Reliability Contract

One Retry on a Failed Browser Request

When you force browser rendering (mode: "js", render_js: true or a forced tier) and the page comes back as a verification screen or times out, the error carries a hint that names the setting to change. The platform can apply that hint for you once: it retries the same job a single time through the standard route, with a fresh identity.

• The retry happens at most once per job and reuses the job id.
• You are billed only for a page that was delivered, at the price of the route that returned it. A failed retry costs nothing.
• It never moves a request to a more expensive route than the one you asked for.

Pacing

Batches are paced from what the platform observes, not from fixed per-site numbers. For each domain in your batch, the number of jobs in flight starts at your plan's limit. If challenges or timeouts rise among your own jobs on that domain, the window shrinks; when they clear, it grows back. It never drops below 1, and a healthy domain is never slowed.

To cap a batch yourself, set max_in_flight_per_domain. It can only lower the window, never raise it. The live value for each domain is returned in the batch status as dispatch_windows:

{
  "dispatch_windows": [
    { "domain": "example.com", "window": 8, "limit": 25, "in_flight": 8, "throttled": true }
  ]
}

Domain Pauses and Reopen Time

After repeated failures on a domain, automatic attempts on it pause for a short time. A pause caused only by requests where you forced the strategy, from a single account, does not pause the domain for other accounts. Every request rejected by a pause returns the same fields, on single scrapes (HTTP 422) and on each batch item (error_detail), and no credits are charged:

{
  "message": "Domain 'example.com' is temporarily unavailable — too many consecutive scrape failures. Retry after 90s. No credits were charged.",
  "code": "domain_circuit_open",
  "retryable": true,
  "retry_after_seconds": 90,
  "reopen_at": "2026-11-10T12:31:30+00:00",
  "scope": "transient",
  "trip_count": 1,
  "next_action": { "type": "wait", "retry_after": 90 }
}