Minor nit: many non-billing metrics are near-real time, e.g. DB concurrents, cloud functions CPU/RAM usage; any metrics that require aggregation (storage, billing) are going to be batched less frequently. This is going to be true across all platforms of non-trivial scale (eventual consistency + batch jobs).

Second note: the number of people who actually do this is very low (a few a year, of hundreds of thousands of developers). The blog posts are scary, but in my ~five years at Firebase, I'm pretty sure we refunded every one. As my boss (James Tamplin, CEO of Firebase) used to say, "There are lots of bad systems, but rarely are there bad people."