This "sandbox" is a place where Code Golf users can get feedback on prospective challenges they wish to post to main. This is useful because writing a clear and fully specified challenge on your first try can be difficult, and there is a much better chance of your challenge being well received if you post it in the sandbox first.

Sandbox FAQ


To post to the sandbox, scroll to the bottom of this page and click "Answer This Question". Click "OK" when it asks if you really want to add another answer.

Write your challenge just as you would when actually posting it, though you can optionally add a title at the top. You may also add some notes about specific things you would like to clarify before posting it. Other users will help you improve your challenge by rating and discussing it.

When you think your challenge is ready for the public, go ahead and post it, and replace the post here with a link to the challenge and delete the sandbox post.


The purpose of the sandbox is to give and receive feedback on posts. If you want to, feel free to give feedback to any posts you see here. Important things to comment about can include:

  • Parts of the challenge you found unclear
  • Comments addressing specific points mentioned in the proposal
  • Problems that could make the challenge uninteresting or unfit for the site

You don't need any qualifications to review sandbox posts. The target audience of most of these challenges is code golfers like you, so anything you find unclear will probably be unclear to others.

If you think one of your posts requires more feedback, but it's been ignored, you can ask for feedback in The Nineteenth Byte. It's not only allowed, but highly recommended! Be patient and try not to nag people though, you might have to ask multiple times.

It is recommended to leave your posts in the sandbox for at least several days, and until it receives upvotes and any feedback has been addressed.


Search the sandbox / Browse your pending proposals

The sandbox works best if you sort posts by active.

To add an inline tag to a proposal, use shortcut link syntax with a prefix: [tag:king-of-the-hill]. To search for posts with a certain tag, include the name in quotes: "king-of-the-hill".

  • \$\begingroup\$ What if I posted on the sandbox a long time ago and get no response? \$\endgroup\$
    – None1
    Commented May 15 at 14:05
  • \$\begingroup\$ @None1 If you don't get feedback for a while you can ask in the nineteenth byte \$\endgroup\$
    – mousetail
    Commented May 29 at 13:27

4703 Answers 4703

118 119
121 122

Product and Sum free set

Given some non-negative integer \$n\$ find a sum- and product-free subset \$S \subset \mathbb N = \{1,2,3,\ldots \}\$ of size \$|S|=n\$

EDIT: with the least possible maximum \$\max_{s\in S} s\$.


  • A set \$S\$ is called sum-free if for all \$a,b\in S\$ their sum \$a+b \not \in S\$.
  • A set \$S\$ is called product-free if for all \$a,b\in S\$ their product \$a\cdot b \not \in S\$.



Note that the sets for each \$n\$ might not necessarily be unique.

  n  S
  1  {2}
  2  {2,3}
  3  {3,4,5}
  5  {4,5,6,7}

EDIT: Ok, no this doesn't work either.

  • \$\begingroup\$ I'm a bit concerned that this would just be output the first n odd primes, but maybe non-golfing languages can do something more interesting than that. \$\endgroup\$
    – Emigna
    Commented Feb 26, 2019 at 10:09
  • \$\begingroup\$ @Emigna Thanks for the feedback, you're right this is not something I considered. Maybe it helps if we allow any number of additions/multiplications? But in this case I think you could just use [k+1,k+2,k+3,...,n] for some large enough k (depending on n). Maybe I should add some condition that requires the numbers to be very small to make it more interesting. It definitely needs some more work. \$\endgroup\$
    – flawr
    Commented Feb 26, 2019 at 10:42
  • \$\begingroup\$ You can make this work by having a secondary scoring algorithm. You could, for example, score the programs by {bytes}*{total score for n=1 to 100}, where the score for an input is the highest number in the output set. \$\endgroup\$ Commented Feb 27, 2019 at 18:28
  • 2
    \$\begingroup\$ Also note that the set {1} is not product free - 1*1=1, which is in {1}. \$\endgroup\$ Commented Feb 27, 2019 at 18:29

Use input letters to create as many words as possible

I have the scrabble letters as magnets on my fridge and I was thinking: how many actual words could I make using only these letters. Obviously it would be hard to work out myself, so I challenge you to create a program which will take a string of characters as input, and your goal is to output as many words as possible using those characters.


  • The characters will be one string
  • Only the 26 letters of the alphabet will appear.
  • Letters may appear more than once
  • Letters can appear in any order. 5 e's might appear next to each other, and there might be other e's in the string.
  • There will always be at least a single letter, but it might not always be possible to create a valid word

Example, using the letters from Scrabble:



A list of valid English words, which you can separate with a space, newline or comma. A trailing newline at the end is fine.

Default loopholes are forbidden.

