diff --git a/requirements.txt b/requirements.txt
index b19dce6..3422064 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,5 +1,5 @@
 pandas>=2.0,<3.0
-pyreadstat>=1.2,<1.3
+pyreadstat>=1.2,<2.0
 numpy>=2.1,<3.0
 pyyaml>=6.0,<7.0
 psycopg2-binary>=2.9,<3.0
diff --git a/utils/s3_download.py b/utils/s3_download.py
new file mode 100644
index 0000000..c987db5
--- /dev/null
+++ b/utils/s3_download.py
@@ -0,0 +1,728 @@
+"""S3-source counterpart to ``generic_loader/load_folder.py``.
+
+Reads a YAML config that points at an S3 bucket + prefix, lists every object
+under that prefix recursively, groups objects into *clusters* using the same
+explicit-pattern + auto-detect rules as ``load_folder.py``, and downloads each
+cluster's files into its own subfolder under a local destination root.
+
+-------------------------------------------------------------------------------
+USAGE
+-------------------------------------------------------------------------------
+
+1. YAML config
+--------------
+::
+
+    bucket: my-bucket               # required
+    prefix: census/2020/raw/        # required; recursive scan under it
+    local_folder: ./downloads       # required; one subfolder per cluster
+    aws_profile: default            # optional; default boto3 chain if omitted
+
+    auto_detect: true               # optional; default true
+    extensions:                     # optional; default sas7bdat/xpt/xport
+      - .sas7bdat
+    on_exists: skip                 # optional; skip | overwrite | error
+    concurrency: 4                  # optional; default 4
+
+    clusters:
+      - pattern: '^group_a\\d+\\.sas7bdat$'
+        name: group_a
+      - pattern: '^group_b\\d+\\.sas7bdat$'
+        name: group_b
+
+2. Command-line interface
+-------------------------
+::
+
+    python s3_download.py --config download_config.yaml [--dry-run]
+                          [--overwrite] [--fail-fast]
+
+Flags:
+    --config PATH   Required. Path to the YAML config above.
+    --dry-run       List discovered clusters and the objects each would
+                    download (with sizes). No GET requests are issued
+                    beyond the initial LIST.
+    --overwrite     Force-redownload every matched object regardless of
+                    local cache state. Equivalent to ``on_exists: overwrite``
+                    for this run.
+    --fail-fast     Abort the whole run on the first per-file download
+                    failure. Default is to log the failure and keep going.
+
+Exit codes:
+    0 - every file downloaded (or skipped) successfully (or dry-run completed)
+    1 - at least one file failed (details on stderr)
+    2 - the LIST returned no objects matching the configured extensions
+
+3. Discovery rules
+------------------
+* Listing is recursive (no S3 ``Delimiter``). Regexes are matched against
+  the *basename* of each key (the part after the last ``/``), so a nested
+  object like ``census/2020/raw/nested/group_c1.sas7bdat`` is grouped by
+  ``group_c1.sas7bdat`` alone.
+* Explicit patterns are tried in order. A key matched by one pattern is
+  removed from the pool before the next pattern runs. Overlap between
+  patterns is flagged as an error at discovery time.
+* Auto-detect groups remaining keys by ``re.sub(r'\\d+$', '', stem)`` with
+  any trailing ``_`` / ``-`` stripped afterward, mirroring
+  ``load_folder.py``. Stems without trailing digits become singleton
+  clusters named after the stem.
+* Within a cluster, files are sorted numerically by the LAST digit group
+  in the stem so ``_9_`` sorts before ``_40_`` regardless of zero-padding.
+
+4. Library usage
+----------------
+::
+
+    from s3_download import load_download_config, list_s3_objects, \
+        discover_clusters, download_cluster, build_s3_client
+
+    cfg = load_download_config("download_config.yaml")
+    s3 = build_s3_client(cfg)
+    objects = list_s3_objects(s3, cfg)
+    for cluster in discover_clusters(cfg, objects):
+        download_cluster(s3, cfg, cluster)
+"""
+
+from __future__ import annotations
+
+import argparse
+import re
+import sys
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple
+
+import boto3
+import yaml
+
+
+DEFAULT_EXTENSIONS: Tuple[str, ...] = (".sas7bdat", ".xpt", ".xport")
+VALID_ON_EXISTS: Tuple[str, ...] = ("skip", "overwrite", "error")
+DEFAULT_CONCURRENCY: int = 4
+
+
+# ---------------------------------------------------------------------------
+# Dataclasses
+# ---------------------------------------------------------------------------
+
+
+@dataclass
+class _ExplicitPattern:
+    """Parsed form of a single ``clusters[*]`` YAML entry."""
