### shishyando's blog

By shishyando, history, 14 months ago, translation,

We are really sorry to make the round unrated. Anyway, we hope that you enjoyed the problems!

1497A - Meximization

Idea: shishyando

Tutorial
Implementation

1497B - M-arrays

Idea: Artyom123

Tutorial
Implementation

1497C1 - k-LCM (easy version)

Idea: shishyando

Tutorial
Implementation

1497C2 - k-LCM (hard version)

Idea: isaf27

Tutorial
Implementation

1497D - Genius

Idea: shishyando

Tutorial
Implementation

1497E1 - Square-free division (easy version)

Idea: Artyom123

Tutorial
Implementation

1497E2 - Square-free division (hard version)

Idea: isaf27

Tutorial
Implementation

• +334

 » 14 months ago, # |   +29 SpeedForces
 » 14 months ago, # | ← Rev. 2 →   +7 Point distribution was weird imo but Questions were really good tho.
 » 14 months ago, # |   +22 For C2 — legend+ary
 » 14 months ago, # | ← Rev. 2 →   +15 Problem A — Implementation basedProblem B — Implementation | Constructive (Great Problem)Problem C1,C2 — Intuitive | Constructive (Make Cases and you are done with both problems)Problem E1 — Simply Math | Implementation (Prime factorization)A great Round shishyando and Artyom123 with great and interesting problems. Hope to see your next round soon !!
 » 14 months ago, # |   +3 E2 and D are interesting problems overall.
•  » » 14 months ago, # ^ |   -68 Lesser number of submissions doesn't make a problem interesting
•  » » 14 months ago, # ^ |   +2 I think so. I am not able solve the problems, but reading the resulotion helps me a lot in understanding usage of DP. Love this round so much!
 » 14 months ago, # |   +2 Really liked the problems.
 » 14 months ago, # |   +418 Though I'm a tester, I want to say that C1 is the very beautiful hint for C2! Without C1, C2 can be the hardest problem in the contest.
•  » » 14 months ago, # ^ | ← Rev. 2 →   -9 Agreed, I literally copied my code from C1, made minor changes, then submitted C2.
•  » » 14 months ago, # ^ |   0 Yes, The magic number of the problem, 3, seems rather arbitrary and unintuitive at first.
•  » » 14 months ago, # ^ |   0 lmao i gave 1.5 hours in upsolving yesterday and couldn't get C2 as i didn't look at C1 First. Now after reading the editorial i can see how intuitive it is.
 » 14 months ago, # |   0 Can someone explain in detail math behind C1?
•  » » 14 months ago, # ^ | ← Rev. 2 →   +11 It was simple first if n is odd then just print 1 n/2 n/2if n is even then check for if n/2 is even then print n/2 n/4 n/4 if n/2 is odd then print (n/2)-1 (n/2)-1 2we can extend this same logic in C2 by simply printing 1 , (k-3) times then taking n=n-(k-3) in C1
•  » » 14 months ago, # ^ |   -15 No math just you have to figure out basic logic that's it the one who did it can easily get through
•  » » 14 months ago, # ^ |   0 n can be of the form 4*k , 4*k+1 , 4*k+2 , 4*k+3 (for some k>=0) case 1: n = 4*k ans = k ,k, 2*k LCM = 2*k case 2: n = 4*k+1 ans = 1 ,2k ,2k LCM = 2kcase 3: n= 4*k+2 ans = 2 , 2k , 2k LCM = 2kcase 4: n = 4k+3 ans = 1 , 2k+1 , 2k+1 LCM = 2k+1
 » 14 months ago, # |   +72 I went straight to C(hard) without looking at C(easy) since I didn't want to be misled by some simple solution that only worked for simple testcases but not hard ones... instead I fumbled around for an hour looking for a general solution without realizing the k=3 case is all you need. That really backfired on me.
•  » » 14 months ago, # ^ | ← Rev. 2 →   +2 The scoring distribution: 500 — 750 — (750 + 500) — 1750 — (1500 + 1500) It was pretty much clear, C1 was supposed to be harder
