1: \begin{abstract}

2: We derive an exact p-value for testing a global null

3: hypothesis in a general adaptive regression setting.  Our

4: approach uses the Kac-Rice formula \citep[as described in][]{RFG}

5: applied to the problem of maximizing a Gaussian process. 

6: The resulting test statistic has a known distribution in

7: finite samples, assuming Gaussian errors.  We

8: examine this test statistic in the case of the lasso, group lasso,

9: principal components, and matrix completion problems. For the lasso

10: problem, our test relates closely to the recently proposed covariance

11: test of \citet{covtest}.   Our approach also yields exact selective

12: inference for the mean parameter at the global maximizer of the process.

13: \end{abstract}

14: