# Traffic

Two numbers decide who sees what:

- **Coverage** (`coverage_bp`, 0 to 10,000 basis points): the share of targeted visitors who enter the experiment. The rest see your original page and are not counted.
- **Weights** (`weight_bp` per variation, adding up to 10,000): how the visitors who enter are split between the control and the variations. The default is an equal split.

Example: coverage 4,000 and weights 5,000 / 5,000 means 40% of targeted visitors enter the test: 20% see the control and 20% see Variation B. The other 60% see the original page and are not counted.

## How visitors are assigned

Every visitor gets a random id when TryFast first sees them (after consent, with the default consent mode). The variation is computed from that id and the experiment, so it is the same on every page and every visit, and the same in the browser and in TryFast's own records. The assignment is also stored in the visitor's browser, so a visitor keeps their variation even after the experiment's weights or coverage change.

A visitor is assigned only when the experiment's [targeting](targeting.md) matches on the current page. From then on they stay in their variation for the whole experiment, and their conversions count from that moment until the experiment stops.

## Ramp up safely

Start with a small coverage while you check that everything works, then raise it:

- **Raising coverage never moves an assigned visitor** to another variation; it only lets more new visitors in.
- **Lowering coverage** stops new visitors from entering; visitors already in the test keep their variation.
- **Coverage 0** on a running experiment lets nobody new in. Pausing is usually clearer.

On a running or paused experiment, coverage is the only traffic setting you can change, with the traffic action (`POST /api/v1/experiments/{id}/traffic`, `set_traffic_allocation` for agents, or the Traffic step in the app). Changing it is a launch-type action: agents with the default permission get an [approval link](lifecycle.md#approvals). Changes reach every visitor within 2 minutes.

Weights and variations cannot change while an experiment runs, because that would mix two different tests in one result. Stop the experiment and duplicate it instead.

## Rules

- Weights must add up to exactly 10,000. A variation other than the control cannot have weight 0 (remove it instead).
- The control always gets the letter A, variations B to H in order.

## When the plan limit is reached

When a workspace reaches its monthly tested users, visitors already in an experiment keep their variation and their conversions keep counting, and new visitors see your original pages without entering any experiment. Your site never breaks. Enrollment resumes on the first day of the next month (UTC) or when the plan changes. At 80% the app shows a notice.