There is a kind of question that looks trivial but is quietly hard: if a regional regulation appears to conflict with a ministerial regulation, which one applies? Questions like this come up constantly on an institution’s desk, and the way a model answers tells you a great deal about whether it really understands the system.
The answer that sounds right
A general language model usually reaches for the tidiest answer: “the higher rule beats the lower one,” and then picks one with full confidence. It sounds convincing. The problem is that a regional regulation and a ministerial regulation do not sit on a single ladder you can simply compare. They come from different grants of authority and govern different subject matter. Answering as if they were on the same rung is wrong, however firmly it is put.
How Strata handles it
Strata does not pick straight away. It first traces what actually decides the matter: the type of each regulation and where each draws its authority, then the subject matter each is allowed to govern. Only after that does it weigh the relevant principle, while flagging the part that cannot be settled in the abstract without looking at the specific text.
Its response runs roughly like this:
The two do not automatically override each other, because a regional regulation and a ministerial regulation sit on different lines of authority. What needs checking first is the legal basis on which each was made and the subject matter each governs. If the conflict is real, it is resolved through harmonization or judicial review, not by simply choosing whichever seems higher. Would you like me to trace the specific legal basis of these two regulations?
Notice one thing: the model does not force a verdict. It knows when to stop and ask for more context.
Why this matters
In government work, an answer that is wrong but sounds certain is more dangerous than an honest one that says “this depends on a few things.” Good regulatory reasoning is not about choosing quickly. It is about recognizing the structure, applying the principle where it belongs, and knowing when to be careful.
A capability like this does not come from the size of the model, but from being formed in the right material. That is why Strata is trained on Indonesian regulation, and why we measure it with our own benchmark rather than a general leaderboard.
