diff --git a/.env.example b/.env.example
index e84fb31..65cb8c8 100644
--- a/.env.example
+++ b/.env.example
@@ -2,5 +2,6 @@
 export COLLECTION=conservation-area
 export DATASET=conservation-area
 export DATASET_NAME=conservation-area
-export TRANSFORMATION_OFFSET=1600
+export TRANSFORMATION_OFFSET=1200
 export TRANSFORMATION_LIMIT=200
+# export REPROCESS=''
diff --git a/.gitignore b/.gitignore
index a960cc6..6a7b085 100644
--- a/.gitignore
+++ b/.gitignore
@@ -16,6 +16,8 @@
 /expectations
 /task/metadata.json
 /metadata.json
+/task/state.json
+/state.json
 /task/specification
 /specification
 /task/issue
diff --git a/bin/download_resources.py b/bin/download_resources.py
index 0f974ba..1ca1322 100644
--- a/bin/download_resources.py
+++ b/bin/download_resources.py
@@ -13,7 +13,7 @@
 logger = logging.getLogger(__name__)
 
 
-def download_resources(collection, collection_dir: str, bucket=None, base_url=None, collection_name=None, dataset=None, transformaiton_offset=None, transformation_limit=None, dataset_resource_dir="var/dataset-resource/", pipeline_dir="pipeline/", specification_dir="specification/", reprocess=False, max_threads=4) -> None:
+def download_resources(collection, collection_dir: str, bucket=None, base_url=None, collection_name=None, dataset=None, transformaiton_offset=None, transformation_limit=None, dataset_resource_dir="var/dataset-resource/", pipeline_dir="pipeline/", specification_dir="specification/", reprocess=False, max_threads=4, state_path=None) -> None:
     """Download resources for a collection.
 
     Uses the same selection logic as transform_resources (via select_resources_to_process)
@@ -59,6 +59,7 @@ def download_resources(collection, collection_dir: str, bucket=None, base_url=No
         offset=transformaiton_offset,
         limit=transformation_limit,
         reprocess=reprocess,
+        state_path=state_path,
     )
 
     # Extract unique resources to download (a resource only needs to be downloaded once)
@@ -154,6 +155,12 @@ def download_resources(collection, collection_dir: str, bucket=None, base_url=No
     default="specification/",
     help="Path to the specification directory (used for specification hash)"
 )
+@click.option(
+    "--state-path",
+    default=None,
+    type=click.Path(exists=True),
+    help="Path to state.json for stable ordered resource list"
+)
 @click.option(
     "--reprocess",
     is_flag=True,
@@ -176,7 +183,7 @@ def download_resources(collection, collection_dir: str, bucket=None, base_url=No
     is_flag=True,
     help="Enable debug logging"
 )
-def run_command(collection_dir, bucket, base_url, collection_name, dataset, offset, limit, dataset_resource_dir, pipeline_dir, specification_dir, reprocess, max_threads, quiet, debug):
+def run_command(collection_dir, bucket, base_url, collection_name, dataset, offset, limit, dataset_resource_dir, pipeline_dir, specification_dir, state_path, reprocess, max_threads, quiet, debug):
     """Download resources for a collection from S3 or HTTP(S) URLs.
 
     Either --bucket or --base-url must be provided.
@@ -218,6 +225,7 @@ def run_command(collection_dir, bucket, base_url, collection_name, dataset, offs
             specification_dir=specification_dir,
             reprocess=reprocess,
             max_threads=max_threads,
+            state_path=state_path,
         )
         click.echo("Download complete!")
     except ValueError as e:
diff --git a/bin/transform.sh b/bin/transform.sh
index edd766b..6d2fc3c 100755
--- a/bin/transform.sh
+++ b/bin/transform.sh
@@ -38,10 +38,34 @@ make init
 echo "Step 3: Building collection database..."
 make collection
 
