diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6d5716db..51c7ab69 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,7 @@
+# v4.2.0
+
+* add several new ``Loadable`` convenience classes. ``ReplayCache`` for accessing random elements from a circlecore database, ``ReplayDir`` for folders of ``.osr`` files, and ``ReplayID`` for when you only know the replay id.
+
 # v4.1.2
 
 * correctly account for skips in replays
diff --git a/circleguard/__init__.py b/circleguard/__init__.py
index b519fa76..ce884e67 100644
--- a/circleguard/__init__.py
+++ b/circleguard/__init__.py
@@ -1,7 +1,9 @@
 import logging
 
 from circleguard.circleguard import Circleguard, set_options
-from circleguard.loadable import Check, Replay, ReplayMap, ReplayPath, Map, User, MapUser, ReplayContainer, LoadableContainer, Loadable
+from circleguard.loadable import (Check, Replay, ReplayMap, ReplayPath, Map, User,
+        MapUser, ReplayDir, ReplayContainer, LoadableContainer, Loadable, ReplayCache,
+        CachedReplay, ReplayID)
 from circleguard.enums import Key, RatelimitWeight, Detect, ResultType
 from circleguard.mod import Mod
 from circleguard.utils import TRACE, ColoredFormatter
@@ -26,7 +28,8 @@
 "Circleguard", "set_options",
 # loadables
 "Check", "ReplayContainer", "LoadableContainer", "Map", "User", "MapUser",
-"Replay", "ReplayMap", "ReplayPath", "Loadable",
+"ReplayCache", "Replay", "ReplayMap", "ReplayPath", "CachedReplay", "Loadable",
+"ReplayID", "ReplayDir",
 # enums
 "Key", "RatelimitWeight", "Detect", "ResultType",
 # mod
diff --git a/circleguard/loadable.py b/circleguard/loadable.py
index dd5f93d4..ed130932 100644
--- a/circleguard/loadable.py
+++ b/circleguard/loadable.py
@@ -1,8 +1,13 @@
 import abc
 import logging
+from pathlib import Path
+import os
 
 import circleparse
 import numpy as np
+import sqlite3
+import random
+import wtc
 
 from circleguard.enums import RatelimitWeight
 from circleguard.mod import Mod
@@ -431,6 +436,105 @@ def __eq__(self, loadable):
                 and self.span == loadable.span)
 
 
+class ReplayCache(ReplayContainer):
+    """
+    Contains replays represented by a circlecore database. Primarily useful
+    to randomly sample these replays, rather than directly access them.
+
+    Parameters
+    ----------
+    path: string
+        The path to the database to load replays from.
+    num_maps: int
+        How many (randomly chosen) maps to load replays from.
+    limit: int
+        How many replays to load for each map.
+    Notes
+    -----
+    :meth:`~.load_info` is an expensive operation for large databases
+    (likely because of inefficient sql queries). Consider using as few instances
+    of this object as possible.
+    """
+    def __init__(self, path, num_maps, num_replays):
+        super().__init__(False)
+        self.path = path
+        self.num_maps = num_maps
+        self.limit = num_replays * num_maps
+        self.replays = []
+        conn = sqlite3.connect(path)
+        self.cursor = conn.cursor()
+
+    def load_info(self, loader):
+        map_ids = self.cursor.execute(
+            """
+            SELECT DISTINCT map_id
+            FROM replays
+            """
+        ).fetchall()
+        # flatten map_ids, because it's actually a list of lists
+        map_ids = [item[0] for item in map_ids]
+        chosen_maps = random.choices(map_ids, k=self.num_maps)
+
+        subclauses = [f"map_id = {chosen_map}" for chosen_map in chosen_maps]
+        where_clause = " OR ".join(subclauses)
+
+        # TODO LIMIT clause isn't quite right here, some maps will have less
+        # than ``num_replays`` stored
+        infos = self.cursor.execute(
+            f"""
+            SELECT user_id, map_id, replay_data, replay_id, mods
+            FROM replays
+            WHERE {where_clause}
+            LIMIT {self.limit}
+            """
+        )
+
+        for info in infos:
+            r = CachedReplay(info[0], info[1], info[4], info[2], info[3])
+            self.replays.append(r)
+
+    def all_replays(self):
+        return self.replays
+
+    def __eq__(self, other):
+        return self.path == other.path
+
+
+class ReplayDir(ReplayContainer):
+    """
+    A folder with replay files inside it.
+
+    Notes
+    -----
+    Any files not ending in ``.osr`` are ignored.
+
+    Warnings
+    --------
+    Nested directories are not support (yet). Any folders encountered will be
+    ignored.
+    """
+    def __init__(self, dir_path, cache=None):
+        super().__init__(cache)
+        self.dir_path = Path(dir_path)
+        if not self.dir_path.is_dir():
+            raise ValueError(f"Expected path pointing to {self.dir_path} to be "
+                              "a directory")
+        self.replays = []
+
+    def load_info(self, loader):
+        for path in os.listdir(self.dir_path):
+            if not path.endswith(".osr"):
+                continue
+            replay = ReplayPath(self.dir_path / path)
+            self.replays.append(replay)
+
+    def all_replays(self):
+        return self.replays
+
+    def __eq__(self, other):
+        return self.dir_path == other.dir_path
+
+
 class Replay(Loadable):
     """
     A replay played by a player.
@@ -779,3 +883,43 @@ def __str__(self):
             return f"Loaded ReplayPath by {self.username} on {self.map_id} at {self.path}"
         else:
             return f"Unloaded ReplayPath at {self.path}"
+
+
+class ReplayID(Replay):
+    def __init__(self, replay_id, cache=None):
+        super().__init__(RatelimitWeight.HEAVY, cache)
+        self.replay_id = replay_id
+
+    def load(self, loader, cache):
+        # TODO file github issue about loading info from replay id,
+        # right now we can literally only load the replay data which
+        # is pretty useless if we don't have a map id or the mods used
+        cache = cache if self.cache is None else self.cache
+        replay_data = loader.replay_data_from_id(self.replay_id, cache)
+        self._process_replay_data(replay_data)
+        self.loaded = True
+
+    def __eq__(self, other):
+        return self.replay_id == other.replay_id
+
+
+class CachedReplay(Replay):
+    def __init__(self, user_id, map_id, mods, replay_data, replay_id):
+        super().__init__(RatelimitWeight.NONE, False)
+        self.user_id = user_id
+        self.map_id = map_id
+        self.mods = Mod(mods)
+        self.replay_data = replay_data
+        self.replay_id = replay_id
+
+    def load(self, loader, cache):
+        if self.loaded:
+            return
+        decompressed = wtc.decompress(self.replay_data)
+        replay_data = circleparse.parse_replay(decompressed, pure_lzma=True).play_data
+        self._process_replay_data(replay_data)
+        self.loaded = True
+
+    def __eq__(self, other):
+        # could check more but replay_id is already a primary key, guaranteed unique
+        return self.replay_id == other.replay_id
diff --git a/circleguard/loader.py b/circleguard/loader.py
index 533ed361..2a1dbe62 100644
--- a/circleguard/loader.py
+++ b/circleguard/loader.py
@@ -358,6 +358,23 @@ def replay_data(self, replay_info, cache=None):
             self.cacher.cache(lzma_bytes, replay_info)
         return replay_data
 
