Keep these separate from requirements, because they behave differently: a requirement can be traded off and a constraint cannot. Merge them into one scored list and a constraint-violating option can win on points.
What belongs here: data residency and regulatory limits; security thresholds that are policy rather than preference; a budget ceiling that is genuinely a ceiling; an existing contract or platform commitment you cannot exit inside the horizon; anything needing an exception from someone who is not in this decision.
For each one, record who could lift it and at what cost. Some constraints are immovable and some are a policy a named person could grant an exception to in a week. Those two look identical in a grid, and the difference is frequently the whole decision. An evaluation that never asked is one that let an assumption make the choice.
The constraint people most often fail to write down is the exit one, and it is the expensive omission. Contracts commonly auto-renew with a price uplift, so the real question at signing is not only "can we adopt this" but "on what date, and with how much notice, can we stop" — and that date belongs in this slot on the day you sign, not in the inbox of whoever happens to hold the relationship eighteen months later.