Centos – debugging multi-processing in bash

I am trying to run a function through multiple processes in bash (on CentOs).
The function does different filtering over curl:

STEP=50
function filtering() {
    local i="$1"
    curl -s -g -X GET 'https://url.com?filter='$i'_filter'
}

for (( i=0; i<=TOTAL; i+=STEP )); do
    filtering "$i" 
done
wait

It works when runs as is. However, when i run the function in 10 processes it behaves odd, bringing every time different results, mostly unreliable (according to the produced numbers in comparison to the serial filtering above). My code is:

PARALLEL_PROCESSES=10
ACTIVE_PROCESSES=0
for (( i=0; i<=TOTAL; i+=STEP )); do
    filtering "$i" > /dev/null 2>&1 &
    (( ACTIVE_PROCESSES++ ))
    if (( ACTIVE_PROCESSES >= PARALLEL_PROCESSES )); then
        wait -n 2> /dev/null
        (( ACTIVE_PROCESSES-- )) 
    fi
done
wait 

When I check all the printed filters everything looks fine. I think curl doesn’t get something correctly.
GNU parallel doesn’t do my task as well. I thought debugging the own paralellisation would be easier.

1. TOTAL is defined programmatically before the iteration starts. It stands to express the total count of the API filters so that further i split it into chunks of 50. Sometimes TOTAL can be 10K and more. The '$i'_filter' itself consists of 50 filtering elements, while this API supports up to 100. The length limit of the request is not reached, the waf and rate limit doesn’t block me.

2. This function retrieves some data of (let’s say) candies. This data being parsed with jq like jq -r '.data[] | .id, .name, '"$label"' | @csv'

3. By unreliable results I mean inconsistently different results between 2 and more closely launched individual iterations. While the serial approach always delivers the same true results. This makes me feel that in the parallelisation something chaotic being submitted to the filter.

I’d very appreciate practical advices how to debug what is going on/wrong inside, and any alternative suggestion for achieving my goal.