 » 14 months ago, # | ← Rev. 4 →   +22 At the risk of my submission soon being systested, I think E2 can be solved greedily.Construct the greedy intervals from E1. Then, at most K times, consider all pairs of adjacent intervals. Compute the number of changes needed to merge all pairs of adjacent intervals in $O(n)$ and then just merge the best one, and update the array so that excess is put at impossible values.It works because in the solution you will never remove a value completely from an interval. So taking suboptimal merger will never help since it isn't actually going to reduce a merging cost.Since you always make at least one change, you repeat at most $k$ times so it is $O(nk)$. I think it can be optimised to $O(n)$.UPD: I think it cannot be optimised to $O(n)$, sorry. I thought that maybe you could save computation by only calculating the effect of the merger rather than recalculating every cost, but that's still $O(n)$. UPD2: WA47, I wonder if greedy is just wrong here or it's a problem with the code...UPD3: Assuming that you are merging intervals, the solution is optimal. However, it is also possible to split up an interval, so it does not work.
•  » » 14 months ago, # ^ |   +11 Can you explain a bit more in detail about splitting intervals? I had the same reasoning and (theoretical) solution as you as well as WA47 and would like to know what the logical error is.
•  » » » 14 months ago, # ^ |   +15 Say there are just three intervals originally. The greedy solution wants to subsume the middle interval into either the left or the right interval. However, it is also possible to subsume the left part of the middle interval into the left interval and the right part of the middle interval into the right interval. It is possible that this is less costly.
•  » » » » 14 months ago, # ^ |   0 That's perfect, thank you!
•  » » 14 months ago, # ^ |   0 what do you mean by best one? does it mean that bigger the interval, the better it is? or anything else? btw approach is nice. Thank you !!
 » 14 months ago, # |   0 Thank you for the problems!
 » 14 months ago, # |   +20 even despite unrated, this was the best round ive taken in a while. clean and concise problem statements. thank you so much for the round. :D
 » 14 months ago, # |   +7 Checkout this problem for a variant on problem A.
•  » » 14 months ago, # ^ |   0 I recalled the problem while giving the round today, lovely coincidence.
 » 14 months ago, # | ← Rev. 3 →   0 Video Tutorial B:https://www.youtube.com/watch?v=l9RQpnyIOXwVideo Tutorial A:https://www.youtube.com/watch?v=-CK_6lqs1hc&t=5s
 » 14 months ago, # |   +1 What are "constructive" problems? can anyone tell. how to become good at these?
•  » » 14 months ago, # ^ |   0 Basically, it is related to problems which is asking you to find any answer(of possibly many) that satisfies the constraints of the question. You can practice these here
•  » » » 14 months ago, # ^ |   +1 Thanks dude! I'll save this reply.
•  » » 14 months ago, # ^ |   0 Constructive Problems asks you to build something like array, graph or matrix which satisfies the given condition in problem. Usually these type of problems have multiple correct solutions and asks you to output any one. For problems you can refer this and this blog.
 » 14 months ago, # |   0 Even though I am just starting out, I must say I loved this contest! Great work shishyando
 » 14 months ago, # |   +23 "...it is easy to see that..."Writing that sentence is like begging for downvotes. Remember, editorials are written for those not able to solve the problem. Telling them how easy it is is no good idea.
 » 14 months ago, # |   +11 From problem D: "Then in binary form weight has its k-th bit set true if and only if j < k < i" Shouldn't it be j <= k < i ? Because we have for example 10000 — 00100 = 01100
•  » » 14 months ago, # ^ |   0 Seems true, thank you. Fixed now.
 » 14 months ago, # |   0 Thanks for the round, although I couldn't participate.Problem D is beautiful.