+
+    pattern: re.Pattern
+    raw_pattern: str
+    name: str
+
+
+@dataclass
+class DownloadConfig:
+    """Top-level configuration parsed from YAML."""
+
+    bucket: str
+    prefix: str
+    local_folder: Path
+    aws_profile: Optional[str] = None
+    auto_detect: bool = True
+    extensions: Tuple[str, ...] = DEFAULT_EXTENSIONS
+    on_exists: str = "skip"
+    concurrency: int = DEFAULT_CONCURRENCY
+    explicit: List[_ExplicitPattern] = field(default_factory=list)
+
+
+@dataclass
+class S3Object:
+    """A single S3 object selected from the LIST response."""
+
+    key: str
+    basename: str
+    size: int
+
+
+@dataclass
+class ClusterSpec:
+    """Resolved per-cluster download settings."""
+
+    name: str
+    objects: List[S3Object]
+    source: str  # "explicit" or "auto"
+    pattern: Optional[str] = None
+
+
+# ---------------------------------------------------------------------------
+# Config loading
+# ---------------------------------------------------------------------------
+
+
+def _validate_on_exists(value: Any, where: str) -> str:
+    s = str(value).lower()
+    if s not in VALID_ON_EXISTS:
+        raise ValueError(
+            f"{where}: on_exists={value!r} is not one of {list(VALID_ON_EXISTS)}"
+        )
+    return s
+
+
+def _parse_extensions(raw_value: Any, where: str) -> Tuple[str, ...]:
+    if raw_value is None:
+        return DEFAULT_EXTENSIONS
+    if isinstance(raw_value, str):
+        items = [raw_value]
+    elif isinstance(raw_value, list):
+        items = list(raw_value)
+    else:
+        raise ValueError(
+            f"{where}: 'extensions' must be a string or list of strings."
+        )
+    out: List[str] = []
+    for i, item in enumerate(items):
+        if not isinstance(item, str) or not item.strip():
+            raise ValueError(
+                f"{where}: 'extensions[{i}]' must be a non-empty string."
+            )
+        ext = item.strip().lower()
+        if not ext.startswith("."):
+            ext = "." + ext
+        out.append(ext)
+    if len(out) != len(set(out)):
+        raise ValueError(f"{where}: 'extensions' contains duplicates.")
+    return tuple(out)
+
+
+def _parse_concurrency(raw_value: Any, where: str) -> int:
+    if raw_value is None:
+        return DEFAULT_CONCURRENCY
+    try:
+        value = int(raw_value)
+    except (TypeError, ValueError):
+        raise ValueError(
+            f"{where}: 'concurrency' must be a positive integer, "
+            f"got {raw_value!r}"
+        )
+    if value <= 0:
+        raise ValueError(
+            f"{where}: 'concurrency' must be a positive integer, got {value}"
+        )
+    return value
+
+
+def load_download_config(path: Path) -> DownloadConfig:
+    """Parse and validate the YAML download config at ``path``."""
+    path = Path(path)
+    with path.open("r", encoding="utf-8") as f:
+        raw = yaml.safe_load(f)
+
+    if not isinstance(raw, dict):
+        raise ValueError(
+            f"Config at {path} must be a YAML mapping at the top level."
+        )
+
+    missing = [
+        k for k in ("bucket", "prefix", "local_folder") if k not in raw
+    ]
+    if missing:
+        raise ValueError(
+            f"Config {path} missing required keys: {', '.join(missing)}"
+        )
+
+    bucket = str(raw["bucket"]).strip()
+    if not bucket:
+        raise ValueError(f"Config {path}: 'bucket' must be a non-empty string.")
+
+    prefix = str(raw["prefix"])
+    # Normalize: strip leading slash, ensure exactly one trailing slash unless
+    # the user explicitly asked for an empty prefix (whole bucket scan).
