Feature Flag Evaluation

This document offers an in-depth explanation of how the SDK determines the value of a feature flag when executing the GetValue function. Understanding this process requires prior knowledge of targeting concepts.

The feature flag's value is determined by:

• The feature flag's rules defined on the Dashboard,
• The User Object provided to the GetValue function, and
• The default value passed to the GetValue function.

The feature flag's value always comes from exactly one rule, following this algorithm:

1. Evaluation of Targeting Rules: If Targeting Rules are present, the SDK evaluates them one by one, from top to bottom. It checks if all the conditions in the rule's IF part are met (i.e. all the conditions evaluate to true).
   • If the conditions are met, the THEN part determines the value to return. However, if the THEN part contains Percentage Options but the Percentage Evaluation Attribute is missing from the User Object, the SDK will skip the Targeting Rule and continue with the next rule - even though the Targeting Rule's conditions are met.
   • If the conditions aren't met, the SDK moves to the next Targeting Rule, or to step 2 (below) if there are no more Targeting Rules.
2. Evaluation of Percentage Options: If a Percentage Options rule exists, the SDK executes the Evaluation of Percentage Options algorithm to determine which Percentage Option applies to the user and returns the value associated with that option. If the Percentage Evaluation Attribute is missing from the User Object, the SDK skips to step 3 (below).
3. Returning simple value: At this stage, the only remaining "rule" is the simple value specified at the end of the feature flag, which the SDK then returns.

In the event of an unexpected error during evaluation, the SDK returns the default value passed to the GetValue function.

Evaluation of a Targeting Rule

The SDK evaluates the conditions in the rule's IF part one by one, from top to bottom. The result of a condition can be one of the following: true, false or cannot evaluate.

The result cannot evaluate occurs when the necessary user attribute is missing, invalid or incorrectly formatted (the SDK logs these issues as warnings).

A Targeting Rule matches only when all its conditions evaluate to true. In any other cases, it doesn't match.

Evaluation of a User Condition

The SDK looks up the comparison attribute (the user attribute referenced by the condition) in the User Object. It compares the attribute value to the comparison value that is set on the Dashboard. The comparison is done according to the selected comparator, resulting in a true or false value. This will be the result of the condition.

The result of the condition will be cannot evaluate in case the comparison attribute is missing (null, undefined, "") or invalid (not of the type expected by the comparator or not formatted properly). In such cases, the Targeting Rule containing the condition will be skipped, and the evaluation will continue with the next rule.

Evaluation of a Flag Condition

Using the same User Object used to evaluate the dependent flag, the SDK evaluates the prerequisite flag (the feature flag referenced by the condition). Then, the result is checked against the comparator. In case the prerequisite flag is not a boolean setting, the result is compared to the comparison value that is set on the Dashboard. The comparison results in a true or false value. This will be the result of the condition.

In case the prerequisite flag is missing or there is a type mismatch between the return value and the comparison value, the evaluation process stops, and the SDK will return the default value. (Though this can happen only when using the flag overrides feature with invalid data.)

Evaluation of a Segment Condition

The SDK looks up the segment referenced by the condition. Then evaluates the condition described by the segment similarly to User Conditions. Finally, the result is checked against the comparator:

• For IS IN SEGMENT, the result of the Segment Condition will be the same as the result of the segment itself.
• For IS NOT IN SEGMENT, the result will be negated (i.e. it will be the opposite of the result of the segment).

If the segment evaluates to cannot evaluate, so does the Segment Condition.

The evaluation process stops if the referenced segment is missing, and the SDK will return the default value. (Though this can happen only when using the flag overrides feature with invalid data.)

Evaluation of Percentage Options

Percentage Options are designed to be consistent and sticky across all SDKs, which means that users with the same attributes fall in the same group and get the same feature flag value across the supported platforms as long as the percentage split is the same.

