I've got some metrics from smartpaymapi.ovoenergy.com/usage/api but I'm slightly puzzled.

Those are usage data for half-hourly periods, each page usually giving 48 figures from midnight to midnight GMT.

For peak/offpeak usage over a 24-hour period, you’d have to check the actual register readings. Mine look like this:
  

The URL for these JSON data is https://api.eu1.prod.kaluza.com/graphql/1, but the request includes details narrowing down its scope. You might find it more profitable to check out Mike Williams’ lovely little OvoData utility - he’s published all his source code for people like you to plunder.

You will never get correspondence between the nominal half-hourly buckets (e.g. from 00:00:00 to 06:59:59) and the difference between two days’ offpeak register readings. This results from the meter’s ‘randomized offset’, which delays the start and end of each offpeak period by a few minutes. In my sample data, yesterday’s offpeak consumption was 582.050 - 581.744 = 0.306 kWh. The half-hourly data for the nominal offpeak period (14 buckets) add up to 0.318 kWh. 

That’s what I’m after, I’ll take a wander through the code and see what I can grab.

I’m not too worried by matching the figures with the bills, I’m much more interested in trends and which appliances are consuming power overnight.

A task for a rainy day I think.

Then it’ll be trying to put ambient temperature in the same dataset.