 » 14 months ago, # |   0 Is it ok that i got TL in E2 with O(nk log n + max(a_i)) complexity ?
•  » » 14 months ago, # ^ |   0 Is your max(a_i) per test case, or in total? If it's per test case, then it would be O(nklogn+max(a_i)t) which is too much
•  » » » 14 months ago, # ^ |   0 it's in total, I need it to precalculate prime numbers which are less than 1e7, I passed E1 that way.
 » 14 months ago, # |   0 Help needed 110230087 for problem 1497B - M-arrays
•  » » 14 months ago, # ^ |   +1 Your Solution is failing for testcase: 1 4 7 3 4 3 4 Your Output: 0 Correct : 1 Corrected 110257624,Just changed m/2 to (m+1)/2 in FOR loop
•  » » » 14 months ago, # ^ |   0 thanku
•  » » 14 months ago, # ^ |   0 You also have one more mistake.If i=0 => m-i=m, but m can not be a reminder by mod m. So you have to check v[i] and v[(m-i)%m].
•  » » » 14 months ago, # ^ |   0 no, I started i from 1.
 » 14 months ago, # |   +5 For me C2 strikes so fast even I hadn't thought of it...it just dropped from the sky :)Simply, considering 1 1 1 1 ... upto k-3 and the C2 becomes C1
•  » » 14 months ago, # ^ |   +4 I think if there was no C1 then people may have a hard time coming to this solution for C2.
•  » » » 14 months ago, # ^ |   0 yes :)))
 » 14 months ago, # | ← Rev. 2 →   0 I used binary search for C1 & C2, didn't think it can be done in O(1) too :( My solution Great contest though! <3
•  » » 14 months ago, # ^ |   0 I am curious while doing binary search how you got the idea that low is the ans instead of mid ??
•  » » » 14 months ago, # ^ |   0 See, the thing about binary search is that is always drifts towards the right answer. Similar thing is happening here.Initially, high = ⌈n/2⌉-1, low = 1 and during first iteration, mid = ⌈n/2⌉/2. Now if you look carefully, for first case (when n is odd) and for second case (when n is even but not divisible by 4) the answer is actually, (n-2*(⌈n/2⌉-1), ⌈n/2⌉-1, ⌈n/2⌉-1) and for third case(when n is even as well as divisible by 4) the ans is (n-⌈n/2⌉, ⌈n/2⌉/2, ⌈n/2⌉/2).So you must have already observed that for 1st & 2nd case, the ans is initial value of high, (n-2*high, high, high) and for third case ans is the initial value of mid, (n-2*mid, mid, mid)For the first two cases, the if condition will execute only once when mid=high, i.e., low=high=mid, and when this happens high will become less than low and it'll come out of the loop, so the ans will be low, [n-2*low, low, low] and for the third case, the if condition will execute only once, when mid is the ans, let's say value of initial mid is mid1, so for all further iterations, the if condition will never execute since high = mid1-1 and the value of low will keep on increasing until it becomes equal to high+1, i.e., low = high+1 = mid1, so again the ans is [n-2*low, low, low].
 » 14 months ago, # |   +1 The dp of E2 can be solved in O(nk). https://codeforces.com/contest/1497/submission/110252728
 » 14 months ago, # |   0 What is the reason for the unusual memory limit in Problem D?
•  » » 14 months ago, # ^ |   +5 I'm assuming the reason was to prevent the more obvious solution with a dp[n][n] table, which would need to store up to 25,000,000 ints = 100 megabytes.
•  » » » 14 months ago, # ^ |   +1 igz Can you please explain the dp[n][n] solution?
 » 14 months ago, # |   0 Why it is necessary to use map instead of unordered_map in 1497B - M-arrays? I submitted both versions but only map got AC but unordered_map one got WA on test 3. map version-> 110262091 unordered_map version-> 110262159
•  » » 14 months ago, # ^ |   -13 It's happening mostly because of collision, refer this to use a custom hash for unordered map. I've used the same in my submission.
