diff --git a/CHANGELOG.md b/CHANGELOG.md
index 25b0353..2017651 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -40,6 +40,22 @@ Note that 0.3.3 was never released; 0.3.4 follows 0.3.2.
 
 ### Fixed
 
+- Docstrings and guides that disagreed with the code are corrected. Most
+  visible:
+  - `SessionConfig.sliding_threshold` now describes renewal once less than
+    that fraction of the TTL remains; it used to say the opposite.
+  - The `cache()` arguments `no_cache`, `stale_ttl`, `private` and `ttl` are
+    described by what they do, and the docstring gains a `Raises:` section.
+  - The per-user `key_builder` example in the HTTP caching guide no longer
+    sets `private=True`, which bypassed the backend and made the key builder
+    unused.
+  - The state quick start catches `StateError`, so a malformed state is a 400
+    instead of a 500.
+  - The `get_session_manager` 500 message and the session dependency
+    docstrings name `FastAPICacheXSessionMiddleware` instead of the
+    deprecated `SessionMiddleware`.
+  - The monitoring routes no longer claim to count cache hits.
+
 - The Redis backend now matches its key prefix and the path given to
   `clear_path()` literally when it builds `SCAN` patterns. Glob characters in
   them used to be live: `clear_path("/files/[draft]")` missed the cached entry
diff --git a/CLAUDE.md b/CLAUDE.md
index f135bc7..61713b9 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -103,6 +103,6 @@ Four non-abstract atomic primitives live on the base class with non-atomic fallb
 - Ruff is configured with `extend-select = ['ALL']` with specific ignores (see `pyproject.toml`). Notable: E501 (line length), B008 (function calls in defaults), FBT001/FBT002 (boolean args — intentional for Cache-Control API).
 - mypy runs in strict mode on the package (not tests).
 - pydocstring convention is Google style.
-- `from __future__ import annotations` is used for forward references.
+- Forward references are mostly quoted annotations with `TYPE_CHECKING` imports; only a couple of modules use `from __future__ import annotations`.
 - All public functions must have complete type annotations.
 - Coverage threshold is 90% (enforced by `pytest-cov`).
diff --git a/docs/DEVELOPMENT.md b/docs/DEVELOPMENT.md
index 42e35a6..8af5e4b 100644
--- a/docs/DEVELOPMENT.md
+++ b/docs/DEVELOPMENT.md
@@ -179,7 +179,9 @@ uv run mypy fastapi_cachex --strict
 
 - Make sure all functions have type annotations
 - Use `Type | None` for parameters that could be None (the codebase uses PEP 604 unions, not `Optional`)
-- Use `from __future__ import annotations` for forward references
+- Write forward references as quoted annotations (`"SessionManager"`), importing the name under
+  `if TYPE_CHECKING:` when it is only needed for typing. Most modules do this; only a couple use
+  `from __future__ import annotations`
 - Keep `fastapi_cachex/py.typed` in place; it is what makes the installed package typed for users
   (it is listed under `[tool.uv.build-backend] include` in `pyproject.toml`)
 
diff --git a/docs/HTTP_CACHING.md b/docs/HTTP_CACHING.md
index 3d88131..6590c65 100644
--- a/docs/HTTP_CACHING.md
+++ b/docs/HTTP_CACHING.md
@@ -120,9 +120,12 @@ HTTP requests.
 >    cache it, and `If-None-Match` revalidation still works against freshly
 >    rendered content.
 > 2. **A key builder that includes the caller's identity** — use this when you
->    do want a server-side cache per user.
+>    do want a server-side cache per user. Leave `private` unset: `private=True`
+>    bypasses the backend, so the key builder would never be used.
 
 ```python
+from fastapi import Request, Response
+
 from fastapi_cachex import cache
 from fastapi_cachex.types import CACHE_KEY_SEPARATOR
 
@@ -149,11 +152,19 @@ def per_user_key(request: Request) -> str:
 
 
 @app.get("/me/dashboard")
-@cache(ttl=60, private=True, key_builder=per_user_key)
-async def my_dashboard(user: CurrentUser):
+@cache(ttl=60, key_builder=per_user_key)
+async def my_dashboard(user: CurrentUser, response: Response):
+    # Without `private`, the response goes out as `Cache-Control: max-age=60`,
+    # which a shared cache (CDN, reverse proxy) may store. Vary on whatever
+    # carries the identity so such a cache keeps one copy per user.
+    response.headers["Vary"] = "Authorization"
     return build_dashboard(user)
 ```
 
