The mine can be thought of as a rooted tree, with one node per cave, the root being node (1), and each tunnel corresponding to an edge.
Assuming (N \gt 1), Minerva's route will consist of driving along the path from node (1) to some other node (a), drilling a tunnel to some other node (b) (potentially with (b = 1)), and finally driving along the path from node (b) to node (1). Any given node pair ((a, b)) is valid if and only if no undirected edge is traversed multiple times (in other words, the path from node (1) to node (a) shares no edges in common with the path from node (b) to node (1)). For such a path, the total weight of gold ore collected is equal to the sum of (C) values across all nodes which are part of at least one of the two paths.
We can observe that a choice of ((a, b)) is valid if and only if either (b = 1), or (a) and (b) lie within the subtrees of different children of node (1). If they were to both lie within the subtree of the same child (c) of node (1), then both paths would include the undirected edge between nodes (1) and (c), which would be invalid.
We can also observe that the only node which will be part of both paths is node (1), and that the direction of each path is irrelevant (with both paths essentially being interchangeable). If we let (F(i)) be the sum of (C) values on the path between node (1) and node (i), then the total weight of gold ore collected will be (F(a) + F(b) - C_1).
Let (G(i)) be the maximum value of (F(j)) for any node (j) in node (i)'s subtree (including (i) itself). We can recursively compute the (F) and (G) values for all (N) nodes in (O(N)) time, based on the recurrences (F(i) = F(P_i) + C_i) (where (P_i) is node (i)'s parent) and (G(i)) equalling the maximum of (F(i)) and of (G(c)) (for each of node (i)'s children (c)).
Putting everything together, if node (1) has fewer than two children, the final answer will be (G(1)), while the answer will otherwise be (G(c_1) + G(c_2) - C_1), where (c_1) and (c_2) are the two distinct children of node (1) which yield the largest (G) values. It's also possible to arrive at a similar result without explicitly computing (F) and (G) separately.