+    prefix = prefix.lstrip("/")
+    if prefix and not prefix.endswith("/"):
+        prefix = prefix + "/"
+
+    local_folder = Path(raw["local_folder"])
+    if not local_folder.is_absolute():
+        candidate = (path.parent / local_folder).resolve()
+        # Mirror load_folder's behavior: prefer the config-relative path when
+        # it exists, otherwise keep what the user wrote. Either way, we'll
+        # mkdir(parents=True) before downloading so non-existence is fine.
+        local_folder = candidate if candidate.parent.exists() else candidate
+
+    aws_profile = raw.get("aws_profile")
+    if aws_profile is not None:
+        aws_profile = str(aws_profile).strip() or None
+
+    auto_detect = bool(raw.get("auto_detect", True))
+    extensions = _parse_extensions(raw.get("extensions"), f"Config {path}")
+    on_exists = _validate_on_exists(
+        raw.get("on_exists", "skip"), f"Config {path}"
+    )
+    concurrency = _parse_concurrency(
+        raw.get("concurrency"), f"Config {path}"
+    )
+
+    explicit: List[_ExplicitPattern] = []
+    seen_names: set = set()
+    clusters_raw = raw.get("clusters") or []
+    if not isinstance(clusters_raw, list):
+        raise ValueError(f"Config {path}: 'clusters' must be a list if present.")
+    for i, entry in enumerate(clusters_raw):
+        where = f"Config {path} clusters[{i}]"
+        if not isinstance(entry, dict):
+            raise ValueError(f"{where} must be a mapping.")
+        if "pattern" not in entry or "name" not in entry:
+            raise ValueError(f"{where} must include 'pattern' and 'name'.")
+        raw_pat = str(entry["pattern"])
+        try:
+            compiled = re.compile(raw_pat)
+        except re.error as e:
+            raise ValueError(
+                f"{where}: invalid regex {raw_pat!r}: {e}"
+            ) from e
+        name = str(entry["name"]).strip()
+        if not name:
+            raise ValueError(f"{where}: 'name' must be a non-empty string.")
+        if name in seen_names:
+            raise ValueError(
+                f"{where}: duplicate cluster name {name!r}"
+            )
+        seen_names.add(name)
+        explicit.append(
+            _ExplicitPattern(
+                pattern=compiled, raw_pattern=raw_pat, name=name,
+            )
+        )
+
+    return DownloadConfig(
+        bucket=bucket,
+        prefix=prefix,
+        local_folder=local_folder,
+        aws_profile=aws_profile,
+        auto_detect=auto_detect,
+        extensions=extensions,
+        on_exists=on_exists,
+        concurrency=concurrency,
+        explicit=explicit,
+    )
+
+
+# ---------------------------------------------------------------------------
+# S3 listing
+# ---------------------------------------------------------------------------
+
+
+def build_s3_client(cfg: DownloadConfig):
+    """Build a boto3 S3 client honoring ``cfg.aws_profile`` if set."""
+    if cfg.aws_profile:
+        session = boto3.Session(profile_name=cfg.aws_profile)
+    else:
+        session = boto3.Session()
+    return session.client("s3")
+
+
+def list_s3_objects(s3_client, cfg: DownloadConfig) -> List[S3Object]:
+    """List all objects under ``cfg.prefix`` recursively, filtered by extension."""
+    paginator = s3_client.get_paginator("list_objects_v2")
+    out: List[S3Object] = []
+    for page in paginator.paginate(Bucket=cfg.bucket, Prefix=cfg.prefix):
+        for entry in page.get("Contents", []):
+            key = entry["Key"]
+            if key.endswith("/"):
+                continue
+            basename = key.rsplit("/", 1)[-1]
+            ext = ("." + basename.rsplit(".", 1)[-1].lower()
+                   if "." in basename else "")
+            if ext not in cfg.extensions:
+                continue
+            out.append(
+                S3Object(key=key, basename=basename, size=int(entry["Size"]))
+            )
+    out.sort(key=lambda o: o.key)
+    return out
+
+
+# ---------------------------------------------------------------------------
+# Cluster discovery (mirrors generic_loader/load_folder.py)
+# ---------------------------------------------------------------------------
+
+
+_TRAILING_DIGIT_RE = re.compile(r"\d+$")
+_DIGIT_GROUP_RE = re.compile(r"\d+")
+
+
+def _auto_prefix(stem: str) -> str:
+    """Cluster key for *stem*: strip trailing digits and any trailing _/-."""