+A per-user entry is only safe from shared caches in front of your app if they
+honour `Vary` for that header. When they don't, or when identity comes from
+something a shared cache cannot see, use option 1 instead.
+
 > [!CAUTION]
 > The key builder decides who sees whose data, so the identity it reads must
 > come from something already verified — a claim from a checked token, a user
diff --git a/docs/STATE.md b/docs/STATE.md
index 0a4a272..a65edbd 100644
--- a/docs/STATE.md
+++ b/docs/STATE.md
@@ -20,7 +20,7 @@ from fastapi.responses import RedirectResponse
 
 from fastapi_cachex import BackendProxy
 from fastapi_cachex.backends import MemoryBackend
-from fastapi_cachex.state import InvalidStateError, StateExpiredError, StateManagerDep
+from fastapi_cachex.state import StateError, StateManagerDep
 
 app = FastAPI()
 BackendProxy.set(MemoryBackend())
@@ -38,7 +38,7 @@ async def login(states: StateManagerDep):
 async def callback(state: str, code: str, states: StateManagerDep):
     try:
         data = await states.consume_state(state)  # one-time: deleted on retrieval
-    except (InvalidStateError, StateExpiredError) as e:
+    except StateError as e:  # unknown, expired or malformed state
         raise HTTPException(status_code=400, detail="Invalid state") from e
 
     # Exchange the code for tokens, create a session ...
diff --git a/fastapi_cachex/backends/base.py b/fastapi_cachex/backends/base.py
index 3565e9a..d9e7d27 100644
--- a/fastapi_cachex/backends/base.py
+++ b/fastapi_cachex/backends/base.py
@@ -77,8 +77,9 @@ async def delete_many(self, keys: Iterable[str]) -> int:
 
         The base implementation deletes one key at a time and reports how
         many were attempted, since ``delete`` does not say whether the key
-        existed. The built-in backends override it with a single batched
-        operation that counts what was actually removed.
+        existed. The memory and Redis backends override it with a single
+        batched operation that counts what was actually removed; Memcached
+        keeps this per-key loop.
         """
         count = 0
         for key in keys:
diff --git a/fastapi_cachex/backends/memcached.py b/fastapi_cachex/backends/memcached.py
index 1cf4d76..f5c0fef 100644
--- a/fastapi_cachex/backends/memcached.py
+++ b/fastapi_cachex/backends/memcached.py
@@ -264,13 +264,16 @@ async def delete(self, key: str) -> None:
     async def clear(self) -> None:
         """Clear all values from cache.
 
-        Note: Memcached's flush_all affects the entire server.
-        Consider using clear_path() with your specific keys instead.
+        Note: Memcached's flush_all affects the entire server, including
+        other applications' keys. Memcached cannot enumerate keys, so there
+        is no way to clear only this namespace; delete keys you know by name
+        with ``delete()``/``delete_many()`` instead.
         """
         warnings.warn(
             "Memcached.clear() flushes ALL cached data from the server, "
-            "affecting other applications. Consider using clear_path() instead "
-            "to selectively remove only this namespace's keys.",
+            "affecting other applications. Memcached cannot enumerate keys, so "
+            "this namespace cannot be cleared on its own; delete known keys "
+            "with delete() or delete_many() instead.",
             RuntimeWarning,
             stacklevel=2,
         )
@@ -280,14 +283,15 @@ async def clear(self) -> None:
     async def clear_path(self, path: str, include_params: bool = False) -> int:
         """Clear cached responses for a specific path.
 
