--- title: "Variants and allocation" description: "Configure the alternatives and traffic split for an Experiment." search: tags: ["Features"] keywords: ["holdout", "control group", "traffic split", "pin variant"] --- Variants let you compare different versions of the same part of an Experience. You choose the alternatives and how traffic is split between them, and Nuxie assigns customers to the test. Each variant has its own name and identity, so its results stay connected to the path customers received. ## Control, treatment, and holdout An Experiment supports two to five variants. Ask Nuxie to prepare the paths and give them names that describe the differences. ### Control The control is the baseline you compare against. This will often be your existing screen or offer, so you can see whether a change improves on what customers already receive. ### Treatment A treatment contains the change you're testing. It needs a complete path through that section of the Experience, including the actions that let the customer continue afterward. ### Holdout A holdout skips the treatment and continues through the common Journey path. It's useful when you want to understand whether a section helps at all. For example, you could compare an onboarding survey with a holdout that goes directly to the next step. Customers in a holdout can still see screens before and after the Experiment. They're skipping the tested section, not the whole Experience. ## Allocation Tell Nuxie what share of customers each variant should receive. The percentages must add up to 100%. For example: > Give the current onboarding 50% of the traffic and the shorter version 50%. To edit the split yourself, select the Experiment step on the Canvas and change the percentages under **Settings → Traffic Split**. Publish the Experience to apply the change. For three alternatives, you could use 34%, 33%, and 33%. A split is a chance of assignment, so a small number of customers won't necessarily produce those exact proportions. Nuxie assigns a variant to the customer, and their Journey follows that path when it reaches the Experiment. If you change the allocation, it applies to new assignments. Journeys already underway keep their selected variant. **Pause** sends new arrivals to control, and **Resume** restores your configured split. ## Device pins Open **Test on a device** on the Experiment card to pin a customer or device ID to a variant. This lets you try each path in your app without waiting to be randomly assigned to it. Pinned customers are excluded from the comparison. Delete the pin when you want to return to normal assignment. As with allocation changes, pinning doesn't replace the variant in an existing Journey. ## Exposure A customer counts as exposed when their Journey reaches the Experiment and selects a variant. For a holdout, that happens when the Journey skips the treatment. Being assigned earlier in the Journey doesn't count until the customer reaches this point. Results use unique exposed customers. Seeing the same screen several times doesn't make someone several people in the test. Nuxie records presentation separately, so you can distinguish reaching the Experiment from actually seeing its content. If a selected treatment later fails to load or present, the customer still counts as exposed. Keeping that result in the comparison means a variant's loading failures don't disappear from its results. ### Fallbacks and mixed variants If an assignment can't be used, the Experience can fall back to control to keep the Journey moving. This fallback traffic is reported separately from the randomized comparison. Customers seen in more than one variant are also kept separate. Nuxie checks these groups and the traffic split before declaring a winner. [Goals and results](/docs/guides/experiment-results) explains how to read the comparison and choose what customers receive after the test.