•  » » 14 months ago, # ^ |   +15 Using cnt[i] may create a new element (cnt[i]=0) in map. And it may cause rehashing of unordered_map, iterators are invalidated (see "Iterator invalidation" here), and for-loop continues working with invalid iterators, which causes undefined behaviour. In usual map creating new element does not invalidate iterators, so UB won't happen. But u still got lucky that these extra zeroes in your map don't change result of your program.
•  » » » 14 months ago, # ^ |   0 thanks for the above link I learned a new thing...
•  » » 14 months ago, # ^ |   0 My solution by using unordered_map 110274693. I just avoid new insertion and deletion or such kind of invalidate iterator operations that mentioned in this Your text to link here... given by KostasKostil.
 » 14 months ago, # |   0 I really like how C1 is a hint for C2.
 » 14 months ago, # |   +3 C1 750ptsC2 500pts
 » 14 months ago, # |   +54 Problem E is similar to this problem.
 » 14 months ago, # |   0 If Only I submitted C1 after My C2 was accepted!
 » 14 months ago, # |   0 i am curious while doing C1,C2 problem how people were able to come up with the idea of for n%2 == 0 ans is n/2,n/2,2 and for n%4 == 0 ans is n/2-1,n/2-1,2 and so on.During the contest I made a pattern for 1 to 11 but i was unable to intutively guess this solution .Can anyone recommend the problems that i can do for increasing my intution for doing this type of problems thanks
•  » » 9 months ago, # ^ |   0 sorry for late response.I made a patter up to 23 before getting the right answer. My cases was if n is a multiple of 3 we done (n / 3, n / 3, n / 3), and it was difficult after that. I sais to myself it the limit is n / 2 i will try to be near to n / 2. and for each number i put n / 2 and see what can be done. I think the best is to go up to a limit (say 50) and if you don't see a pattern maybe it's not the right solution.
 » 14 months ago, # |   +3 There is an $O(nk)$ dp solution in Problme E2.At first, we can let $a[i]=mask(a[i])$. Assume $dp[i][j]$ denotes we consider the first $i$ positions, after making $j$ changes, the minimum answer we can get and the maximum position where the begining of the last segment at. So we can just maintain $last[i]$ which denotes the maximum position $j$ where $a_j=a_i$ (expecially, if $a_i=0,last[i]=i-1$) and make transfers.The key observation in the solution is that if we change the number $a_i=x$, and for the first $j>i,a_j=x$, we should let $last[j]=last[i]$. But actually, there is no need to do that, we can still get the right answer.See the code for details.
•  » » 14 months ago, # ^ |   0 Stuck in E1. In sample test case 1, should't answer be 2 ? Isn't it valid -> (18, 6, 4), (2, 1) ?
•  » » » 14 months ago, # ^ |   0 We can't change the order of the sequence while dividing.
 » 14 months ago, # |   0 Can anyone tell me how to reach to the solution for $C2$. I am asking this because $C1$ made a path for $C2$. So, can someone give some mathematical proof or logical way to approach such problem
•  » » 9 months ago, # ^ |   0 Achieving lcm of n / 2 lead intuitively (to me) to take n / 2 everywhere. now we can see that we can put n / 2 at most 2 time and i will remain a 1 case left.(that leads to k = 3 case). you can also think i put the same number some number of time and i put 1 everywhere since 1 does not increase the lcm. But i think it's very difficult to figure out alone without a hint.
 » 14 months ago, # |   0 for the tutorial for c2, (1+1+...+1)+(a+b+c)=(k−3)+(n−k−3)=n. shouldn't it be (k-3) + n-(k-3) or (k-3) + n — k + 3?