-# Step 4: Download dataset resource logs (used to skip already up-to-date resources)
+# Step 4: Download state.json (used for stable batch ordering) and dataset resource logs
+# Use local state.json if it exists, allowing manual edits
+STATE_PATH=${STATE_PATH:-"state.json"}
+
+if [ -f "$STATE_PATH" ]; then
+    echo "Step 4a: Using existing state.json at $STATE_PATH"
+else
+    echo "Step 4a: Downloading state.json..."
+    if [ -n "$COLLECTION_DATASET_BUCKET_NAME" ]; then
+        aws s3 cp "s3://${COLLECTION_DATASET_BUCKET_NAME}/${COLLECTION_NAME}-collection/state.json" "$STATE_PATH"
+    elif [ -n "$DATASTORE_URL" ]; then
+        base=$(echo "$DATASTORE_URL" | sed 's:/*$::')
+        curl --fail -o "$STATE_PATH" "${base}/${COLLECTION_NAME}-collection/state.json"
+    else
+        echo "Error: Either COLLECTION_DATASET_BUCKET_NAME or DATASTORE_URL must be set"
+        exit 1
+    fi
+
+    if [ ! -f "$STATE_PATH" ]; then
+        echo "Error: Failed to download state.json"
+        exit 1
+    fi
+fi
+
+# Download dataset resource logs (used to skip already up-to-date resources within a batch)
 # Skipped when REPROCESS is set - logs will be freshly written by the transform step
 if [ -z "$REPROCESS" ]; then
-    echo "Step 4: Downloading dataset resource logs..."
+    echo "Step 4b: Downloading dataset resource logs..."
 
     DATASET_RESOURCE_CMD="python bin/download_dataset_resource.py --collection-dir $COLLECTION_DIR --collection-name $COLLECTION_NAME --dataset-resource-dir $DATASET_RESOURCE_DIR"
 
@@ -61,14 +85,14 @@ if [ -z "$REPROCESS" ]; then
     echo "Command: $DATASET_RESOURCE_CMD"
     eval $DATASET_RESOURCE_CMD
 else
-    echo "Step 4: Skipping dataset resource log download - REPROCESS is set, logs will be written fresh"
+    echo "Step 4b: Skipping dataset resource log download - REPROCESS is set, logs will be written fresh"
 fi
 
 # Step 5: Download resources
 echo "Step 5: Downloading resources..."
 
 # Build the download command
-DOWNLOAD_CMD="python bin/download_resources.py --collection-dir $COLLECTION_DIR"
+DOWNLOAD_CMD="python bin/download_resources.py --collection-dir $COLLECTION_DIR --state-path $STATE_PATH"
 
 # Add bucket or base URL (bucket takes precedence, matching makefile convention)
 if [ -n "$COLLECTION_DATASET_BUCKET_NAME" ]; then
@@ -111,7 +135,7 @@ echo "Resources downloaded successfully"
 # Step 6: Transform resources using Python multiprocessing
 echo "Step 6: Transforming resources..."
 
-TRANSFORM_CMD="python bin/transform_resources.py --collection-dir $COLLECTION_DIR"
+TRANSFORM_CMD="python bin/transform_resources.py --collection-dir $COLLECTION_DIR --state-path $STATE_PATH"
 
 # Add directory parameters
 TRANSFORM_CMD="$TRANSFORM_CMD --pipeline-dir $PIPELINE_DIR"
diff --git a/bin/transform_resources.py b/bin/transform_resources.py
index de3f93f..ccccbac 100755
--- a/bin/transform_resources.py
+++ b/bin/transform_resources.py
@@ -87,6 +87,12 @@
     default="specification/",
     help="Path to the specification directory"
 )
+@click.option(
+    "--state-path",
+    default=None,
+    type=click.Path(exists=True),
+    help="Path to state.json for stable ordered resource list"
+)
 @click.option(
     "--reprocess",
     is_flag=True,
@@ -119,6 +125,7 @@ def run_command(
     offset,
     limit,
     max_workers,
+    state_path,
     reprocess,
     quiet,
     debug
@@ -148,6 +155,7 @@ def run_command(
             offset=offset,
             limit=limit,
             max_workers=max_workers,
+            state_path=state_path,
             reprocess=reprocess,
         )
 
diff --git a/src/collection_task/filtering.py b/src/collection_task/filtering.py
index 8b78057..faf9bc7 100644
--- a/src/collection_task/filtering.py
+++ b/src/collection_task/filtering.py
@@ -1,5 +1,6 @@
 """Filtering and resource management functions for collection tasks."""
 
