FourA Digest — Oct 16 to Oct 23, 2026 FourA's Single, Proxy Finder and Browser endpoints now return HTTP 400 with an `Invalid request:` error listing up to five faulty fields, replacing the generic 503 that previously invited pointless retries, according to the company's Oct 16 to Oct 23, 2026 digest. The change classifies bad requests as a `client_error` outcome that costs 0 credits, and the Budget view now separates billed credits from total spent credits with sortable per-key, per-domain and per-client-IP columns. FourA also made password changes and sign-outs end all other Dashboard sessions, added gateway-redirect detection to Proxy Finder, and moved api.foura.ai and eu.api.foura.ai to multiple answering servers that take over new calls within a couple of minutes if one stops responding. Highlights A request you got wrong is now your 400, not our 503. Single, Proxy Finder and Browser name each field at fault, charge nothing for it, and no longer invite a retry that can't work. The Budget view also splits the credits you pay for from everything your requests used, and changing your password now signs out every other Dashboard session. You'll find the first two on the changelog https://updates.foura.ai . What's New A bad request says what's wrong Send Single, Proxy Finder or Browser a body it can't run a missing url , a timeout ms that isn't a number and you get HTTP 400. The error starts with Invalid request: and lists each field at fault with the reason, up to five of them. It's a client error outcome and costs 0 credits. Until this week those requests came back as a generic 503, which reads like trouble on our side and tells a retry loop to go again. If your client treats a 400 as final the sample in our error reference https://foura.ai/docs/api/errors does , a typo now stops the run on the first call: python import requests r = requests.post "https://api.foura.ai/api/single", headers={"X-API-Key": "pk live ..."}, json={"url": "https://example.com/data", "timeout ms": "fast"}, if r.status code == 400: "Invalid request: timeout ms: ..." means fix the call, don't retry it raise ValueError r.json "error" Auto works as before here. A field it can't accept still comes back as HTTP 200 with status: 400 and the reason in error , before any attempt and at no cost. Billed and spent credits, side by side The Budget view under Metrics now leads with billed credits, the ones only successful requests use, and shows everything spent right underneath as "of N spent". Cost per request is in billed credits too. And the breakdown by key, domain or client IP has a sortable column for each number. Spent counts every outcome, because a failed request still takes an exit or a browser to find out it failed. So the gap between the two columns shows, domain by domain, how much work went into requests that didn't succeed. Signing out means signed out Change your password in Settings, reset it from the sign-in page, or press Sign Out Others, and every other browser signed in to your account is signed out. The one you made the change from stays in. Signing out now ends that session on our side as well, not just in your browser. The password reset page now asks for the same verification step as sign-in and sign-up. It stays out of sight unless it needs you. Proxy Finder and gateway redirects Some exits sit behind a company network gateway. The previous digest covered the kind that answers with its own login page. Proxy Finder now spots the kind that answers with a redirect to its sign-in service too, counts it as a failed attempt and moves on to another exit, instead of handing that redirect back as the site's answer. Smaller Dashboard fixes - When the owner you've picked Everything, Personal or an organization has no API keys yet, Overview and Metrics say so under the period and point you to create one. - Every number on Overview sits right under its label in one size, so a row of panels reads straight across at any width. The Auto attempts panel is always shown, reading zero for a period without Auto calls. - Subscribe drops the Monthly/Annual switch for a plan that's sold only one way. - On a trial we set up for you, choosing a different plan on the upgrade page now asks for a card and starts a full period that day. The trial ends once the payment goes through. Under the Hood Calls to api.foura.ai and eu.api.foura.ai are now answered by several servers at once instead of one. If one stops answering, new calls go to the others within a couple of minutes. Auto got lighter. Picking a saved session for a call takes four internal lookups where it used to take dozens, and every server runs more Auto capacity than before, so Auto calls spend less time queued behind each other at busy hours. Every engine moved to the current long-term-support runtime with October's security release, and we patched libraries across the API, the Dashboard, the docs and the blog. On Single and Proxy Finder the new runtime keeps one exit that stops reading mid-request from slowing down the other requests on its server. Before a server restarts for a new version, it stops receiving new calls, and it rejoins once it passes its checks. Retries belong to the sites that make you wait, not to a typo in your own request.