 » 14 months ago, # | ← Rev. 2 →   +13 Here is another $\mathcal{O}(nk)$ solution to 1497E2 - Square-free division (hard version).For the first part, after normalization, instead of left[i][j] which is a bit bothering, we only need to find pre[i], which is the rightmost index to the left of the current index that shares the same value as the current index. This is very simple.Then for the second part, for each change number j, we store best[j] denoting the minimal pair (segs,-last_start), and second_best[j] denoting the second minimal pair (segs,-last_start). Here last_start is the start index of the last open segment, and segs is the number of segments.During the DP, if the last open segment can contain the new element, everything remains the same. Otherwise, we choose either to start a new segment (so that j does not change), or to change the current element (so that j becomes j+1).Note that there would be cases that second_best becomes better than best, in which they should be swapped.Code: 110279809
•  » » 14 months ago, # ^ |   0 Can we solve it using binary search + greedy?
•  » » » 14 months ago, # ^ |   0 I do not know whether it is possible, but I think that would be hard.
•  » » » » 14 months ago, # ^ | ← Rev. 3 →   0 Like we will fix the max number of elements per segment and then try to put the elements greedily. Whenever we find a repeated value we will replace it with a very big prime number.(of course considering the restriction of k)If it is possible then we set lo=mid+1 else hi=mid-1 Then the answer would be ceil(n/hi)
•  » » 14 months ago, # ^ |   0 You may solve our problem here using your observation.
 » 14 months ago, # |   0 Can someone explain how left[i][j] is calculated in E2. I am not too comfortable with two pointers, perhaps there is any other way to calculate it? I understood the second part of the solution, the first part bothers me.
•  » » 14 months ago, # ^ |   0 Umm.............anyone? It will be really helpful if anyone could.
 » 14 months ago, # |   0 is there any way to solve E1 problem using binary search??
•  » » 14 months ago, # ^ |   0 No need of binary search just fill the elements greedily, you will get the answer
•  » » » 14 months ago, # ^ |   0 BhanuPratap Stuck in E1. In sample test case 1, should't answer be 2 ? Isn't it valid -> (18, 6, 4), (2, 1) ?
•  » » » » 14 months ago, # ^ |   0 You cannot reorder the elements. The first test case is 18 6 2 4 1
 » 14 months ago, # |   0 For C1 wasn't there suppose to be a case when n % 3 == 0.The answear would be then n / 3 , n / 3, n / 3?
•  » » 14 months ago, # ^ | ← Rev. 2 →   0 You could analyze that, but that is not needed. The solution explained in the editorial says if n is odd you do A and if n is even you do B, no need to add more cases. This is a typical bad practice from beginners.If n is a multiple of 13 you could answer 6*n/13, 6*n/13, n/13. But it would be silly to add such a case
 » 14 months ago, # |   0 Now let's relax dp values. When we consider an edge {i,j},tagi≠tagj we try to solve problem i after solving dpj problems ending with j, and problem i after solving dpi problems ending with i. It means that dpi=max(dpi,dpj+p), dpj=max(dpj,dpi+p) at the same time, where p=|si−sj|.Can anyone explain this part of the editorial of problem D?
•  » » 14 months ago, # ^ |   0 An important part that the editorial misses is that we can only jump back once that is from j to i ;j>i, try to prove it yourself.
•  » » 14 months ago, # ^ |   0 Even i am not able to get this part....
 » 14 months ago, # |   0 I am getting wrong answer for E1 in test 3. I cannot find out where I am making a mistake. Can someone please help? my submission is : 110286186
•  » » 14 months ago, # ^ |   0 Can you tell me why (18, 6, 4), (2, 1) are not valid sequence for first test case of E1??
 » 14 months ago, # |   0 Thank you for C1 and E1 which give me some ideas about hard version
•  » » 14 months ago, # ^ |   0 RadestionAdtinium Can you tell me why (18, 6, 4), (2, 1) are not valid sequence for first test case of E1??
•  » » » 14 months ago, # ^ |   0 You can't change the array
»
14 months ago, # |
-14