-        Note: Memcached does not support pattern-based queries.
-        This method can only delete keys if the exact key is provided,
-        or will try to match keys in memory if include_params=True.
-        For better pattern support, consider using Redis backend.
+        Note: Memcached does not support pattern-based queries, so this
+        only deletes the key that is exactly ``path``. HTTP route keys
+        (``method|||host|||path|||query``) are not matched. For path-based
+        clearing, use the Redis or memory backend.
 
         Args:
-            path: The path to clear cache for
-            include_params: Currently unsupported (Memcached limitation)
+            path: The exact key to delete
+            include_params: Unsupported; emits a ``RuntimeWarning`` and is
+                otherwise ignored
 
         Returns:
             Number of cache entries cleared (0 or 1 for exact match only)
diff --git a/fastapi_cachex/cache.py b/fastapi_cachex/cache.py
index c5e30bb..66a5f49 100644
--- a/fastapi_cachex/cache.py
+++ b/fastapi_cachex/cache.py
@@ -437,24 +437,45 @@ def cache(
 ) -> Callable[[HandlerCallable], AsyncResponseCallable]:
     """Cache decorator for FastAPI route handlers.
 
+    Only GET requests go through the cache; other methods run the handler
+    unchanged.
+
     Args:
-        ttl: Time-to-live in seconds for cache entries, sent as ``max-age``.
+        ttl: How long, in seconds, a stored response may be served without
+            running the handler. The same value is sent as ``max-age``.
             ``ttl=0`` sends ``max-age=0`` and, like ``None``, keeps the entry
             only for ETag revalidation: the body is never served from the
             cache, but a matching ``If-None-Match`` still gets a 304. Negative
             values are rejected.
-        stale_ttl: Additional time-to-live for stale cache entries
-        stale: Stale response handling strategy ('error' or 'revalidate')
-        no_cache: Whether to disable caching
-        no_store: Whether to prevent storing responses
-        public: Whether responses can be cached by shared caches
-        private: Whether responses are for single user only
-        immutable: Whether cached responses never change
-        must_revalidate: Whether to force revalidation when stale
-        key_builder: Custom function to build cache keys. If None, uses default_key_builder
+        stale_ttl: Seconds sent with the directive chosen by ``stale``. It only
+            shapes the ``Cache-Control`` header; the backend entry still
+            expires after ``ttl``. Must be given together with ``stale``.
+        stale: ``"revalidate"`` sends ``stale-while-revalidate=<stale_ttl>``,
+            ``"error"`` sends ``stale-if-error=<stale_ttl>``.
+        no_cache: Run the handler on every request and send ``no-cache``. The
+            response is still stored and ``If-None-Match`` still gets a 304
+            when it matches the fresh ETag. The header then carries only
+            ``no-cache`` (plus ``must-revalidate`` when set); ``ttl``,
+            ``stale``, ``public``/``private`` and ``immutable`` are left out.
+        no_store: Run the handler, store nothing, and send ``no-store``. Takes
+            precedence over every other option.
+        public: Send ``public``. Mutually exclusive with ``private``.
+        private: Send ``private`` and bypass the shared backend entirely: the
+            handler runs on every request and nothing is read or stored. ETag
+            revalidation still works against the freshly rendered response.
+            Mutually exclusive with ``public``.
+        immutable: Send ``immutable``.
+        must_revalidate: Send ``must-revalidate``.
+        key_builder: Custom function to build cache keys. If None, uses
+            ``default_key_builder``.
 
     Returns:
         Decorator function that wraps route handlers with caching logic
+
+    Raises:
+        CacheXError: When the decorator is applied, if ``stale`` and
+            ``stale_ttl`` are not given together, if ``public`` and
+            ``private`` are both set, or if ``ttl`` is negative.
     """
 
     def decorator(func: HandlerCallable) -> AsyncResponseCallable:
diff --git a/fastapi_cachex/manager.py b/fastapi_cachex/manager.py
index ac9398a..7b935fc 100644
--- a/fastapi_cachex/manager.py
+++ b/fastapi_cachex/manager.py
@@ -42,6 +42,8 @@ def __init__(
                 without an explicit ttl. None means no expiry by default.
 
