Elasticsearch: use a decay function in function_score on field coupled with random sorting on another field

In my clusters, I have documents containing, among others, a field created_at which is the date at which they were inserted and a field category which is a string describing the category of the document: its values are in list ["a", "b", "c", ...].

I got two problems:

1. I want to rank items by recency (se newer first)
2. I also want to prevent documents with the same category to appear near to each other

For problem 1. alone, I would simply use a decay (either among the linear/gauss/exp available) on created_at, setting its parameters, say as in

functions: [
    {
        gauss: {
            created_at : {
                "origin": 'now',
                "scale": '12h',
                "offset": '0s',
                "decay": 0.5
            }
        }
    }
]

Problem 2. is particularly severe as for reasons related to the data, many documents with same category get inserted at the same time. For this problem alone, I would use the random_score field in ES to resort results randomly, say as in

functions: [
    random_score:
        { 
          seed: 2          // seed: ID of user
        }               
    }
]

What I want though is a query where results are both sorted by recency but also the same category values do not appear clustered. I tried combining them as

functions: [
    {
        gauss: {
            created_at : {
                "origin": 'now',
                "scale": '12h',
                "offset": '0s',
                "decay": 0.5
            }
        }
    },
    { random_score:
        { seed: 2 }
    },
]

but the results are not really satisfactory on the side of the category.

Is the best I can do playing with the weights of these two functions, or is there a better way to achieve my goal?