+    # TODO make this check cache for the replay
+    def replay_data_from_id(self, replay_id, cache):
+        """
+        Retrieves replay data from the api, given a replay id.
+
+        Parameters
+        ----------
+        replay_id: int
+            The id of the replay to retrieve data for.
+        """
+        response = self.api.get_replay({"s": replay_id})
+        Loader.check_response(response)
+        lzma = base64.b64decode(response["content"])
+        replay_data = circleparse.parse_replay(lzma, pure_lzma=True).play_data
+        # TODO cache the replay here if the api ever gives us the info we need
+        return replay_data
+
     @lru_cache()
     @request
     def map_id(self, map_hash):
diff --git a/circleguard/result.py b/circleguard/result.py
index ede139b9..7dbe1e3e 100644
--- a/circleguard/result.py
+++ b/circleguard/result.py
@@ -66,10 +66,12 @@ class StealResult(ComparisonResult):
         The other replay involved.
     earlier_replay: :class:`~circleguard.loadable.Replay`
         The earlier of the two replays (when the score was made). This is a
-        reference to either replay1 or replay2.
+        reference to either replay1 or replay2, or ``None`` if one of the
+        replays did not provide a timestamp.
     later_replay: :class:`~circleguard.loadable.Replay`
         The later of the two replays (when the score was made). This is a
-        reference to either replay1 or replay2.
+        reference to either replay1 or replay2, or ``None`` if one of the
+        replays did not provide a timestamp.
     similarity: int
         The similarity of the two replays (the lower, the more similar).
         Similarity is, roughly speaking, a measure of the average pixel
@@ -82,6 +84,10 @@ class StealResult(ComparisonResult):
     def __init__(self, replay1: Replay, replay2: Replay):
         super().__init__(replay1, replay2, ResultType.STEAL)
 
+        # can't compare ``None`` timestamps
+        if not self.replay1.timestamp or not self.replay2.timestamp:
+            return
+
         if self.replay1.timestamp < self.replay2.timestamp:
             self.earlier_replay: Replay = self.replay1
             self.later_replay: Replay = self.replay2
diff --git a/circleguard/version.py b/circleguard/version.py
index 13ffcf42..0fd7811c 100644
--- a/circleguard/version.py
+++ b/circleguard/version.py
@@ -1 +1 @@
-__version__ = "4.1.2"
+__version__ = "4.2.0"