Why a Campaign Keeps Flipping Learning/Limited

What to check before you touch anything

How many actual changes have been made to the campaign over the last month? Flip-flopping between statuses is almost always a sign of frequent edits (bids, budget, target, conversion actions, structure), not the algorithm "behaving erratically" on its own.
Who's making these changes, and how often? If multiple people (or a person plus automated rules/scripts) are editing the campaign independently, learning can keep resetting without anyone realizing there's a single underlying cause.
How much conversion volume does the campaign have? At low volume, the system itself is inherently less stable and can flip between statuses even with no outside changes, simply from a lack of data.
Do the status swings line up with seasonal spikes in demand? A sudden, temporary shift in traffic volume (not a human action) can also trigger a move between Learning and Limited.
Are there automated rules or scripts adjusting budget or bids on a schedule? Their impact is often overlooked when diagnosing this, since they don't feel like "manual intervention."

If frequent manual edits are the cause, the most direct fix is agreeing on a minimum interval between meaningful changes (say, not touching budget/target more often than every 1–2 weeks) to give the strategy time to complete its learning cycle.
If multiple specialists are editing independently, it often helps simply to set up a single shared change calendar/checklist with a clear owner, to rule out unintentional duplicate edits.
Google generally recommends minimizing changes specifically while a campaign hasn't yet exited "Learning" — this isn't a hard rule, but it's a frequent recommendation baked into the system's own prompts.
If the swings are driven by low data volume rather than team actions, instead of trying to "stabilize" things with frequent small tweaks, it's probably smarter to either widen reach (budget/keywords) or temporarily loosen the target to get the system more data for more predictable behavior.
If you find automated rules or scripts regularly changing campaign parameters, it's worth explicitly deciding whether they're compatible with smart bidding logic on a target, and either disabling them or limiting them to parameters that don't affect learning.