#include<bits/stdc++.h>
#define fi first
#define se second
#define all(v) begin(v), end(v)
#define rall(v) rbegin(v), rend(v)
#define compact(v) v.erase(unique(all(v)), end(v))
#define sz(v) (int)(v).size()
#define endl "\n"
#define dbg(x) "[" #x " = " << (x) << "]"
using namespace std;
template<class T> bool minimize(T& a, T b){
if(a > b) return a = b, true;
return false;
}
template<class T> bool maximize(T& a, T b){
if(a < b) return a = b, true;
return false;
}
typedef long long ll;
const int MAX = 2e5 + 5;
const int lg = 17;
int N, Q, H[MAX];
int l[MAX][lg+1], r[MAX][lg+1], nxt[MAX][lg+1], spt[MAX][lg+1];
int combine(int i, int j){
return H[i] < H[j] ? j : i;
}
void init(int _N, vector<int> _H){
N = _N;
for(int i = 0; i < N; i++) H[i+1] = _H[i];
stack<int> st;
for(int i = 1; i <= N; i++){
while(!st.empty() && H[i] > H[st.top()]) st.pop();
l[i][0] = st.empty() ? -1 : st.top();
st.push(i);
}
while(!st.empty()) st.pop();
for(int i = N; i >= 1; i--){
while(!st.empty() && H[i] > H[st.top()]) st.pop();
r[i][0] = st.empty() ? -1 : st.top();
st.push(i);
}
for(int i = 1; i <= N; i++){
spt[i][0] = i;
nxt[i][0] = -1;
if(l[i][0] != -1 && r[i][0] != -1) nxt[i][0] = combine(l[i][0], r[i][0]);
else if(l[i][0] != -1) nxt[i][0] = l[i][0];
else if(r[i][0] != -1) nxt[i][0] = r[i][0];
}
for(int j = 1; j <= lg; j++){
for(int i = 1; i <= N; i++){
l[i][j] = (l[i][j-1] == -1 ? -1 : l[l[i][j-1]][j-1]);
r[i][j] = (r[i][j-1] == -1 ? -1 : r[r[i][j-1]][j-1]);
nxt[i][j] = (nxt[i][j-1] == -1 ? -1 : nxt[nxt[i][j-1]][j-1]);
if(i + (1 << j) - 1 <= N){//sparse table max
spt[i][j] = combine(spt[i][j-1], spt[i + (1 << (j-1))][j-1]);
}
}
}
}
int get(int l, int r){
if(l > r) return -1;
int k = __lg(r - l + 1);
return combine(spt[l][k], spt[r - (1 << k) + 1][k]);
}
int minimum_jumps(int a, int b, int c, int d){
a++, b++, c++, d++;
int answer = 0;
int idx_max_cd = get(c, d);
int idx_max_bc = get(b + 1, c - 1);
int limit_end = H[idx_max_cd];
int limit_middle = (idx_max_bc == -1 ? -1 : H[idx_max_bc]);
int st = b;
for(int j = lg; j >= 0; j--){
if(l[st][j] != -1 && l[st][j] >= a && H[l[st][j]] < limit_end){
st = l[st][j];
}
}
for(int j = lg; j >= 0; j--){
if(nxt[st][j] != -1 && H[nxt[st][j]] < limit_middle){
answer += (1 << j);
st = nxt[st][j];
}
}
auto inside = [&](int l, int r, int x) -> bool{
return l <= x && x <= r;
};
// H[st] > H[middle] -> try to jump in 1
if(r[st][0] != -1 && inside(c, d, r[st][0])) return answer + 1;
// H[st] < H[middle] -> try to step back one and then go for the clutch
// more than 1 step back are useless since the more we go left, the value will increase -> out of bound [c, d]
if(l[st][0] != -1 && inside(c, d, r[l[st][0]][0])) return answer + 2;
//st --> middle --> r[middle][0]
for(int j = lg; j >= 0; j--){
if(r[st][j] != -1 && H[r[st][j]] <= limit_middle){
answer += (1 << j);
st = r[st][j];
}
}
return inside(c, d, r[st][0]) ? answer + 1 : -1;
}
#ifdef LOCAL
void fastIO(){
ios_base::sync_with_stdio(NULL); cin.tie(0);
freopen("task.INP", "r", stdin);
freopen("task.OUT", "w", stdout);
}
signed main(){
fastIO();
init(7, {3, 2, 1, 6, 4, 5, 7});
cout << minimum_jumps(0, 1, 2, 2);
}
#endif
