Dutton and Painter's finding, chapter 20, is the one that has to be understood before anything else in this chapter makes sense: attachment strength was predicted by intermittency and by power imbalance, not by severity.
That means the standard outsider question — how bad does it have to get before you leave — is the wrong measure. Severity is not what holds people.
The mechanism is a partial-reinforcement schedule with fear in it. Arousal from threat is relieved by the same person who produced it, and relief conditioned to the source of the threat produces attachment that intensifies after incidents rather than before them. Chapter 19's variable-ratio schedule is maximally resistant to extinction because the absence of reward is indistinguishable from an ordinary gap.
Three consequences follow and each is a defensive instruction.
First, chart the cycle rather than the incident. The incident is what gets discussed and it is the least informative thing available. What is diagnostic is a dated log across months: when warmth arrived, when it was withdrawn, what preceded each. What the log usually shows is that the warmth was not responsive to anything you did — which is the point, because a reward contingent on your behavior would be trainable and therefore escapable.
Second, the craving on separation is a physiological signature and not a verdict. People who leave report intrusive longing, preoccupation, and a pull to return that is strongest in the first weeks. That is withdrawal from an intermittent schedule, it has a course, and mistaking it for evidence that the relationship was good is the single most common reason people go back.
Third — and this is why the chapter has to say it plainly rather than mechanically — separation is the period of highest risk. Homicide and serious assault data across jurisdictions consistently show the weeks around leaving as the most dangerous, which means a departure should be planned rather than improvised, and chapter 350 is the chapter on that.
The counter-response is structured separation: a defined no-contact period long enough for extinction to begin, external support that knows the situation, and someone who holds the record so that the good periods do not overwrite the rest.
Arguments do not reach this. The bond is conditioned rather than reasoned, which is why just leave is not advice.
This counters Law 14.