         Raises:
+            BackendNotFoundError: If ``backend`` is None and no backend has
+                been set with ``BackendProxy.set()``.
             ValueError: If ``default_ttl`` is zero or negative.
         """
         self.backend = backend if backend is not None else BackendProxy.get()
diff --git a/fastapi_cachex/routes.py b/fastapi_cachex/routes.py
index ce15686..9a93c65 100644
--- a/fastapi_cachex/routes.py
+++ b/fastapi_cachex/routes.py
@@ -22,7 +22,7 @@
 
 @dataclass
 class CacheHitRecord:
-    """Record for a single cache hit."""
+    """One cached route entry. Despite the name, hits are not counted."""
 
     cache_key: str
     method: str
@@ -36,7 +36,7 @@ class CacheHitRecord:
 
 @dataclass
 class CacheHitSummary:
-    """Summary of cache hit statistics."""
+    """Summary of the cached route entries."""
 
     total_cached_entries: int
     active_entries: int
@@ -45,7 +45,10 @@ class CacheHitSummary:
 
 @dataclass
 class CacheHitsResponse:
-    """Response for cached hits endpoint."""
+    """Response for the ``/cached-hits`` endpoint.
+
+    The ``*_hits`` fields count cached entries, not requests served from them.
+    """
 
     cached_hits: list[CacheHitRecord]
     total_hits: int
@@ -241,9 +244,10 @@ def add_routes(
 ) -> None:
     """Add cache monitoring routes to the FastAPI application.
 
-    This function allows users to optionally add cache monitoring routes
-    to their FastAPI application. Users can call this function to enable
-    cache hit tracking and cache record display.
+    Mounts two read-only routes that report what the configured backend
+    currently holds. They inspect stored entries; nothing counts cache hits.
+    The routes have no authentication of their own, so pass ``dependencies``
+    in production.
 
     Args:
         app: FastAPI application instance
@@ -261,14 +265,16 @@ def add_routes(
                       expiry are still reported. Defaults to True.
 
     Example:
+        ```python
         from fastapi import FastAPI
         from fastapi_cachex import add_routes
 
         app = FastAPI()
         add_routes(app)  # Routes at /cached-hits and /cached-records
 
-        # Or with prefix
-        add_routes(app, prefix="/api/cache")  # Routes at /api/cache/cached-hits and /api/cache/cached-records
+        # Or with a prefix: /api/cache/cached-hits and /api/cache/cached-records
+        add_routes(app, prefix="/api/cache")
+        ```
     """
 
     @app.get(
@@ -277,13 +283,14 @@ def add_routes(
         dependencies=dependencies,
     )
     async def get_cached_hits() -> CacheHitsResponse:
-        """Return cached hit records.
+        """List the cached route entries.
 
-        Shows cache statistics including which routes are frequently being cached,
-        hit counts, and cache key information.
+        Splits every cached key into method, host, path and query, with its
+        ETag and expiry, plus counts of valid and expired entries and the
+        distinct cached paths. Cache hits are not counted.
 
         Returns:
-            CacheHitsResponse containing cache hit records and statistics
+            CacheHitsResponse describing the cached entries
         """
         return _cached_hits(_parse_entries(await _cache_data()))
 
diff --git a/fastapi_cachex/session/config.py b/fastapi_cachex/session/config.py
index ef900a0..c8f4683 100644
--- a/fastapi_cachex/session/config.py
+++ b/fastapi_cachex/session/config.py
@@ -87,7 +87,7 @@ class SessionConfig(BaseModel):
         default=0.5,
         ge=0.0,
         le=1.0,
-        description="Fraction of TTL that must pass before sliding refresh (0.5 = refresh after half TTL)",
+        description="Renew once less than this fraction of session_ttl remains (0.5 = renew in the second half of the TTL)",
     )
 
     # Token settings
diff --git a/fastapi_cachex/session/dependencies.py b/fastapi_cachex/session/dependencies.py
index 1e47184..066c0db 100644
--- a/fastapi_cachex/session/dependencies.py
+++ b/fastapi_cachex/session/dependencies.py
@@ -32,7 +32,8 @@ def get_optional_session(
     """Get session from request state (optional).
 