+    stripped = _TRAILING_DIGIT_RE.sub("", stem)
+    stripped = stripped.rstrip("_-")
+    return stripped or stem
+
+
+def _basename_stem(basename: str) -> str:
+    if "." in basename:
+        return basename.rsplit(".", 1)[0]
+    return basename
+
+
+def _cluster_sort_key(obj: S3Object) -> Tuple[int, str]:
+    """Sort key for ordering objects within a cluster by trailing digits."""
+    stem = _basename_stem(obj.basename)
+    digits = _DIGIT_GROUP_RE.findall(stem)
+    n = int(digits[-1]) if digits else -1
+    return (n, stem)
+
+
+def discover_clusters(
+    cfg: DownloadConfig, objects: List[S3Object]
+) -> List[ClusterSpec]:
+    """Bucket *objects* into clusters using explicit patterns then auto-detect."""
+    clusters: List[ClusterSpec] = []
+
+    for i, p_i in enumerate(cfg.explicit):
+        for j in range(i + 1, len(cfg.explicit)):
+            p_j = cfg.explicit[j]
+            for obj in objects:
+                if (p_i.pattern.search(obj.basename)
+                        and p_j.pattern.search(obj.basename)):
+                    raise ValueError(
+                        f"Object {obj.basename!r} matches multiple explicit "
+                        f"patterns: {p_i.raw_pattern!r} and "
+                        f"{p_j.raw_pattern!r}"
+                    )
+
+    remaining = list(objects)
+    for patt in cfg.explicit:
+        matched = [o for o in remaining if patt.pattern.search(o.basename)]
+        if not matched:
+            clusters.append(
+                ClusterSpec(
+                    name=patt.name,
+                    objects=[],
+                    source="explicit",
+                    pattern=patt.raw_pattern,
+                )
+            )
+            continue
+        remaining = [o for o in remaining if o not in matched]
+        clusters.append(
+            ClusterSpec(
+                name=patt.name,
+                objects=sorted(matched, key=_cluster_sort_key),
+                source="explicit",
+                pattern=patt.raw_pattern,
+            )
+        )
+
+    if cfg.auto_detect and remaining:
+        buckets: Dict[str, List[S3Object]] = {}
+        for obj in remaining:
+            key = _auto_prefix(_basename_stem(obj.basename))
+            buckets.setdefault(key, []).append(obj)
+        for key in sorted(buckets):
+            clusters.append(
+                ClusterSpec(
+                    name=key,
+                    objects=sorted(buckets[key], key=_cluster_sort_key),
+                    source="auto",
+                )
+            )
+
+    return clusters
+
+
+# ---------------------------------------------------------------------------
+# Download
+# ---------------------------------------------------------------------------
+
+
+def _local_path(
+    cfg: DownloadConfig,
+    cluster: ClusterSpec,
+    obj: S3Object,
+    basename_collisions: set,
+) -> Path:
+    """Resolve the on-disk destination path for *obj*.
+
+    Falls back to a key-derived filename when two objects in the same cluster
+    share a basename (possible under recursive scan).
+    """
+    cluster_dir = cfg.local_folder / cluster.name
+    if obj.basename in basename_collisions:
+        safe = obj.key.replace("/", "__")
+        return cluster_dir / safe
+    return cluster_dir / obj.basename
+
+
+def _basename_collisions(cluster: ClusterSpec) -> set:
+    """Return the set of basenames that appear more than once in *cluster*."""
+    seen: Dict[str, int] = {}
+    for obj in cluster.objects:
+        seen[obj.basename] = seen.get(obj.basename, 0) + 1
+    return {name for name, count in seen.items() if count > 1}
+
+
+def _decide_action(
+    local_path: Path, obj: S3Object, on_exists: str
+) -> Tuple[str, Optional[str]]:
+    """Return ``(action, message)`` where action is 'download' or 'skip'."""