The SDK looks up the Percentage Evaluation Attribute in the User Object, then:

• The SDK creates a hash from the combination of the specific feature flag's key and the Percentage Evaluation Attribute's value (by default, it is the user identifier).
• The hashing algorithm assigns the user a number between 0 and 99.
• The assigned number determines which group the user falls into, i.e. which Percentage Option applies to the user.

The fact that the above algorithm is implemented across all SDKs guarantees stickiness, consistency and randomness.

info

The evaluation process is entirely implemented within the SDKs, meaning your users' sensitive data never leaves your system. The data flow is one-way – from ConfigCat CDN servers to your SDKs – and ConfigCat does not receive or store any attributes of the User Object passed to the SDKs. This design prioritizes the privacy and security of user data.

Example scenarios for Percentage Options

The following scenarios show how Percentage Options enable controlled and gradual rollouts of features.

Let's imagine the following setup:

• you have two users: Jane and Joe,
• you have two features flags: isTwitterSharingEnabled and isFacebookSharingEnabled,
• the feature flags use percentage-based targeting,
• the Percentage Evaluation Attribute is the user identifier (which is simply the users' name in this example).

First, the users are assigned a number between 0 and 99 based on the hash of the feature flag's key and their identifier. According to the Percentage Options specified on the Dashboard, this number determines which group they are assigned to, i.e. whether or not the feature is enabled for them.

Let's assume the hash algorithm produces the following numbers:

	isTwitterSharingEnabled	isFacebookSharingEnabled
Jane	hash('isTwitterSharingEnabled' + 'Jane') mod 100 results in 8	hash('isFacebookSharingEnabled' + 'Jane') mod 100 results in 64
Joe	hash('isTwitterSharingEnabled' + 'Joe') mod 100 results in 32	hash(isFacebookSharingEnabled' + 'Joe') mod 100 results in 12

1. Initial Setup: 0% ON / 100% OFF

	isTwitterSharingEnabled 0% ON / 100% OFF	isFacebookSharingEnabled 0% ON / 100% OFF
Jane	8 >= 0 → OFF	64 >= 0 → OFF
Joe	32 >= 0 → OFF	12 >= 0 → OFF

2. Adjustment to 10% ON / 90% OFF In this case both feature flags are enabled for only 10% of users.

	isTwitterSharingEnabled 10% ON / 90% OFF	isFacebookSharingEnabled 10% ON / 90% OFF
Jane	8 < 10 → ON	64 >= 10 → OFF
Joe	32 >= 10 → OFF	12 >= 10 → OFF

info

The isTwitterSharingEnabled feature flag is enabled only for Jane because Joe's user identifier places him in a different group.

However, isFacebookSharingEnabled is disabled for both of them even though both feature flags are set to 10% ON / 90% OFF. This happens because the feature flag key is also involved in computing the hash.

3. Increasing isTwitterSharingEnabled to 40% ON / 60% OFF

	isTwitterSharingEnabled 40% ON / 60% OFF	isFacebookSharingEnabled 10% ON / 90% OFF
Jane	8 < 40 → ON	64 >= 10 → OFF
Joe	32 < 40 → ON	12 >= 10 → OFF

4. Rolling Back to a Safer 10% ON / 90% OFF

Same setup as in Step 2.

There are cases when you want to roll back a feature flag to a safer state. In this case, you can change the Percentage Options to 10% ON / 90% OFF again. The sticky nature of percentage-based targeting ensures that the same userbase is served ON as in Step 2, not another random 10% of users.

	isTwitterSharingEnabled 10% ON / 90% OFF	isFacebookSharingEnabled 10% ON / 90% OFF
Jane	8 < 10 → ON	64 >= 10 → OFF
Joe	32 >= 10 → OFF	12 >= 10 → OFF

5. Final Step: Moving to 100% ON / 0% OFF

After testing the feature flags, you can safely move them to 100% ON. This enables the features for all users.

	isTwitterSharingEnabled 100% ON / 0% OFF	isFacebookSharingEnabled 100% ON / 0% OFF
Jane	8 < 100 → ON	64 < 100 → ON
Joe	32 < 100 → ON	12 < 100 → ON