     This dependency automatically displays the authorization input box in OpenAPI/Swagger UI.
-    The actual authentication is handled by SessionMiddleware; the credentials parameter
+    The actual authentication is handled by the session middleware
+    (``FastAPICacheXSessionMiddleware``); the credentials parameter
     is only used to generate the OpenAPI security scheme.
 
     Args:
@@ -52,7 +53,8 @@ def get_session(
     """Get session from request state (required).
 
     This dependency automatically displays the authorization input box in OpenAPI/Swagger UI.
-    The actual authentication is handled by SessionMiddleware; the credentials parameter
+    The actual authentication is handled by the session middleware
+    (``FastAPICacheXSessionMiddleware``); the credentials parameter
     is only used to generate the OpenAPI security scheme.
 
     Args:
@@ -79,7 +81,8 @@ def get_session_manager(request: Request) -> "SessionManager":
     """Get SessionManager instance from app state.
 
     This dependency allows you to access the SessionManager instance
-    that was registered via SessionMiddleware. Use this when you need
+    that the session middleware (``FastAPICacheXSessionMiddleware``)
+    registered on ``app.state`` when it handled its first request. Use this when you need
     to perform session operations like create, delete, or regenerate.
 
     Example:
@@ -103,7 +106,8 @@ async def login(
         SessionManager instance
 
     Raises:
-        HTTPException: 500 if SessionManager not found in app state
+        HTTPException: 500 if no session middleware has registered a
+            SessionManager yet
     """
     manager: SessionManager | None = getattr(
         request.app.state, "__fastapi_cachex_session_manager", None
@@ -111,7 +115,10 @@ async def login(
     if manager is None:
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="SessionManager not initialized. Ensure SessionMiddleware is added to the app.",
+            detail=(
+                "SessionManager not initialized. Ensure "
+                "FastAPICacheXSessionMiddleware is added to the app."
+            ),
         )
     return manager
 
diff --git a/fastapi_cachex/session/manager.py b/fastapi_cachex/session/manager.py
index 63aebe8..83a6c25 100644
--- a/fastapi_cachex/session/manager.py
+++ b/fastapi_cachex/session/manager.py
@@ -402,8 +402,10 @@ async def clear_expired_sessions(self) -> int:
     async def _iter_sessions(self) -> AsyncIterator[tuple[str, Session]]:
         """Yield every readable session under this manager's prefix with its key.
 
-        Backends without key enumeration (which raise NotImplementedError)
-        simply yield nothing.
+        Backends that cannot enumerate keys yield nothing: the built-in
+        Memcached backend returns ``[]`` from ``get_all_keys()`` with a
+        ``RuntimeWarning``, and a custom backend may raise
+        ``NotImplementedError`` instead.
         """
         try:
             all_keys = await self.backend.get_all_keys()
diff --git a/fastapi_cachex/state/manager.py b/fastapi_cachex/state/manager.py
index fa77f8e..278dd6c 100644
--- a/fastapi_cachex/state/manager.py
+++ b/fastapi_cachex/state/manager.py
@@ -63,6 +63,8 @@ def __init__(
             default_ttl: Default time-to-live in seconds for state
 
         Raises:
+            BackendNotFoundError: If ``backend`` is None and no backend has
+                been set with ``BackendProxy.set()``.
             ValueError: If ``default_ttl`` is zero or negative.
         """
         validate_ttl(default_ttl)