+    if on_exists == "overwrite":
+        return ("download", None)
+    if not local_path.exists():
+        return ("download", None)
+    local_size = local_path.stat().st_size
+    if local_size == obj.size:
+        return (
+            "skip",
+            f"  skip {obj.key} -> {local_path} (size {local_size} matches)",
+        )
+    if on_exists == "error":
+        raise RuntimeError(
+            f"Local file {local_path} exists with size {local_size} but S3 "
+            f"object s3://{obj.key} has size {obj.size} (on_exists=error)"
+        )
+    return (
+        "download",
+        f"  re-download {obj.key} -> {local_path} "
+        f"(local size {local_size} != S3 {obj.size})",
+    )
+
+
+def download_cluster(
+    s3_client,
+    cfg: DownloadConfig,
+    cluster: ClusterSpec,
+    *,
+    on_exists_override: Optional[str] = None,
+    fail_fast: bool = False,
+) -> Tuple[int, int, int, List[Tuple[str, Exception]]]:
+    """Download every object in *cluster* into ``cfg.local_folder/cluster.name``.
+
+    Returns ``(downloaded, skipped, bytes_downloaded, failures)`` where
+    *failures* is a list of ``(key, exception)`` tuples.
+    """
+    if not cluster.objects:
+        return (0, 0, 0, [])
+
+    on_exists = on_exists_override or cfg.on_exists
+    cluster_dir = cfg.local_folder / cluster.name
+    cluster_dir.mkdir(parents=True, exist_ok=True)
+
+    collisions = _basename_collisions(cluster)
+
+    plans: List[Tuple[S3Object, Path, str]] = []
+    skipped = 0
+    for obj in cluster.objects:
+        local_path = _local_path(cfg, cluster, obj, collisions)
+        action, message = _decide_action(local_path, obj, on_exists)
+        if message:
+            print(message, file=sys.stderr)
+        if action == "skip":
+            skipped += 1
+            continue
+        plans.append((obj, local_path, action))
+
+    downloaded = 0
+    bytes_downloaded = 0
+    failures: List[Tuple[str, Exception]] = []
+
+    if not plans:
+        return (0, skipped, 0, failures)
+
+    def _do_one(item):
+        obj, local_path, _action = item
+        local_path.parent.mkdir(parents=True, exist_ok=True)
+        s3_client.download_file(cfg.bucket, obj.key, str(local_path))
+        return obj
+
+    if cfg.concurrency <= 1 or len(plans) == 1:
+        for plan in plans:
+            obj = plan[0]
+            try:
+                _do_one(plan)
+                downloaded += 1
+                bytes_downloaded += obj.size
+                print(
+                    f"  ok   {obj.key} -> {plan[1]} ({obj.size:,} bytes)",
+                    file=sys.stderr,
+                )
+            except Exception as exc:
+                failures.append((obj.key, exc))
+                print(
+                    f"  FAIL {obj.key}: {exc}", file=sys.stderr,
+                )
+                if fail_fast:
+                    break
+    else:
+        with ThreadPoolExecutor(max_workers=cfg.concurrency) as pool:
+            future_to_plan = {pool.submit(_do_one, p): p for p in plans}
+            for fut in as_completed(future_to_plan):
+                plan = future_to_plan[fut]
+                obj = plan[0]
+                try:
+                    fut.result()
+                    downloaded += 1
+                    bytes_downloaded += obj.size
+                    print(
+                        f"  ok   {obj.key} -> {plan[1]} "
+                        f"({obj.size:,} bytes)",
+                        file=sys.stderr,
+                    )
+                except Exception as exc:
+                    failures.append((obj.key, exc))
+                    print(
+                        f"  FAIL {obj.key}: {exc}", file=sys.stderr,
+                    )
+                    if fail_fast:
+                        for other in future_to_plan:
+                            other.cancel()
+                        break
+
+    return (downloaded, skipped, bytes_downloaded, failures)
+
+
+# ---------------------------------------------------------------------------
+# CLI
+# ---------------------------------------------------------------------------
+
+
+def _build_argparser() -> argparse.ArgumentParser:
+    p = argparse.ArgumentParser(
+        description=(
+            "Download S3 objects under a prefix into a local folder, "
+            "grouping objects into clusters that each become one subfolder."
+        ),
+    )
+    p.add_argument(
+        "--config", required=True, type=Path, help="Path to YAML config",
+    )
+    p.add_argument(
+        "--dry-run",
+        action="store_true",
+        help=(
+            "List discovered clusters and the objects each would download. "
+            "No GET requests are issued beyond the initial LIST."
+        ),
+    )
+    p.add_argument(
+        "--overwrite",
+        action="store_true",
+        help=(
+            "Force-redownload every matched object regardless of local "
+            "cache state. Equivalent to on_exists=overwrite."
