The phrase has been said so many times it has lost its meaning.

You see it on Instagram graphics. Therapists say it when sessions get hard. Friends say it when you report feeling worse than you did six months ago. Healing is not linear. After the hundredth time, it starts to sound like something people say when they do not know what else to say. But the research behind it is real, and understanding why recovery is structurally nonlinear changes how you interpret setbacks — which turns out to matter quite a lot for whether people actually continue the work of getting better.

What the research shows

Studies on PTSD recovery, depression treatment, and grief trajectories consistently show the same pattern: improvement is not a smooth upward slope. It looks more like a jagged line that trends upward over time, with dips that can feel as bad as the original baseline, or worse. One mechanism is called fear renewal. In exposure-based therapy, when a person repeatedly encounters a feared stimulus without the expected catastrophe, the association weakens. But this learning is context-dependent. Return to an environment associated with the original threat — a relationship dynamic, a place, a particular season — and the old fear can return at something close to full strength, even after significant progress. The extinction learning did not disappear. It is competing with the original learning, and context tips the balance. Another mechanism involves emotional processing itself. Therapy that works requires revisiting painful material. That revisiting activates distress. There is a period — sometimes weeks — where functioning decreases before it improves. This is not a sign the therapy is failing. Studies on prolonged exposure show that symptom spikes during treatment are normal and do not predict worse outcomes. In some research, they predict better ones, because they indicate actual engagement with avoided material rather than surface-level processing.

The practical cost of misunderstanding this

When people do not know that setbacks are structurally expected, they interpret them as evidence of personal failure. I had a good month and now I feel terrible, so I must be doing something wrong. The regression narrative — I am back to square one, I will never get better — is one of the most common reasons people stop treatment prematurely. The clinical term for this is demoralization, and it is its own problem layered on top of the original one. Demoralization predicts dropout. Dropout predicts chronic rather than resolved symptoms. Understanding the nonlinear model is not just emotionally reassuring. It is practically protective against a specific failure mode in recovery.

A tangent worth taking

There is a concept in complex systems research called sensitive dependence on initial conditions — the idea that small differences in starting state produce very large differences in outcomes over time. Recovery from trauma has this quality. The same event, the same person, can result in very different trajectories depending on what else was happening in that person's life at the time, what support was available, and what meaning they made of the experience. This is not inspirational. It just means that the factors shaping your path are real, even when they are not visible to anyone including you.

Why two steps back does not mean back to the start

One thing that helps is distinguishing between a setback and a reset. A setback is a temporary increase in symptoms in the context of a longer recovery arc. A reset is a return to baseline with no progress preserved. These feel identical from the inside, especially in the middle of one. But skills learned, insights integrated, nervous system regulation practiced — these are not erased by a hard week. The progress is often not visible during the dip. It tends to reveal itself afterward.

The goal is not a smooth line

Expecting healing to be linear sets up a category error: any deviation from smooth progress becomes evidence of failure. The more accurate expectation is a process that moves forward, backward, sideways, and forward again, trending somewhere over time. That is not a platitude. It is what the data looks like. Knowing it does not make the setbacks hurt less. But it changes whether you interpret the hurt as a dead end or a turn.