Product Update: The Refine Dashboard Is Now 3× Faster

We spent the last few weeks doing the unglamorous work: measuring where Refine was slow, fixing what the measurements pointed at, and re-measuring in production until the numbers proved it. This post is the honest changelog.

The headline numbers

On our largest production workspace — 43 connected AWS accounts — here's the before and after, measured on the live product, not a benchmark rig:

Login to dashboard: 12.5 seconds → under 4 seconds.

API requests on first load: 146 → 28.

Security findings screen: from over 12 seconds to about 4 — and most of that remaining time is the data actually rendering, not waiting.

What was actually slow

The interesting part of performance work is that the villain is rarely where you'd guess. Our security screen felt slow, but the security code was fine — the time was going to a boot-time pattern where the app eagerly fetched a full report for every connected account before showing you anything. With one account, unnoticeable. With 43, that was roughly ninety network round-trips before first paint, on every login, whether or not you'd ever open most of those accounts.

The fix was philosophical as much as technical: the app now loads what you're looking at, and nothing else. Account lists load in one paginated sweep instead of per-account fan-outs, reports load when you select the month that needs them, and screens that only need a summary no longer pay for the full detail behind it.

Faster is worthless if it's wrong

Alongside the speed work, we fixed something we care about more: every surface now agrees on the same numbers. A few of you noticed that the anomaly count in your email digest didn't always match the dashboard, or that the month-end projection differed between screens. You were right, and the cause was subtle: different surfaces had grown their own counting and forecasting code, and near-duplicate logic drifts.

Those calculations now live in exactly one shared engine each — one place that decides what counts as an anomaly, one that computes the month-end projection, one that sums open-recommendation savings. The digest, the dashboard, and the reports all call the same code, so they can no longer disagree. We verified the fix against live production data before shipping it.

Smaller things you'll notice

The security screen got a redesign along the way — severity cards are now one-tap filters, large accounts page through findings smoothly instead of loading everything at once, and loading states announce themselves properly to screen readers.

The "data processing" banner learned to stand down. It used to linger even after your month's report was ready; it now checks the actual report status for the month you're viewing.

Settings are now linkable. Tabs live in the URL, so you can bookmark or share a direct link to the exact settings screen you mean.

The measurement habit

None of this involved exotic engineering — it involved refusing to guess. Every change above started with a production measurement, shipped behind a verification, and ended with a re-measurement confirming the number moved. It's the same discipline we apply to your AWS bill: measure first, then act, then prove the saving landed.

If you haven't looked at Refine lately, log in — the difference is most visible on workspaces with many accounts. And if you're not connected yet, connecting now takes about two minutes.