+        ),
+    )
+    p.add_argument(
+        "--fail-fast",
+        action="store_true",
+        help=(
+            "Abort on the first per-file download failure. Default is to "
+            "log the failure and keep going."
+        ),
+    )
+    return p
+
+
+def _format_bytes(n: int) -> str:
+    units = ("B", "KiB", "MiB", "GiB", "TiB")
+    size = float(n)
+    for unit in units:
+        if size < 1024 or unit == units[-1]:
+            return f"{size:,.1f} {unit}" if unit != "B" else f"{int(size):,} B"
+        size /= 1024
+    return f"{n} B"
+
+
+def _describe_cluster(cluster: ClusterSpec) -> str:
+    src = cluster.source
+    if cluster.pattern:
+        src += f" pattern={cluster.pattern!r}"
+    if not cluster.objects:
+        return (
+            f"cluster {cluster.name!r} [{src}]\n  objects: (no matching keys)"
+        )
+    total = sum(o.size for o in cluster.objects)
+    lines = [
+        f"cluster {cluster.name!r} [{src}] "
+        f"{len(cluster.objects)} object(s), {_format_bytes(total)}"
+    ]
+    for obj in cluster.objects:
+        lines.append(f"  - {obj.key} ({_format_bytes(obj.size)})")
+    return "\n".join(lines)
+
+
+def main(argv: Optional[List[str]] = None) -> int:
+    args = _build_argparser().parse_args(argv)
+
+    cfg = load_download_config(args.config)
+
+    s3 = build_s3_client(cfg)
+    objects = list_s3_objects(s3, cfg)
+
+    if not objects:
+        print(
+            f"error: no objects matching extensions "
+            f"{list(cfg.extensions)} found under "
+            f"s3://{cfg.bucket}/{cfg.prefix}",
+            file=sys.stderr,
+        )
+        return 2
+
+    clusters = discover_clusters(cfg, objects)
+    loadable = [c for c in clusters if c.objects]
+
+    print(
+        f"discovered {len(loadable)} cluster(s) "
+        f"({sum(len(c.objects) for c in loadable)} object(s)) under "
+        f"s3://{cfg.bucket}/{cfg.prefix}:"
+    )
+    for c in clusters:
+        print(_describe_cluster(c))
+
+    if args.dry_run:
+        return 0
+
+    cfg.local_folder.mkdir(parents=True, exist_ok=True)
+
+    on_exists_override = "overwrite" if args.overwrite else None
+
+    totals: List[Tuple[str, int, int, int]] = []  # (name, dl, skip, bytes)
+    failures: List[Tuple[str, str, Exception]] = []  # (cluster, key, exc)
+    aborted = False
+    for cluster in loadable:
+        if aborted:
+            break
+        print(
+            f"\n>>> downloading cluster {cluster.name!r} "
+            f"({len(cluster.objects)} object(s))"
+        )
+        try:
+            dl, sk, by, fails = download_cluster(
+                s3, cfg, cluster,
+                on_exists_override=on_exists_override,
+                fail_fast=args.fail_fast,
+            )
+        except Exception as exc:
+            failures.append((cluster.name, "<cluster>", exc))
+            print(
+                f"  !! cluster {cluster.name!r} aborted: {exc}",
+                file=sys.stderr,
+            )
+            if args.fail_fast:
+                aborted = True
+            continue
+        totals.append((cluster.name, dl, sk, by))
+        for key, exc in fails:
+            failures.append((cluster.name, key, exc))
+        if fails and args.fail_fast:
+            aborted = True
+
+    print("\n=== summary ===")
+    for name, dl, sk, by in totals:
+        print(
+            f"  {name}: downloaded {dl}, skipped {sk}, "
+            f"{_format_bytes(by)}"
+        )
+    for cname, key, exc in failures:
+        print(f"  FAIL {cname} {key}: {exc}", file=sys.stderr)
+
+    return 1 if failures else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/utils/sample_s3_download_config.yaml b/utils/sample_s3_download_config.yaml
new file mode 100644
index 0000000..4ae715e
--- /dev/null
+++ b/utils/sample_s3_download_config.yaml
@@ -0,0 +1,105 @@
+# Example S3 download config for utils/s3_download.py.