# include<bits/stdc++.h>

using namespace std; int main(){ int t; cin>>t; while(t--){ int n; cin>>n; int tag[n]; int i,j; for(i=0;i<n;i++){ cin>>tag[i]; } int s[n]; for(i=0;i<n;i++){ cin>>s[i]; }

long long int ans=0;
for(i=1;i<n;i++){
for(j=0;j<i;j++){
if(tag[i]!=tag[j]){
ans=ans+abs(s[i]-s[j]);
}
}
}
cout<<ans<<endl;

} return 0; }

//can some one explain why logic is not true. for the same test case i am attaching a picture

•  » » 14 months ago, # ^ |   -8 if j1 is greater than j2 then 2^j1-i is always greater than 2^j2-k for all i and k less than j2 .if someone still didn't get the logic you can message me but please help me
 » 14 months ago, # |   -7 Guys I'm really sorry I know posting your code and asking why it's giving me an error is a sure-fire way to get downvoted but I really have no choice, I have commented neatly and explained the code for problem D, but I keep getting a run-time error on TC-2, can anyone please help me outHere is the code : Python 3Thank you in advance :)
•  » » 14 months ago, # ^ |   0 it's because in your code first of all you didn't memoized all the states and 2ndly recursion tree for your code does not form a DAG(because yours is cyclic) which is necessary condition for any dp/recursion problem. In simpler words you end up calling a state which initially made you call these states.
 » 14 months ago, # |   0 For a case where n = 3z ,z being an odd natural number, isn't the answear (z, z ,z), not (1, n / 2, n/ 2) as it is specified in the editorial ?
 » 14 months ago, # |   -8 Guys is there an algorithm involved in B problem. Or is it an only math based problem? What should be said ?
•  » » 14 months ago, # ^ |   0 Both i guess
•  » » » 14 months ago, # ^ |   -8 which algorithm ?
 » 14 months ago, # | ← Rev. 2 →   -8 I can't understand what is the wrong here can you help me 110316296
 » 14 months ago, # |   0 Hello Everyone, Can some one tell me, where I am getting wrong in problem E2? I am getting WA on 8th testcase. My solution. Thank you in advance !!
 » 14 months ago, # |   0 Can someone provide better explanation for D and E2?
 » 14 months ago, # |   -15 C1, C2 sucks :((
 » 14 months ago, # | ← Rev. 2 →   0 Hi, guys! May u help pls find out why do I have TL in E1 ?I run the sieve to get least devisors of all numbers from 1 to 1e7. Which is 1e7 * log(log(1e7)) Then for each a[i] I get mask whish is 2*1e5 * log(1e7) in worst case. So totally it is less then 1e9. But what's wrong?My submission: https://codeforces.com/contest/1497/submission/110540450
 » 14 months ago, # |   0 Can anyone explain the reason why am i getting TLE on test case 102 on this submission but the exact same code gets AC here when i change 'long long' to 'int' with no other changes for problem E1.
 » 14 months ago, # |   0 Can anyone explain this line in editorial of div2 B "In this array the amount of x and the amount of m−x should differ not more than by 1, that's why we need to make max(0,|cntx−cntm−x|−1) arrays, containing a single number (x or m−x) that is more common."
 » 14 months ago, # |   0 In div2B why the counterpart of x is m-x. If m=5 and x=13 then counterpart of x will be 5-13 = -8 but we know that we need only 2 to add to 13 to make their sum divisible by 5. Instead it should be m-(x%m). Can anyone tell whether I'm correct or going wrong ??
•  » » 10 months ago, # ^ |   0 its actually m-(x%m),that's why it is written that take each number modulo m.
 » 14 months ago, # |   0 Omg. I was amazed when I read the tutorial for Problem C2. I accepted C1. After that, I tried to think and find solution for C2 for 1 hour before I gave up. That's amazing. That's unbelievable. That's incredible.
 » 13 months ago, # | ← Rev. 2 →   0 Can somebody tell why my submission for problem E1 fails on test case 2:https://codeforces.com/contest/1497/submission/113704368Thank you!Edit: I found my mistake. The function returning the reduced integer was wrong. A silly mistake!
 » 12 months ago, # |   0 For B, why is it $max(0, |cnt_x-cnt_{m-x}|-1)$? If the counts differ by 1, won't we have one additional array? Shouldn't this be $max(1, |cnt_x-cnt_{m-x}|)$?
•  » » 10 months ago, # ^ |   0 If the counts differ by 1 we can always include that extra element in our array. It will not violate the rules since we need adjacent sum to be divisible
 » 12 months ago, # |   0 in problem {1497B — M-arrays} I think unordered_map works well instead of map but it gives wrong answer , don't understand why?
 » 10 months ago, # |   0
»
9 months ago, # |
0

1497C1 — k-LCM (easy version) shishyando I tried problem and found a solution which is best and easy to understand

# include<bits/stdc++.h>

using namespace std;

# define rep(i,a,b) for(int i = a; i < b; i++)

void solve() { int t; cin>>t;

while(t--){
int n,k;
cin>>n>>k;

int res[k];

if(n == 3){
res[0] = 1;
res[1] = 1;
res[2] = 1;
}
else if(n %2 == 0){
if((n/2) % 2 == 0){
res[0] = n/2;
res[1] = n/4;
res[2] = n/4;
}
else{
res[0] = (n/2)-1;
res[1] = (n/2)-1;
res[2] = 2;
}
}
else if(n%2 != 0){
res[0] = n/2;
res[1] = n/2;
res[2] = 1;
}
for(int i=0;i<k;i++){
cout<<res[i]<<" ";
}
cout<<endl;
}

}

int main() { solve(); return 0; }

•  » » 9 months ago, # ^ |   0 Best solution indeed!
 » 7 months ago, # |   0 Why my code is giving TLE, I think it is completely right Time Comlecity is O(n*sqrt(a[i])) -> <=10power 9 Please Help me #include using namespace std; #define int long long void solve(){ // int MOD=998244353; int n,k;cin>>n>>k; map>vertices; for(int i=0;i>a>>b; vertices[a].insert(b); vertices[b].insert(a); } if((n%2==0 and k>=n/2) or (n%2!=0 and k>n/2)){ cout<<0<>t; while (t--){ solve(); } return 0; } 
 » 7 months ago, # |   0 Can someone please help me understand why this solution for E2 works on all tests except 47?