Constraint Stacking

Constraint stacking refers to loading a single prompt with more and more individually reasonable requirements, expecting them to add up to a precise description of what you want — but past a certain point, these conditions start pulling against each other, and Claude can't satisfy all of them at once, so it makes trade-offs you may not even know it made. This differs from the normal practice of writing requirements clearly — in the ordinary case, more detail usually helps. Constraint stacking describes the point where that stops being true: when conditions carry real tension against each other (concise versus complete, professional versus warm, fast versus careful), the more you stack, the harder it gets for Claude to judge which one to prioritize, and the output becomes less stable rather than more precise.

This happens because when people prepare a prompt, they typically add one requirement at a time as it comes to mind, rather than checking upfront whether those requirements conflict with each other. Think 'needs to be professional,' add a line. Think 'shouldn't be too stiff,' add another. Think 'keep it short,' add one more — each addition feels reasonable at the moment, because each one is, in fact, true on its own. Nobody pauses mid-process to ask whether 'short' and 'cover every key point' might inherently be in tension with each other. Only once the output comes out looking slightly off does anyone go back and notice the prompt had accumulated seven or eight requirements, some of which flatly contradicted each other, and Claude had no choice but to pick one way of resolving that contradiction — a resolution that isn't necessarily the one you wanted.

In practice there are two ways to check for this. First, after writing the prompt, list out every constraint and check each pair for tension — common conflicting pairs are 'concise' versus 'cover every detail,' 'professional and formal' versus 'warm and natural,' 'give a quick conclusion' versus 'fully argue every step.' If a list turns up this kind of opposing pair, the prompt is already asking Claude to do something impossible to satisfy at once, and a priority needs to be set — you can't have both equally. Second, if the conditions genuinely have a priority order, state it explicitly in the prompt — which one leads, which ones are secondary — rather than laying all conditions out in the same flat tone and letting Claude guess which one to sacrifice. Without a stated priority, Claude's trade-off may not even be consistent — ask the same prompt twice and the priority it lands on can differ each time, which is exactly the kind of instability constraint stacking tends to produce.

For you, the moment constraint stacking is most likely to happen is when a prompt keeps growing longer, one more small requirement tacked on each time it comes to mind — that kind of cumulative writing feels like it's making the instruction more precise, but it may actually be quietly burying conditions that contradict each other. The habit worth building is spending ten seconds after writing a prompt to check through every condition and ask yourself: if all of these were genuinely satisfied at once, could such a thing even be written? If the answer is no, there's a conflict in the list that needs a stated priority, rather than everything being crammed in and left for Claude to sort out on its own. Worth noting: this problem doesn't get solved by adding more conditions. When output falls short, the instinctive reaction is often 'let me add one more clarifying line,' but if the root cause is that existing conditions already contradict each other, adding a new condition usually just deepens the conflict rather than sharpening the result.