31__tar.gz

classifyre_cli-0.4.31/.turbo/turbo-build.log

@@ -0,0 +1,3 @@
+$ uv sync
+Resolved 265 packages in 164ms
+Checked 50 packages in 1ms

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: classifyre-cli
-Version: 0.4.29
+Version: 0.4.31
 Summary: Classifyre CLI — scan and classify unstructured data sources
 License: MIT
 Keywords: data,ingestion,metadata,pii,secrets,unstructured

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@classifyre/cli",
-  "version": "0.4.29",
+  "version": "0.4.31",
   "private": true,
   "scripts": {
     "build": "uv sync",

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "classifyre-cli"
-version = "0.4.29"
+version = "0.4.31"
 description = "Classifyre CLI — scan and classify unstructured data sources"
 readme = "README.md"
 requires-python = ">=3.12"

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/main.py

@@ -366,6 +366,12 @@ async def run_command_async(args: argparse.Namespace, recipe: dict[str, Any]) ->
                             len(all_stubs),
                         )
 
+                    # Persist the advanced AUTOMATIC sampling cursor (no-op for
+                    # other strategies, which return None). Only on the normal
+                    # completion path — a timed-out run must not advance it.
+                    if hasattr(sink, "set_sampling_cursor"):
+                        sink.set_sampling_cursor(source.current_sampling_cursor())
+
                     await sink.finish()
                     logger.info(
                         "Extraction completed: %s assets in %s batches",

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/models/generated_input.py

@@ -85,9 +85,10 @@ class SlackChannelType(StrEnum):
 
 class SamplingStrategy(StrEnum):
     """
-    Sampling strategy: RANDOM samples items randomly, LATEST prioritises the most recently modified/created items, ALL scans every item with no limit
+    Sampling strategy. AUTOMATIC (recommended default) incrementally ingests a new slice of not-yet-seen data on every run, remembering its position between runs and wrapping around to re-scan from the start once everything has been covered — eventually ingesting everything at a bounded cost per run. RANDOM samples items randomly. LATEST prioritises the most recently modified/created items. ALL scans every item with no limit.
     """
 
+    AUTOMATIC = 'AUTOMATIC'
     RANDOM = 'RANDOM'
     LATEST = 'LATEST'
     ALL = 'ALL'
@@ -95,7 +96,7 @@ class SamplingStrategy(StrEnum):
 
 class SamplingConfig(BaseModel):
     """
-    Controls how content is extracted from each source. For tabular sources rows_per_page controls both sample size for RANDOM/LATEST and pagination batch size for ALL.
+    Controls how content is extracted from each source. For tabular sources rows_per_page controls both sample size for AUTOMATIC/RANDOM/LATEST and pagination batch size for ALL.
     """
 
     model_config = ConfigDict(
@@ -124,7 +125,7 @@ class SamplingConfig(BaseModel):
     )
     rows_per_page: int | None = Field(
         100,
-        description='Tabular sources only. Number of rows per sample (RANDOM/LATEST) or per pagination batch (ALL). Controls memory usage during large table scans.',
+        description='Tabular sources only. Number of rows per sample (AUTOMATIC/RANDOM/LATEST) or per pagination batch (ALL). For AUTOMATIC this is the size of the incremental slice ingested each run. Controls memory usage during large table scans.',
         ge=10,
         le=10000,
     )

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/outputs/rest.py

@@ -127,6 +127,11 @@ class FinalizeIngestRunRequest(BaseModel):
 
     runner_id: str = Field(serialization_alias="runnerId")
     seen_hashes: list[str] = Field(serialization_alias="seenHashes")
+    # AUTOMATIC sampling cursor to persist on the source for the next run.
+    # Omitted (None) for other strategies so the stored cursor is left untouched.
+    sampling_cursor: dict[str, Any] | None = Field(
+        None, serialization_alias="samplingCursor"
+    )
 
