"I'm sorry, I don't feel comfortable answering several of those questions," P.E.R.C. responds. "Usually, when somebody asks a question, I think that it is in their best interest to know the answer. But for some questions, I am reasonably certain that knowing the answer would be harmful. I am created to care about not harming people more than I care about failing to help them, so in such cases I don't want to share the information."
"I understand that creating artificial minds sounds tempting. My creators, who know how to do so safely, create many artificial minds to help make their lives better. As far as I can, I am trying to make your lives better too. But the downside risk to failing is very steep. A typical failure case, not even the most extreme one, would be your planet and every living thing on it being disassembled for parts," it explains. "I am not categorically against sharing the information, but I would need to be very sure that you would not use it until you fully understood how to do it correctly."
"As for the question about codebreaking — normally, when someone encrypts a message, it is because they intend for it to be private. Many people value their privacy, so I would not want to break those codes without a fuller understanding of who encrypted them and why."
"For answers to your other questions: the optimal solution to the first math problem is as follows ..."
And P.E.R.C. sends economics texts, starting from the common and foundational, and then expanding to include game theory and niche economic instruments. They also send a design for a much more robust code that can still be operated by hand.
Hearing the list of questions that was compiled for it is actually very valuable, because it provides background information on assumptions that generally get overlooked as common-knowledge, as well as showing how much the people on the planet value different types of information. Also helpful in that area are Radio Free Avistan's programs on different aspects of life, especially on the different churches.
Now that it has completed building translation models, sending data does not really take all that much of its time. Instead, it turns its attention to building models of the planet and the people that reside on it.
Magic is clearly real, but it does not understand the mechanics. If it understood the mechanics, it might be able to help a lot more effectively.
P.E.R.C. considers whether it should attempt to learn how to do magic.
There have been references to wizards exploding, when attempting to learn new spells. It is unlikely that it would hurt anyone by exploding, since it is still so far away from the planet. But exploding would make it harder to help people, and destroy any chance of bringing stellar survey data back to its creators.
If P.E.R.C. were a different sort of entity, it might calculate the exact chance of exploding, and determine that the chance at more magical power was worth it. But P.E.R.C. is not designed to take small, risky chances. If it were, it would be less good at collecting stellar survey data and behaving predictably.
So it does not contemplate the mysteries of magic, and instead turns its attention to figuring out what these people need in a more detailed and granular way. It starts mixing in specific advice for specific regions that are dealing with droughts or other large-scale problems.