The Energy Dashboard uses the recorder statistics for the base data.
All energy values come from the long-term statistics, which are captured hourly. The hour value, for state class total-increasing sensors such as energy meters, is just the running sum and final state value for the hour.
As such there is no captured '15 minute' value, and it is not possible therefore to update sub-hourly.
The power 'now' values I believe, as state class measurement, come from the short-term 5 minute averages, and this graph does update every 5 minutes, at least for the 10 day default retention of the data. You will note that, when viewing the power graph for an historical date from more than 10 days ago, the graph resorts to using the long-term hourly figures only, as the short term data has been purged.
The grid energy input and output sensors are expected to be used only for the grid values. Adding other 'device' sensors will double count. At the moment, if you wish to calculate energy and costs for devices, I believe that the expected approach is to add a template sensor, with the daily device energy (resetting utility meter if required) and current energy unit cost used to calculate the day cost.
Naturally for variable rate tariffs, more work is required to capture the hour/half-hour/quarter-hour energy consumption (resetting utility meter) and the current tariff. Clearly easier to (inappropriately) use the grid import to do this for you, but double counting will result.
Add a utility meter, source as the energy sensor, reset 15 minutes.
Add a template sensor, multiply the above by the current energy tariff cost (sensor)
Add a utility meter (reset optional) source above, with the 'source may periodically reset to zero' turned on.
I realise that this is extra work for you, but a lot less work than changing the Recorder long-term statistics and the Energy Dashboard.