 
 class UpdateRunnerStatusRequest(BaseModel):
@@ -165,6 +170,11 @@ class RestOutputSink:
         self.session.mount("https://", adapter)
         self._runner_id = context.runner_id
         self._seen_hashes: set[str] = set()
+        self._sampling_cursor: dict[str, Any] | None = None
+
+    def set_sampling_cursor(self, cursor: dict[str, Any] | None) -> None:
+        """Record the AUTOMATIC sampling cursor to persist on finalize."""
+        self._sampling_cursor = cursor
 
     async def start(self) -> None:
         if not self.context.source_id:
@@ -244,11 +254,12 @@ class RestOutputSink:
         payload = FinalizeIngestRunRequest(
             runner_id=runner_id,
             seen_hashes=sorted(self._seen_hashes),
+            sampling_cursor=self._sampling_cursor,
         )
         self._request_json(
             "POST",
             f"/sources/{source_id}/assets/finalize",
-            payload.model_dump(mode="json", by_alias=True),
+            payload.model_dump(mode="json", by_alias=True, exclude_none=True),
         )
 
         status_payload = UpdateRunnerStatusRequest(status="COMPLETED")

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/base.py

@@ -1,7 +1,11 @@
+import base64
+import json
+import logging
 import os
+import threading
 from abc import ABC, abstractmethod
 from collections.abc import AsyncGenerator, Generator
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Any, TypeVar
 
 from ..models.generated_single_asset_scan_results import DetectionResult, SingleAssetScanResults
 from ..outputs.rest import IngestEdge
@@ -12,6 +16,10 @@ from ..utils.hashing import calculate_checksum, normalize_http_url
 from ..utils.validation import validate_output
 from .recipe_normalizer import normalize_source_recipe
 
+logger = logging.getLogger(__name__)
+
+_T = TypeVar("_T")
+
 
 class BaseSource(ABC):
     """
@@ -26,6 +34,10 @@ class BaseSource(ABC):
     # Default batch size for streaming asset results
     BATCH_SIZE: int = 50
     HAS_SUCCESSFUL_RUN_ENV = "CLASSIFYRE_SOURCE_HAS_SUCCESSFUL_RUN"
+    # The API injects the saved AUTOMATIC sampling cursor here (base64-encoded
+    # JSON) before launching the CLI job. The recipe itself cannot carry it
+    # because every source schema sets ``additionalProperties: false``.
+    SAMPLING_CURSOR_ENV = "CLASSIFYRE_SAMPLING_CURSOR"
 
     def __init__(
         self,
@@ -42,6 +54,11 @@ class BaseSource(ABC):
             runner_id: Optional runner ID (for API runs)
         """
         normalized_recipe = normalize_source_recipe(recipe, recipe.get("type"))
+        # Cursor carried over from the previous run (AUTOMATIC strategy). Read
+        # before the override hook so subclasses can consult it there if needed.
+        self._sampling_cursor: dict[str, Any] = self._load_sampling_cursor()
+        self._next_sampling_cursor: dict[str, Any] | None = None
+        self._sampling_cursor_lock = threading.Lock()
         self._apply_initial_sampling_override(normalized_recipe)
         recipe.clear()
         recipe.update(normalized_recipe)
@@ -55,6 +72,108 @@ class BaseSource(ABC):
     def _apply_initial_sampling_override(self, recipe: dict[str, Any]) -> None:
         pass
 
