The input-output machine produces behaviour from inputs, and the most powerful inputs are environmental. The room, the time pressure, the presence of authority, the physical accessibility of options---all of these load responses before character has a chance to weigh in. This is why the same person is generous at home and ruthless in the office, or courageous in training and paralysed in an unfamiliar operational environment. The person hasn't changed; the inputs have.

The military principle of "train as you fight" is, underneath its practical wisdom, a statement about neural architecture. Pathways learned in one context retrieve in that context. If you learn ethical decision-making in a calm classroom, those pathways are encoded with calm-classroom cues. Under operational stress, those cues are absent, and the pathways don't fire. The behaviour you need must be practised under the conditions where you'll need it.

So what can you do? Audit the situation, not just the person. When someone behaves badly, ask what the situation made easy before asking what the person's character is like. And when you want better behaviour, design the situation: make the ethical option the obvious one, reduce the barriers to the right choice, and ensure the environmental inputs---supervision, time, resources, options---support the behaviour you want.

When you enter a new situation, the prediction engine immediately starts loading predictions about what's expected, what's normal, what will happen next. These predictions are built from every previous experience of similar situations. Walk into a library and the system predicts quiet; walk into a football stadium and it predicts noise. Walk into a unit where corner-cutting is routine and the system predicts---and prepares for---corner-cutting.

The predictions don't just describe what will happen; they shape what you do. If the system predicts that everyone else is bending the rules, rule-following feels abnormal and costly. If the system predicts that the authority figure expects compliance, questioning feels risky. The situation loads the prediction, and the prediction loads the behaviour.

So what can you do? Change what the situation predicts. If you want people to question bad orders, create situations where questioning is normal---practise it, reward it, make it part of the routine. The prediction engine will update, and once "people question things here" is the loaded prediction, the behaviour follows. But the update requires consistent experience, not just one-off training events.

The lazy controller conserves resources by defaulting to whatever action requires the least deliberation. In a well-designed situation, the default is the right thing: the form requires a signature, the procedure mandates a check, the layout makes the safe option the obvious one. In a badly designed situation, the default is whatever's fastest: skip the check, take the shortcut, follow the crowd.

This is why environmental design is so much more effective than exhortation. Telling people to "be more careful" asks the controller to spend resources overriding defaults. Redesigning the process so that care is built in---so that the careful option is the default---costs the controller nothing.

So what can you do? Design defaults that align with desired behaviour. Checklists, mandatory pauses, physical layouts that route people through review steps, supervision at key decision points---all of these make the right action the easy action. Don't rely on people to override bad defaults through willpower. Make the default the thing you want them to do.

When situations are loosely designed---when people have wide discretion, few constraints, and little oversight---behaviour varies enormously. Some people will act well, some badly, and the difference between them is mostly noise: mood, fatigue, recent experiences, who happens to be in the room. Reducing noise through situational design---standardised procedures, clear expectations, consistent enforcement---makes behaviour more predictable. This is usually a good thing, because it reduces the worst-case outcomes. But it also reduces the best-case outcomes: tight design constrains heroism as well as villainy.

So what can you do? Match the design to the stakes. High-consequence situations---use of force, financial decisions, safety-critical procedures--- warrant tight design that eliminates noise. Lower-stakes situations can tolerate more discretion, because the cost of a bad outcome is lower. And be honest about the trade-off: tight design means less individual judgement, which means less adaptation to unusual circumstances. The goal isn't to eliminate discretion but to deploy it where it helps and constrain it where it hurts.