Resetting an experiment clears out the posterior data (pulls and successes) for the experiment as a whole, as well as for individual variants within the experiment.
While the underlying source data is NOT deleted (we maintain a full log of everyone who was ever bucketed into a test), the cached values for pulls and successes (which are used to create the Beta distribution) are forcefully reset to zero.
With Bayesian tests, you do not need to reset an experiment as a result of adding a new variant (the same is not true of frequentist A/B tests).
Each experiment type (e.g., landing page) will define the API for resetting the underlying experiment.