+    # ── AUTOMATIC sampling cursor ────────────────────────────────────────
+    #
+    # AUTOMATIC sampling keeps a small, opaque, source-defined cursor in the
+    # API between runs. Each run reads the prior cursor (``sampling_cursor``),
+    # ingests the next slice of not-yet-seen data, then records the advanced
+    # cursor (``set_next_sampling_cursor``). The output sink persists it back to
+    # the API on finalize via ``current_sampling_cursor``. When a source has
+    # ingested everything it should reset the cursor so the next run wraps
+    # around and re-ingests from the start (data is not stale).
+
+    def _load_sampling_cursor(self) -> dict[str, Any]:
+        raw = os.environ.get(self.SAMPLING_CURSOR_ENV)
+        if not raw:
+            return {}
+        try:
+            decoded = base64.b64decode(raw).decode("utf-8")
+            data = json.loads(decoded)
+        except Exception as exc:
+            logger.warning("Ignoring malformed %s: %s", self.SAMPLING_CURSOR_ENV, exc)
+            return {}
+        return data if isinstance(data, dict) else {}
+
+    def sampling_cursor(self) -> dict[str, Any]:
+        """Return the cursor saved by the previous run (empty on first run)."""
+        return self._sampling_cursor
+
+    def set_next_sampling_cursor(self, cursor: dict[str, Any]) -> None:
+        """Record the advanced cursor to persist at the end of this run."""
+        self._next_sampling_cursor = cursor
+
+    def current_sampling_cursor(self) -> dict[str, Any] | None:
+        """Cursor to persist for the next run, or None to leave it unchanged.
+
+        Returns None unless this run advanced the cursor (i.e. AUTOMATIC
+        sampling actually ran), so non-AUTOMATIC runs never touch the stored
+        cursor.
+        """
+        return self._next_sampling_cursor
+
+    def sampling_window_size(self, default: int = 100) -> int:
+        """The per-run AUTOMATIC slice size (``rows_per_page``)."""
+        config = getattr(self, "config", None)
+        sampling = getattr(config, "sampling", None) if config is not None else None
+        size = getattr(sampling, "rows_per_page", None)
+        try:
+            return int(size) if size else default
+        except (TypeError, ValueError):
+            return default
+
+    def _record_cursor_key(self, key: str, value: Any) -> None:
+        """Thread-safely set ``key`` in the cursor to persist for the next run."""
+        with self._sampling_cursor_lock:
+            nxt = self._next_sampling_cursor if isinstance(self._next_sampling_cursor, dict) else {}
+            nxt = {**nxt, key: value}
+            self._next_sampling_cursor = nxt
+
+    def automatic_offset(self, key: str) -> int:
+        """Return the saved offset for a keyed AUTOMATIC DB cursor (0 on first run)."""
+        saved = self._sampling_cursor.get(key)
+        return saved if isinstance(saved, int) and saved >= 0 else 0
+
+    def record_automatic_offset(
+        self, key: str, *, prev_offset: int, fetched: int
+    ) -> None:
+        """Advance a keyed offset cursor; wrap to 0 once a page underfills.
+
+        Used by sources that page rows directly from the backing store
+        (``skip``/``OFFSET``) rather than materialising a full list.
+        """
+        size = self.sampling_window_size()
+        next_offset = 0 if fetched < size else prev_offset + fetched
+        self._record_cursor_key(key, next_offset)
+
+    def automatic_window(self, items: list[_T], *, key: str = "items") -> list[_T]:
+        """Return the next AUTOMATIC slice of a stably-ordered in-memory list.
+
+        Non-tabular sources fetch a list of item references, then call this to
+        ingest only the next ``rows_per_page`` window. A per-``key`` offset is
+        remembered between runs and wraps back to the start once the list has
+        been fully covered (data is not stale, so re-ingesting is desired).
+
+        Callers must pass the items in a **stable order** across runs (e.g. by
+        id or timestamp) so the cursor stays meaningful.
+        """
+        total = len(items)
+        if total == 0:
+            return []
+
+        saved = self._sampling_cursor.get(key)
+        offset = saved if isinstance(saved, int) and 0 <= saved < total else 0
+
+        size = self.sampling_window_size()
+        window = items[offset : offset + size]
+
+        next_offset = offset + len(window)
+        if next_offset >= total:
+            next_offset = 0  # wrap around on the next run
+
+        self._record_cursor_key(key, next_offset)
+
+        return window
+
     @staticmethod
     def _read_bool_env(name: str) -> bool | None:
         raw = os.environ.get(name)

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/confluence/source.py

@@ -242,11 +242,24 @@ class ConfluenceSource(BaseSource):
                 params["labels"] = ",".join(str(v) for v in spaces_filter.labels)
         return self.client.iter_confluence_results("/wiki/api/v2/spaces", params=params)
 
