Have you considered creating 'totalQuery' as a recording rule as well?

[mimitantono]

Sloth generates recording rules like slo:sli_error:ratio_rate5m, but when considering only the error ratio, we lack important context. For a service that experiences varying request rates, a 50% error ratio over a 5-minute period could signify errors in just 2 out of 4 requests, a stark contrast to a scenario where thousands of requests occurred in another 5-minute interval.

We can enhance accuracy by including the total query in the recording rule. This enables more precise calculations when someone needs to reapply the ratio rate to different time windows, eliminating concerns about inaccuracies in averaging the averages.