Starting in CF 1.11 and up to the current draft version of 1.13, there is a contradiction regarding the use of the units_metadata attribute:

A portion of the fourth paragraph of section 3.1.2 states:

A variable must not have a units_metadata attribute if it has no units attribute or if its units do not involve a temperature unit.

Section 4.3.3 describes the use of leap seconds in temporal variables and then suggests the use of the units_metadata:

The standard, proleptic_gregorian, and julian calendars each have two variants. In one variant the timeline does not include leap seconds. In the other variant, the timeline includes leap seconds, even though they are not included in the valid set of datetimes. To resolve the ambiguity between the variants of these calendars, the units_metadata attribute should be defined as well as the calendar attribute, as described later in this section.

Valid time variables do (always?) have a units attribute, but then this is contradicted by the first section mentioned which suggests that units_metadata should only be present for temperature related variables.