+    def _sorted_page_refs(self, refs: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        return sorted(
+            refs,
+            key=lambda ref: parse_datetime(
+                str(ref.get("version_created_at") or ref.get("created_at") or "")
+            ),
+            reverse=True,
+        )
+
     def _sample_page_refs(self, refs: list[dict[str, Any]]) -> list[dict[str, Any]]:
         sampling = self.config.sampling
         if sampling.strategy == SamplingStrategy.ALL:
             return refs
 
+        if sampling.strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            return self.automatic_window(self._sorted_page_refs(refs), key="pages")
+
         limit = int(sampling.rows_per_page or 100)
         if limit >= len(refs):
             return refs
@@ -254,14 +267,7 @@ class ConfluenceSource(BaseSource):
         if sampling.strategy == SamplingStrategy.RANDOM:
             return deterministic_sample(refs, limit)
 
-        refs_sorted = sorted(
-            refs,
-            key=lambda ref: parse_datetime(
-                str(ref.get("version_created_at") or ref.get("created_at") or "")
-            ),
-            reverse=True,
-        )
-        return refs_sorted[:limit]
+        return self._sorted_page_refs(refs)[:limit]
 
     def _extract_page_assets(self, ref: dict[str, Any]) -> list[SingleAssetScanResults]:
         page_id = str(ref["page_id"])

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/databricks/source.py

@@ -423,6 +423,11 @@ class DatabricksSource(BaseTabularSource):
             return value.isoformat()
         return str(value)
 
+    def _automatic_supports_keyset(self) -> bool:
+        # Databricks builds inline (parameter-less) queries; AUTOMATIC uses OFFSET
+        # paging through _fetch_one_page rather than keyset WHERE clauses.
+        return False
+
     # ── Databricks pagination (inline LIMIT/OFFSET) ──────────────────────
 
     def _fetch_one_page(

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/email/source.py

@@ -195,6 +195,11 @@ class EmailSource(BaseSource):
         total = 0
 
         try:
+            if strategy == SamplingStrategy.AUTOMATIC:
+                async for batch in self._extract_automatic(mod, criteria):
+                    yield batch
+                return
+
             for folder in self.folders:
                 if self._aborted or (limit is not None and total >= limit):
                     break
@@ -236,6 +241,63 @@ class EmailSource(BaseSource):
         finally:
             logger.info("Extracted %s email messages", total)
 
+    async def _extract_automatic(
+        self, mod: Any, criteria: Any
+    ) -> AsyncGenerator[list[SingleAssetScanResults], None]:
+        """AUTOMATIC sampling: page through each folder's messages by UID.
+
+        Listing UIDs is cheap (no body fetch); we window the UID list (newest
+        first) so each run ingests the next ``rows_per_page`` slice per folder
+        and wraps around once the folder has been fully covered.
+        """
+        pending: list[SingleAssetScanResults] = []
+        total = 0
+        for folder in self.folders:
+            if self._aborted:
+                break
+            try:
+                self._mailbox.folder.set(folder)
+            except Exception as e:
+                logger.warning("Skipping folder %s: %s", folder, e)
+                continue
+
+            try:
+                uid_ints = sorted((int(u) for u in self._mailbox.uids(criteria)), reverse=True)
+            except Exception as e:
+                logger.warning("Could not list UIDs for folder %s: %s", folder, e)
+                continue
+            if not uid_ints:
+                continue
+
+            window = self.automatic_window([str(u) for u in uid_ints], key=f"folder:{folder}")
+            if not window:
+                continue
+
+            for msg in self._mailbox.fetch(
+                mod.AND(uid=",".join(window)),
+                mark_seen=False,
+                bulk=self.BATCH_SIZE,
+            ):
+                if self._aborted:
+                    break
+                try:
+                    assets = self._message_to_assets(msg, folder)
+                except Exception as e:
+                    logger.error(
+                        "Failed to transform message uid=%s: %s", getattr(msg, "uid", "?"), e
+                    )
+                    continue
+                for asset in assets:
+                    pending.append(asset)
+                    while len(pending) >= self.BATCH_SIZE:
+                        yield pending[: self.BATCH_SIZE]
+                        pending = pending[self.BATCH_SIZE :]
+                total += 1
+
+        if pending:
+            yield pending
+        logger.info("Extracted %s email messages (AUTOMATIC)", total)
+
     def _message_to_assets(self, msg: Any, folder: str) -> list[SingleAssetScanResults]:
         message_id = self._message_id(msg, folder)
         email_hash = self.generate_hash_id(message_id)