+#
+# Shape mirrors what `s3_download.py` expects:
+#
+#   python s3_download.py --config sample_s3_download_config.yaml --dry-run
+#   python s3_download.py --config sample_s3_download_config.yaml
+#
+# Relative paths (e.g. local_folder) are resolved against this config file's
+# directory first, falling back to the current working directory if that
+# doesn't exist.
+
+# ---------------------------------------------------------------------------
+# Required: where to read from and write to
+# ---------------------------------------------------------------------------
+
+bucket: my-bucket
+
+# Listing is recursive under this prefix - no S3 Delimiter is used. A nested
+# object like `census/2020/raw/nested/group_c1.sas7bdat` will be considered.
+# Regex patterns below match against the object BASENAME only (the part
+# after the last `/`), so subfolder location does not affect matching.
+prefix: census/2020/raw/
+
+# Root destination on disk. One subfolder per cluster is created beneath it.
+# If two objects in the same cluster share a basename (possible under the
+# recursive scan), the second one is renamed to a key-derived filename
+# (slashes replaced with `__`) so neither file overwrites the other.
+local_folder: ./downloads
+
+# ---------------------------------------------------------------------------
+# Optional: AWS credentials
+# ---------------------------------------------------------------------------
+
+# Named profile from ~/.aws/credentials. Omit to use the default boto3
+# credential chain (env vars, instance role, SSO, etc.).
+# aws_profile: default
+
+# ---------------------------------------------------------------------------
+# Optional: discovery behavior
+# ---------------------------------------------------------------------------
+
+# When true (default), any object not matched by an explicit pattern below is
+# auto-grouped with its peers by stripping trailing digits (and any trailing
+# _ / -) from the basename stem. Stems without trailing digits become their
+# own singleton cluster.
+#
+# Auto-detect only recognizes *trailing* digit runs. If your basenames put
+# the varying number in the middle of the stem (e.g. surrounded by year,
+# region, and detail components), auto-detect will NOT group them - each
+# object becomes its own singleton cluster. Use an explicit pattern instead;
+# see the embedded-digit example near the bottom of this file.
+auto_detect: true
+
+# Object extensions to consider. Anything else under the prefix is ignored.
+# Default (when this key is omitted): .sas7bdat, .xpt, .xport (matches
+# generic_loader/load_folder.py).
+# extensions:
+#   - .sas7bdat
+#   - .xpt
+
+# ---------------------------------------------------------------------------
+# Optional: download behavior
+# ---------------------------------------------------------------------------
+
+# What to do when the destination file already exists locally.
+#   skip      - (default) if the local file's byte size matches the S3
+#               object's Size, reuse it. If sizes differ, re-download with a
+#               warning. Same byte-size cache rule used by
+#               utils/file_viewer.py::_ensure_local_copy.
+#   overwrite - always re-download.
+#   error     - abort the run if any local file exists with a different size.
+#               (Equal sizes still skip.)
+on_exists: skip
+
+# Parallel download workers. boto3 clients are thread-safe. Default: 4.
+# concurrency: 4
+
+# ---------------------------------------------------------------------------
+# Explicit cluster patterns
+# ---------------------------------------------------------------------------
+#
+# Each pattern is matched against the S3 object BASENAME (last path segment
+# of the key). Objects matched by a pattern are pulled out of the
+# auto-detect pool, so explicit and auto clusters compose cleanly.
+#
+# `name` becomes both the subfolder name under `local_folder` and the label
+# used in the discovery / summary output. Names must be unique across
+# explicit clusters.
+clusters:
+  - pattern: '^group_a\d+\.sas7bdat$'
+    name: group_a
+
+  # - pattern: '^group_b\d+\.sas7bdat$'
+  #   name: group_b
+
+  # Embedded-digit example. When the varying number sits in the MIDDLE of
+  # the basename stem (e.g. year2020_regionA_40_detail.sas7bdat,
+  # year2020_regionA_41_detail.sas7bdat, ...), auto-detect will NOT group
+  # them - each object becomes its own singleton cluster. An explicit
+  # pattern bucketizes them correctly. The \d+ matches any width, and
+  # objects within the cluster are sorted numerically by the last digit
+  # group in the stem, so _9_ sorts before _40_ regardless of zero-padding.
+  #
+  # - pattern: '^year2020_regionA_\d+_detail\.sas7bdat$'
+  #   name: year2020_regionA_detail