Budget for the work you need.
A successful page extraction costs one pageflock credit. Contacts, product data and YouTube metadata are one credit each. A successful search costs five, including a valid search with no results. A company profile reserves five and settles at the pages it actually reads (one to four). If you know the URL already, you can go directly to extraction. A collection charges per successful page rather than a flat job fee.
The API reserves the cost before starting work. Failed and blocked operations refund that reservation. Usage and Activity let you reconcile what happened. Paid plans renew monthly and add credits after each successful payment. Unused credits roll over, including earlier one-time purchases. Cancel in Plans & billing before renewal to stop the next charge; your existing balance does not start a subscription.
Rate and concurrency are different.
A request-per-minute limit controls how quickly you can start work. A concurrency limit controls how much work can be in progress at once. A collection also has its own page and active-job limits. Read the current values in Usage or Pricing instead of hard-coding assumptions into your application.
All keys on an account share its limits and balance. Creating another key does not create another allowance. Start with one request at a time, then increase concurrency only within the limits your account reports.
Retry deliberately.
A 401 response means the key or session needs attention. A 402 means there are not enough credits. For 429, respect Retry-After when the response includes it. Inspect a 403 or a source failure instead of retrying indefinitely.
If your connection drops after sending a request, check Activity before sending it again. Synchronous extraction and search do not currently offer an idempotency key, so the first request may have succeeded and been billed. Collections support an account-scoped client_id to identify creation retries; use the same value only for the same intended job.