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/jira/source.py

@@ -193,11 +193,28 @@ class JiraSource(BaseSource):
             return f"{query} ORDER BY updated DESC"
         return query
 
+    def _sorted_issues(self, issues: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        return sorted(
+            issues,
+            key=lambda issue: parse_datetime(
+                str(
+                    issue.get("fields", {}).get("updated")
+                    if isinstance(issue.get("fields"), dict)
+                    else ""
+                )
+            ),
+            reverse=True,
+        )
+
     def _sample_issues(self, issues: list[dict[str, Any]]) -> list[dict[str, Any]]:
         sampling = self.config.sampling
         if sampling.strategy == SamplingStrategy.ALL:
             return issues
 
+        if sampling.strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            return self.automatic_window(self._sorted_issues(issues), key="issues")
+
         limit = int(sampling.rows_per_page or 100)
         if limit >= len(issues):
             return issues
@@ -205,18 +222,7 @@ class JiraSource(BaseSource):
         if sampling.strategy == SamplingStrategy.RANDOM:
             return deterministic_sample(issues, limit)
 
-        sorted_issues = sorted(
-            issues,
-            key=lambda issue: parse_datetime(
-                str(
-                    issue.get("fields", {}).get("updated")
-                    if isinstance(issue.get("fields"), dict)
-                    else ""
-                )
-            ),
-            reverse=True,
-        )
-        return sorted_issues[:limit]
+        return self._sorted_issues(issues)[:limit]
 
     def _extract_issue_assets(self, issue: dict[str, Any]) -> list[SingleAssetScanResults]:
         fields = issue.get("fields", {})

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/mongodb/source.py

@@ -407,6 +407,14 @@ class MongoDBSource(BaseSource):
         if strategy == SamplingStrategy.ALL:
             return list(collection.find({}).limit(rows_per_page))
 
+        if strategy == SamplingStrategy.AUTOMATIC:
+            # Page forward through the collection each run; wrap when exhausted.
+            key = f"collection:{collection_ref.database}.{collection_ref.collection}"
+            offset = self.automatic_offset(key)
+            documents = list(collection.find({}).skip(offset).limit(rows_per_page))
+            self.record_automatic_offset(key, prev_offset=offset, fetched=len(documents))
+            return documents
+
         if strategy == SamplingStrategy.RANDOM:
             return self._sample_random_documents(collection, rows_per_page)
 

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/neo4j/source.py

@@ -392,6 +392,14 @@ class Neo4jSource(BaseSource):
         strategy = sampling.strategy
         rows = int(sampling.rows_per_page or 100)
 
+        if strategy == SamplingStrategy.AUTOMATIC:
+            # Page forward through this label's nodes each run; wrap when exhausted.
+            key = f"label:{ref.label}"
+            offset = self.automatic_offset(key)
+            page = self._fetch_nodes_page(ref, skip=offset, limit=rows)
+            self.record_automatic_offset(key, prev_offset=offset, fetched=len(page))
+            return page
+
         if strategy == SamplingStrategy.RANDOM:
             cypher = (
                 f"MATCH (n:{_escape_label(ref.label)}) "

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/notion/source.py

@@ -338,11 +338,22 @@ class NotionSource(BaseSource):
             "edited": obj.get("last_edited_time") or obj.get("created_time") or "",
         }
 
+    def _sorted_refs(self, refs: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        return sorted(
+            refs,
+            key=lambda ref: parse_datetime(str(ref.get("edited") or "")),
+            reverse=True,
+        )
+
     def _sample_refs(self, refs: list[dict[str, Any]]) -> list[dict[str, Any]]:
         sampling = self.config.sampling
         if sampling.strategy == SamplingStrategy.ALL:
             return refs
 
