The roommate phase has a precise feel to it: same bed, same mortgage, a calendar full of logistics — and conversations that could be conducted between polite colleagues. When couples land here, both partners privately run the same investigation: who changed? Who stopped trying first? The most useful answer research has to offer is that the question itself is malformed.

The loop nobody chose

Family therapy's core contribution to understanding stuck couples is the systems view: partners aren't two independent people exchanging behaviors, they're a feedback system — and systems produce outcomes neither member chose. Watch a typical loop run. One partner asks for something less warmly than before (they're tired). The other hears criticism in the flatness and engages a little less (self-protection). The first reads the distance and asks more sharply next time (alarm). The second confirms their read — see, hostility — and withdraws further. Two reasonable people, each responding sensibly to the other's response, jointly manufacture a result both of them hate. Nobody drove. The loop drove.

This is why the roommate phase almost never traces back to a betrayal or a decision. It grows out of a thousand tiny moments of exhaustion, each one individually innocent. And it's why the standard remedies misfire: "who started it" is unanswerable because loops are circles, and insight alone changes little because understanding a loop doesn't exit it.

What it looks like day to day

Kisses that became pecks — brief enough to carry no information except "we still do this." Conversations that are 90% household operations. Evenings that end with two phones and two backs. A vague loneliness that makes no sense on paper, because the person is right there. If several of these are familiar, you're not reading a description of a dying relationship. You're reading a description of an unattended system — which is a different diagnosis with a much better prognosis.

The property that should give you hope

Feedback systems have one exploitable weakness: change any single element consistently, and the loop cannot continue as before. Not "might not" — cannot. A system with one changed input produces changed output, by definition. That is the entire theory of one-sided repair, and it's why it isn't naive. You're not trying to change your partner — that's the doomed project everyone attempts first. You're changing the inputs their behavior feeds on: sharp asks replaced with soft ones, withdrawal with steady presence, criticism with deposits. The system reorganizes around the new inputs. Slowly, then noticeably.

Tonight

Pick one input and change it once. The classic: at the next hello or goodbye, hold the kiss for six full seconds — long enough to stop being logistics, short enough to be safe. Then change the same input again tomorrow. Consistency, not size, is what loops respond to. If you'd rather not improvise the sequence, Amendi turns this exact science into one 60-second mission a day for the partner working alone; the first week — the 7-Day Restart Kit — is free.

The loop got you here without anyone deciding. One person deciding is how you leave.