Lately I’m seeing way more volatility than usual—between random outages, “AI enhancements” toggling themselves on, and the whole Andromeda era messing with budget distribution. If your results suddenly tanked, here’s a short checklist that’s helped me stabilize two accounts this week: **1) Re-confirm what’s actually serving.** Open your top ad sets and check each live ad’s preview + destination. I’ve seen FB start showing the wrong image/video via “related media,” which obviously confuses users and trashes CVR. If you spot an off-creative, duplicate the ad with the exact asset you want and re-publish. (Saw a thread calling this out here recently—worth a look.) **2) Kill stealth “AI boosts” during testing.** While you’re isolating variables, turn off any auto-enhancements / Advantage+ style helpers so one rogue toggle doesn’t bias the test. There are a few posts here ranting about forced AI features—honestly, I get it. **3) Assume the platform’s having a day.** If delivery nosedives or metrics freeze, check status communities before you start tearing down structure—some of the recent dips were just… outages. Don’t overreact if it’s a platform hiccup. **4) Don’t let Andromeda pick one “hot” ad and starve the rest.** When I need exploration, I’ll go ABO with 1–2 ads per ad set and guardrails (min spend caps or CPA caps) so the system can’t funnel 90% to a single creative in hour one. Multiple folks here have shared similar experiences. **5) Re-QA the whole path on mobile.** Obvious but easy to skip: thumb-scroll the ad → LP → cart on a real device. Check load times, stock/variants, default shipping, and pixel/Conversions API events firing cleanly. **6) Refresh hooks, not just looks.** If performance faded, I rotate *angles* before I rotate *formats.* New first 2–3 seconds, tighter promise, harder proof (UGC quote, stat, demo). Same template, different hook. **7) Protect your winners with clones.** When a top ad fatigues, I’ll clone it with a fresh intro line and slightly different thumb/first frame. Keep 80% the same; change the scroller’s reason to stop. **8) Separate broad “scale” from “testing.”** Keep a tidy testing campaign where losers can die fast without poisoning your scale set. It’s tempting to dump 50 ads into one big ASC and “let it learn,” but that’s how exploration dies. Plenty of discussion here on this exact pain. **9) Check audience expansion vs. fixed targeting.** When diagnosing swings, I’ll temporarily lock gender/geo/age or turn off expansion to see if the algo wandered. Several members reported better control with expansion off during tests. **10) Read the comments (but faster).** Often the fix is hiding in plain sight: repeat pre-purchase objections in your comments. I skim for patterns (confusion about variants, sizing, bundle rules, etc.) and then address those head-on in the ad copy and first reply. On that last point, I’ve been using PostInsight ai as a quick helper to analyze recent Facebook posts and bubble up comment themes I should answer in ads. It’s not a magic bullet, but it’s useful for post-level feedback and for drafting on-brand reply suggestions when a thread’s blowing up—saves me a bunch of time on busy days. Curious how everyone else is forcing exploration post-Andromeda without burning budget—any structural tweaks you’ve liked (min spends, ABO vs. CBO, separate creative sandboxes)? Also, anyone found a clean way to stop “related media” from sneaking the wrong asset into live ad delivery?