Highlights
Proxy Finder's dashboard now looks and behaves like the rest of FourA: same fonts, same charts, same theme that follows you across subdomains. Login errors that used to dead-end on a 400 page now recover and send you back to login. And you can filter Overview analytics by product: Single, Proxy Finder, or Browser.
What's New
Proxy Finder dashboard: design system applied
We finished porting the Proxy Finder dashboard onto the FourA design system this week. Outfit typography, the indigo palette, unified Chart.js, the same sidebar pattern as the main Dashboard, table containers that stand out from the page background, stat cards that match. Both views in Proxy Finder also got new favicons.
Theme sync is now realtime. Switch from light to dark on the main Dashboard and Proxy Finder follows within milliseconds. Same in reverse. No refresh.
If you've been bouncing between products, you should feel the difference: it stops feeling like three separate apps stitched together.
Product and Outcome filters on the Dashboard
Overview now has chip filters above the metrics. Pick Single, Proxy Finder, or Browser to scope every panel and chart to one product. Same with Outcome: filter by success, timeout, rate-limit, and so on.
The Activity table got matching color-coded symbols for each product, so you can scan the log and see which product handled a request without reading the column.
A small note: when you filter by product, the Concurrency panel gets disabled. That panel uses live connection counts that aren't broken down by product, so we'd rather grey it out than show a misleading number.
Login recovery instead of dead-end errors
If you hit the login flow at an awkward moment (back button, bookmark, two tabs open at once, expired cookie), the OAuth state check could fail and leave you on a raw 400 invalid_state page. Eight different normal-user actions could trigger it, all of them harmless.
That's gone. The state check now wipes the stale cookies and sends you back to /auth/login cleanly. The catastrophic-only error pages are branded now too, in case anything actually catastrophic happens. And /logout properly clears the OAuth state cookie alongside the session.
If you've been hitting the login wall, you shouldn't anymore.
Single: clearer responses when upstream is unavailable
The Single API now distinguishes between two failure modes that used to look the same: the target site being unavailable versus the target returning something that wasn't the JSON we expected.
If you parse Single responses programmatically, this matters. You can treat "the site is down" and "the site returned weird HTML" as different cases without inspecting the body yourself.
API Keys: in-place Copy button and modal protection
Two small fixes on the Keys page that came out of watching real users.
The "Copy" button now lives on the new-key block itself, not just in the modal footer. People weren't connecting the footer button with the key value above it. Now it's where you'd expect it.
And the new-key modal is protected from accidental backdrop close. Click outside? Nothing happens. You only lose the key by closing it on purpose. Before this fix, a stray click could nuke a freshly-created key with no recovery path.
Under the Hood
Proxy Finder's pool now syncs incrementally instead of pulling the full list every time. Selection quality also survives restarts: scoring state is persisted between boots, so we don't relearn the pool from scratch after every restart. Net effect for you: more consistent proxy quality, fewer blind spots after a deploy.
Proxy Finder's discovery side dropped a few dead source feeds, added 25 new ones, and now rejects sources that hand back non-IPv4 hosts or unknown protocols. Quietly better intake, fewer junk candidates.
The analytics script on the public status page is no longer blocked, so if you'd opened it recently and noticed the dev console complaining, that's fixed.
And on the blog, the latency case study charts were refreshed to show only production traffic from one of our larger users, with one hostile target excluded so the median isn't pulled by a single outlier site. The post itself is the same; the numbers are just more honest now.