An open API is where application-level metrics matter most, because you've handed outsiders a focused, routine way to consume your data. APIs are more efficient than screen-scraping (a scraper pulls the whole page — CSS, JS, everything — for one datum), but that efficiency cuts both ways: it's far easier to issue an enormous volume of API calls than browser requests, so an untracked API is an open door to abuse. Treat per-key, per-method usage tracking as **mandatory**. Each call carries a unique API key identifying the app and developer; record, hourly, how many calls each key made and the details of each. With that you can see which keys drive the most traffic, correlate API load to resource use, and adjust per-key rate limits as your capacity landscape changes. Flickr auto-invalidates any key that appears to be abusing the API under its Terms of Service. --- *Source: [[The Art of Capacity Planning]] (John Allspaw, O'Reilly 2008) — Ch 3 — Measurement: Units of Capacity*