1: \begin{abstract}

2: % \editor{From the abstract it is not clear what the contribution of this paper is w.r.t. earlier work.

3: % [\dots] These should be better highlighted in the abstract.}

4: We study connections between ordinary differential equation (ODE)

5: solvers and probabilistic regression methods in statistics.

6: We provide a new view of probabilistic ODE solvers as active

7: inference agents operating on stochastic differential equation

8: models that estimate the unknown initial value problem (IVP)

9: solution from approximate observations of the \edit{solution derivative},

10: as provided by the ODE dynamics.

11: Adding to this picture, we show that several multistep methods

12: of Nordsieck form can be recasted as Kalman filtering on

13: $q$-times integrated Wiener processes.

14: Doing so provides a family of IVP solvers that return a Gaussian

15: posterior measure, rather than a point estimate.

16: We show that some such methods have low computational overhead,

17: nontrivial convergence order, and that the

18: posterior has a calibrated concentration rate.

19: Additionally, we suggest a step size adaptation algorithm

20: which completes the proposed method to a practically useful

21: \edit{implementation}, which we experimentally evaluate using a

22: representative set of standard codes \edit{in} the DETEST benchmark set.

23: \end{abstract}