Firebase A/B测试：重复运行同实验且变体数量不同时原有变体的变化

Great question—let’s break down exactly how this works, since Firebase’s user allocation logic depends on a few key details.

Core Background: How Firebase Assigns Users #

Firebase A/B Testing uses a deterministic hashing system based on two critical pieces of data:

• The user’s unique ID (either their authenticated user ID or anonymous ID, depending on your setup)
• A unique "seed" tied directly to the specific experiment you create

This means a user will always get the same variant in the same experiment, but their allocation can shift across different experiments—even if you reuse the same Remote Config parameter name.

Your Scenario: Two Possible Outcomes #

Let’s cover the two most common ways you might re-run the experiment, and what happens to your original Var B users in each case:

1. You create a brand new experiment (same Remote Config parameter, new experiment entry) #

If you stop the original 3-variant experiment and build an entirely new 2-variant experiment (Control + Var B, 50/50 split), Firebase will assign this new experiment a unique seed.

When the new experiment goes live:

• The 33.3% of users originally assigned to Var B will have their ID hashed with this new seed. This means 50% of them will be reassigned to Control, and 50% will stay in Var B—so yes, a portion will end up in a different variant.
• The same random reallocation applies to users who were in Control or Var A in the original experiment.

2. You modify the original experiment (remove Var A, adjust traffic split) and restart it #

If you instead edit the existing experiment (delete Var A, set Control and Var B to 50% each) and restart it, the experiment’s original seed remains intact.

Here’s how allocation plays out here:

• Original Control users (0-33.3% of the hash range): Still fall within the new Control range (0-50%), so they stay in Control.
• Original Var A users (33.3-66.6%): Now split—those in the 33.3-50% range move to Control, while those in 50-66.6% shift to Var B.
• Original Var B users (66.6-100%): Fall entirely within the new Var B range (50-100%), so they will stay in Var B with 100% certainty.

Pro Tip: Keeping Original Var B Users in Place #

If your goal is to ensure all users who were in Var B stay there in the updated experiment, modifying the original experiment (option 2) is the most reliable approach. Alternatively, if you must create a new experiment, you could use Firebase’s user targeting to:

• Exclude users who were in Var B from the Control variant, or
• Force-assign those users to Var B using custom user properties (you’d need to have tracked their original variant assignment first)

Hope that clears up the confusion!

内容的提问来源于stack exchange，提问作者iamronak