diff --git a/configuration/authorization-storage.md b/configuration/authorization-storage.md
index 2a067d9..84a2713 100644
--- a/configuration/authorization-storage.md
+++ b/configuration/authorization-storage.md
@@ -112,7 +112,39 @@ Such authorizations are typically created in the authorization code flow to link
 so that they can be automatically revoked if the authorization code was redeemed multiple times (which may indicate a token leakage).
 In the same vein, ad-hoc authorizations are also created when a refresh token is returned during a resource owner password credentials grant request.
 
-> [!INFO]
+> [!NOTE]
 > When using the [OpenIddict.Quartz](https://www.nuget.org/packages/OpenIddict.Quartz/) integration, ad-hoc authorizations are automatically
 > removed from the database after a short period of time (14 days by default). Unlike ad-hoc authorizations, permanent authorizations
-> never removed from the database.
\ No newline at end of file
+> are never removed from the database.
+
+## Enabling authorization entry validation at the API level
+
+**For performance reasons, OpenIddict 3.0 doesn't check, by default, the status of an authorization entry when receiving an API request**: access tokens are considered
+valid even if the attached authorization was revoked. For scenarios that require immediate authorization revocation, the OpenIddict validation handler can be configured
+to enforce authorization entry validation for each API request:
+
+> [!NOTE]
+> Enabling authorization entry validation requires that the OpenIddict validation handler have a direct access to the server database where authorizations are stored, which makes it
+> better suited for APIs located in the same application as the authorization server. For external applications, consider using introspection instead of local validation.
+>
+> In both cases, additional latency – caused by the additional DB request and the HTTP call for introspection – is expected.
+
+```csharp
+services.AddOpenIddict()
+    .AddValidation(options =>
+    {
+        options.EnableAuthorizationEntryValidation();
+    });
+```
+
+## Disabling authorization storage
+
+While STRONGLY discouraged, authorization storage can be disabled in the server options:
+
+```csharp
+services.AddOpenIddict()
+    .AddServer(options =>
+    {
+        options.DisableAuthorizationStorage();
+    });
+```
\ No newline at end of file
diff --git a/configuration/toc.yml b/configuration/toc.yml
index a2e9af9..ae712c8 100644
--- a/configuration/toc.yml
+++ b/configuration/toc.yml
@@ -4,8 +4,11 @@
 - name: Application permissions
   href: application-permissions.md
 
+- name: Authorization storage
+  href: authorization-storage.md
+
 - name: Token formats
   href: token-formats.md
 
-- name: Authorization storage
-  href: authorization-storage.md
\ No newline at end of file
+- name: Token storage
+  href: token-storage.md
\ No newline at end of file
diff --git a/configuration/token-formats.md b/configuration/token-formats.md
index 6614971..49d9020 100644
--- a/configuration/token-formats.md
+++ b/configuration/token-formats.md
@@ -1,5 +1,11 @@
 # Token formats
 
+> [!NOTE]
+> In OpenIddict 3.0, being able to revoke a token is not tied to the token format and doesn't require enabling reference tokens:
+> regular JWT or ASP.NET Core Data Protection tokens can be revoked as long as token storage is not explicitly disabled by the developer
+>
+> For more information about reference tokens, read [Token storage](token-storage.md).
+
 ## JSON Web Token
 
 OpenIddict 3.0 implements the [JSON Web Token](https://tools.ietf.org/html/rfc7519), [JSON Web Signature](https://tools.ietf.org/html/rfc7515)
diff --git a/configuration/token-storage.md b/configuration/token-storage.md
new file mode 100644
index 0000000..8c4ccdc
--- /dev/null
+++ b/configuration/token-storage.md
@@ -0,0 +1,71 @@
+# Token storage
+
+To keep track of all the tokens produced by its server services, OpenIddict 3.0 creates a token entry in the database for each generated token.
+A token entry contains metadata like the subject of the token, the client identifier of the application it was issued to or its creation and expiration dates.
+
+By default, the token payload – generated using either the
+[Azure Active Directory IdentityModel Extensions for .NET library](https://github.com/AzureAD/azure-activedirectory-identitymodel-extensions-for-dotnet/) for JWT tokens or
+[ASP.NET Core Data Protection](https://docs.microsoft.com/en-us/aspnet/core/security/data-protection/introduction) for Data Protection tokens – is never stored in the database,
+except for authorization codes (that are short-lived), device and user codes (exclusively used in the device code flow).
+
+Such tokens – called reference tokens – are not returned as-is to the caller: instead, their payload is stored in the database entry and a crypto-secure random 256-bit identifier
+– called reference identifier – is returned as a base64url-encoded string and serves as the "final" token used by the client application when communicating with OpenIddict's endpoints
+or with resource servers (if reference access tokens are enabled in the server options).
+
+> [!NOTE]
+> In OpenIddict 3.0, being able to revoke a token is not tied to the token format and doesn't require enabling reference tokens:
+> regular JWT or ASP.NET Core Data Protection tokens can be revoked as long as token storage is not explicitly disabled by the developer.
+
+## Enabling reference access and/or refresh tokens
+
+Reference access and refresh tokens can be manually enabled in the server options for developers who prefer returning
+shorter access and/or refresh tokens or need to deal with limits that would prevent sending large tokens over the wire.
+
+> [!CAUTION]
+> When enabling reference access and/or refresh tokens support, it is STRONGLY recommended to either:
+> - Use the ASP.NET Core Data Protection format for access and refresh tokens, as they benefit from additional security measures that would prevent them from being sent as-is if
+> they were stolen from the database. For more information on how to enable ASP.NET Core Data Protection, read [Token formats](token-formats.md).
+> - Enable column encryption/data at rest encryption to protect the `Payload` column of token entries.
+
+```csharp
+services.AddOpenIddict()
+    .AddServer(options =>
+    {
+        options.UseReferenceAccessTokens()
+               .UseReferenceRefreshTokens();
+    });
+```
+
+## Enabling token entry validation at the API level
+
+**For performance reasons, OpenIddict 3.0 doesn't check, by default, the status of a token entry when receiving an API request**: access tokens are considered valid until they expire.
+For scenarios that require immediate access token revocation, the OpenIddict validation handler can be configured to enforce token entry validation for each API request:
+
+> [!NOTE]
+> Enabling token entry validation requires that the OpenIddict validation handler have a direct access to the server database where tokens are stored, which makes it
+> better suited for APIs located in the same application as the authorization server. For external applications, consider using introspection instead of local validation.
+>
+> In both cases, additional latency – caused by the additional DB request and the HTTP call for introspection – is expected.
+
+```csharp
+services.AddOpenIddict()
+    .AddValidation(options =>
+    {
+        options.EnableTokenEntryValidation();
+    });
+```
+
+## Disabling token storage
+
+While STRONGLY discouraged, token storage can be disabled in the server options:
+
+```csharp
+services.AddOpenIddict()
+    .AddServer(options =>
+    {
+        options.DisableTokenStorage();
+    });
+```
+
+> [!WARNING]
+> Disabling token storage prevents reference access or refresh tokens support from being enabled, as this requires storing the tokens in the database.
\ No newline at end of file