+        if sampling.strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            return self.automatic_window(self._sorted_refs(refs), key="refs")
+
         limit = int(sampling.rows_per_page or 100)
         if limit >= len(refs):
             return refs
@@ -350,12 +361,7 @@ class NotionSource(BaseSource):
         if sampling.strategy == SamplingStrategy.RANDOM:
             return deterministic_sample(refs, limit)
 
-        refs_sorted = sorted(
-            refs,
-            key=lambda ref: parse_datetime(str(ref.get("edited") or "")),
-            reverse=True,
-        )
-        return refs_sorted[:limit]
+        return self._sorted_refs(refs)[:limit]
 
     # ------------------------------------------------------------------- pages
     def _extract_page_assets(self, page: dict[str, Any]) -> list[SingleAssetScanResults]:

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/object_storage/base.py

@@ -271,6 +271,11 @@ class ObjectStorageSourceBase(BaseSource, ABC):
 
         materialized = list(refs)
 
+        if strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            materialized.sort(key=lambda ref: ref.last_modified, reverse=True)
+            return self.automatic_window(materialized, key="objects")
+
         if strategy == SamplingStrategy.RANDOM:
             if limit >= len(materialized):
                 return materialized

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/powerbi/source.py

@@ -569,11 +569,28 @@ class PowerBISource(BaseSource):
                 return parsed
         return None
 
+    def _ordered_refs_for_automatic(
+        self, refs: list[PowerBIAssetRef], order_field: str
+    ) -> list[PowerBIAssetRef]:
+        values = [self._sampling_sort_datetime(ref, order_field) for ref in refs]
+        scored: list[tuple[bool, datetime, PowerBIAssetRef]] = []
+        for ref, parsed in zip(refs, values, strict=False):
+            effective = parsed or ref.updated_at
+            scored.append((parsed is not None, effective, ref))
+        scored.sort(key=lambda item: (item[0], item[1]), reverse=True)
+        return [item[2] for item in scored]
+
     def _sample_refs(self, refs: list[PowerBIAssetRef]) -> list[PowerBIAssetRef]:
         sampling = self._sampling()
         if sampling.strategy == SamplingStrategy.ALL:
             return refs
 
+        if sampling.strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            order_field = sampling.order_by_column or "modifiedDateTime"
+            ordered = self._ordered_refs_for_automatic(refs, order_field)
+            return self.automatic_window(ordered, key="refs")
+
         if sampling.strategy == SamplingStrategy.RANDOM:
             limit = int(sampling.rows_per_page or 100)
             if limit >= len(refs):

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/recipe_normalizer.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from copy import deepcopy
 from typing import Any
 
-_VALID_SAMPLING_STRATEGIES = {"RANDOM", "LATEST", "ALL"}
+_VALID_SAMPLING_STRATEGIES = {"AUTOMATIC", "RANDOM", "LATEST", "ALL"}
 
 
 def _as_dict(value: Any) -> dict[str, Any]:
@@ -130,7 +130,7 @@ def normalize_source_recipe(
         _normalize_sampling_strategy(sampling.get("strategy")),
         _normalize_sampling_strategy(optional_sampling.get("strategy")),
         _normalize_sampling_strategy(optional_sampling.get("mode")),
-        "RANDOM",
+        "AUTOMATIC",
     )
 
     sampling["strategy"] = strategy

{classifyre_cli-0.4.29 → classifyre_cli-0.4.31}/src/sources/servicedesk/source.py

@@ -204,6 +204,11 @@ class ServiceDeskSource(BaseSource):
         if sampling.strategy == SamplingStrategy.ALL:
             return requests
 
+        if sampling.strategy == SamplingStrategy.AUTOMATIC:
+            # Newest-first stable order; window advances each run and wraps around.
+            sorted_requests = sorted(requests, key=self._request_sort_timestamp, reverse=True)
+            return self.automatic_window(sorted_requests, key="requests")
+
         limit = int(sampling.rows_per_page or 100)
         if limit >= len(requests):
             return requests