The two versions have different constraints on (k), (c). Solving one of the two versions does not necessarily solve the other. You may want to read both versions. Hacks are disabled in both versions. This is an interactive problem. During the time of the Ottoman Empire, many scientists were at work. You decided to study their daily life and stumbled upon an interesting game that was popular at that time. One scientist would think of an integer (x), such that (1 \leq x \leq c). Another scientist would try to guess it. He could specify the base of the numeral system (2 \leq b \leq c) and would receive as a response the sum of the digits of the number (x) in base (b), if (x \geq b), or (-1) if (x < b). You wrote a program that can think of a number (x) and respond to the question. Now you want to play with it and learn to guess using (\leq k) queries. The first line contains three integers (t), (k), (c) ((1 \leq t \leq 10^4), (\mathbf{k = 4}), (\mathbf{c = 4 \cdot 10^{18}})). You must guess the number from (1) to (c) (t) times, using (\leq k) queries. All (t) games will be played consecutively and are independent. To make a query, you must output ? (b) for some (2 \leq b \leq c). Then you must read the response — an integer (a) ((-1 \leq a < c)) — the response to the query. To provide the guessed number, you must output ! (x) , where (1 \leq x \leq c) — your answer. Then you must read the result (r) ((0 \leq r \leq 1)). If (r = 0) — your answer is incorrect, you must immediately terminate the program in this case. Otherwise, if (r = 1) — your answer is correct, you can proceed to the next game or terminate the program if (t) games have already been played. It is guaranteed that the interactor is not adaptive. That is, all chosen numbers are fixed in advance and do not change during the games. After printing each query do not