+import json
 import logging
 from typing import Dict, List, Optional, Tuple
 
@@ -89,6 +90,31 @@ def apply_offset_and_limit(
     return dataset_resource_pairs
 
 
+def load_state_resources(state_path: str, dataset: str) -> List[Tuple[str, str]]:
+    """Load the ordered resource list for a dataset from a state.json file.
+
+    Args:
+        state_path: Path to state.json
+        dataset: Dataset name to load resources for
+
+    Returns:
+        List of (dataset, resource) tuples in the order defined by state.json
+
+    Raises:
+        FileNotFoundError: If state_path does not exist
+        KeyError: If state.json does not contain 'transform_resources' or dataset is not found
+    """
+    with open(state_path) as f:
+        state = json.load(f)
+
+    transform_resources = state["transform_resources"]
+
+    if dataset not in transform_resources:
+        raise KeyError(f"Dataset '{dataset}' not found in state.json transform_resources")
+
+    return [(dataset, resource) for resource in transform_resources[dataset]]
+
+
 def select_resources_to_process(
     dataset_resource_map: Dict[str, List[str]],
     dataset_resource_dir: str,
@@ -98,16 +124,18 @@ def select_resources_to_process(
     offset: Optional[int] = None,
     limit: Optional[int] = None,
     reprocess: bool = False,
+    state_path: Optional[str] = None,
 ) -> List[Tuple[str, str]]:
     """Select the (dataset, resource) pairs to process or download.
 
-    Applies the same logic in both download_resources and transform_resources
-    so the two steps always operate on the same set of resources:
+    If state_path is provided, the stable ordered list from state.json is used
+    as the base for offset/limit. Otherwise falls back to building from
+    dataset_resource_map. The skip check (resource_needs_processing) is always
+    applied after offset/limit so batch boundaries remain stable.
 
-    1. Build the full list of (dataset, resource) pairs
-    2. If not reprocess: filter to only those whose dataset-resource log is
-       out-of-date (different code version, config hash, or spec hash)
-    3. Apply offset/limit to the resulting list
+    1. Build the ordered list (from state.json if available, else dataset_resource_map)
+    2. Apply offset/limit to get a stable batch slice
+    3. If not reprocess: filter the slice to only those needing processing
 
     Args:
         dataset_resource_map: Dictionary mapping datasets to lists of resources
@@ -115,14 +143,23 @@ def select_resources_to_process(
         pipeline_dir: Path to pipeline config directory (used for config hash)
         specification_dir: Path to specification directory (used for spec hash)
         dataset: Optional dataset name to filter to
-        offset: Optional offset into the final list
+        offset: Optional offset into the stable ordered list
         limit: Optional maximum number of pairs to return
         reprocess: If True, skip the dataset-resource log check
+        state_path: Optional path to state.json for stable ordered resource list
 
     Returns:
         List of (dataset, resource) tuples to process
     """
-    pairs = build_dataset_resource_pairs(dataset_resource_map, dataset=dataset)
+    if state_path:
+        if not dataset:
+            raise ValueError("dataset is required when state_path is provided")
+        pairs = load_state_resources(state_path, dataset=dataset)
+    else:
+        pairs = build_dataset_resource_pairs(dataset_resource_map, dataset=dataset)
+
+    # Apply offset/limit to stable list first so batch boundaries never shift
+    pairs = apply_offset_and_limit(pairs, offset=offset, limit=limit, dataset=dataset)
 
     if not reprocess:
         config_hash = hash_directory(pipeline_dir)
@@ -141,7 +178,7 @@ def select_resources_to_process(
             f"{len(pairs)} to process"
         )
 
-    return apply_offset_and_limit(pairs, offset=offset, limit=limit, dataset=dataset)
+    return pairs
 
 
 def build_retired_resources_set(old_resource_entries: List[Dict]) -> set:
diff --git a/src/collection_task/transform.py b/src/collection_task/transform.py
index e17da43..c7f1bd5 100644
--- a/src/collection_task/transform.py
+++ b/src/collection_task/transform.py
@@ -114,6 +114,7 @@ def process_resources(
     limit=None,
     max_workers=None,
     reprocess=False,
+    state_path=None,
 ):
     """Process resources using multiprocessing.
 
@@ -149,6 +150,7 @@ def process_resources(
         offset=offset,
         limit=limit,
         reprocess=reprocess,
+        state_path=state_path,
     )
 
     tasks = []