Meta discussion (not apart of the question)

  • Is a decent challenge for this? I'm more interested in more words being output than I am short code, but how best can I frame this challenge that the best result would be most words created and fewest characters left?
  • Should I restrict the input list to just the letters from Scrabble? I mean how can I evaluate whether a program prints as many as possible?
  • Should I provide a word list? I know there are challenges on this site where a word list is provided 1 and 2 and I'm sure there are a few others which I could not find.
  • This is my third question, so any other feedback
  • 1
    \$\begingroup\$ I would highly recommend including the wordlist as an input and including a few trivial test cases so that people can easily test their answers. \$\endgroup\$
    – Jo King Mod
    Commented Apr 26, 2019 at 7:45
  • 1
    \$\begingroup\$ Not including the word list as input would benefit languages which have dictionary built-ins - everyone else would have to have a list of all English words in the code, which might be a little long, even given really good compression... \$\endgroup\$
    – Matthew
    Commented Apr 26, 2019 at 8:28
  • \$\begingroup\$ Allow any reasonable output format rather than just text. Pass in word lists and letter sets as arguments. Making both constant turns this into a kolmogorov-complexity challenge. \$\endgroup\$
    – Beefster
    Commented May 1, 2019 at 17:53

Who's serving in table tennis?

Table tennis, also known as ping pong, is a well-known game in which players hit a ball back and forth across a table. One player serves the ball twice, and each time the ball is served, either player can score a point. Once the first player serves the ball twice, the second player gets to serve twice. Every two serves, the player which serves the ball switches. The game ends when one player has a score of 11 or more points and is two or more points higher than the other player.

Your task is to take the current score as input, and output a value depending on if player 1 should serve, player 2 should serve, or the game is over.

Input: The score of the first player to serve followed by the score of the other player. This can be as separate values or in a list, or another convenient data type.

Output: A distinct value for each possible case of output. For the test cases, I will use 1 to represent the player who serves first, 2 to represent the other player, and 0 to represent that the game is over.

Test cases:

player1score, player2score -> output

2, 0    ->  2
0, 0    ->  1
11, 0   ->  0
4, 4    ->  1
3, 2    ->  1
6, 9    ->  2
81, 82  ->  2
0, 29   ->  0
16, 13  ->  0

Help the library sort books

A library needs to sort its books, located in separate sections by language. Of course, each language has its own rule for collating strings, so your method should handle that ability.

Most languages will want to treat titles case-insensitively, but unfortunately, the library has a section of Klingon books. And Klingon's romanisation, in case you don't already know, is case-sensitive. As a result, you need to give a way to have the sorter treat two given characters as the same.

Some of the titles also contain numbers, and you need to deal with those as well. Fortunately, the friendly people at the library have written some functions for you to deal with that, so all you have to do is take in a handle to a function as one of your inputs.

Your challenge: Sort a list of book titles, given a collating sequence and a function handle that converts a number to its string representation in the respective language.

  • The collating sequence is a list of lists of strings. Given two strings A and B that appear respectively in the entries at indices i and j, respectively:
    • if i > j, then A should sort after B
    • if i = j, then A should sort the same as B
  • In case some elements of the sequence have multiple characters, use the max-munch rule to tokenise the string: e.g. if both l and ll appear in the collating sequence, then lll should be tokenised to ll l.
  • Any string of digits should be parsed as a number and passed to the callback. Each such string should be replaced by the result.
  • Any characters not appearing in the collating sequence should be ignored. If two strings differ only by such characters, then sort them however you like.

Test cases


english_letters = [["A", "a"], ["B", "b"], ["C", "c"], ["D", "d"], ["E", "e"], ["F", "f"], ["G", "g"], ["H", "h"], ["I", "i"], ["J", "j"], ["K", "k"], ["L", "l"], ["M", "m"], ["N", "n"], ["O", "o"], ["P", "p"], ["Q", "q"], ["R", "r"], ["S", "s"], ["T", "t"], ["U", "u"], ["V", "v"], ["W", "w"], ["X", "x"], ["Y", "y"], ["Z", "z"]]
arka_letters = [["t"], ["k"], ["x"], ["s"], ["n"], ["v"], ["f"], ["m"], ["d"], ["g"], ["p"], ["b"], ["h"], ["y"], ["c"], ["r"], ["z"], ["j"], ["w"], ["l"]]
function english_num_callback(num) {
  // convert a number to its representation in English
function arka_num_callback(num) { // handles nonnegative integers up to 9999
  if (num == 0) return "yuu";
  let kot = 0 | (num / 1000);
  let gal = 0 | (num / 100 % 10);
  let on = 0 | (num / 10 % 10);
  let ko = num % 10;
  let digits = ["yuu", "ko", "ta", "vi", "val", "lin", "kis", "nol", "ten", "los"];
  return (kot == 0 ? "" : digits[kot] + "kot") +
    (gal == 0 ? "" : digits[gal] + "gal") +
    (on == 0 ? "" : digits[on] + "on") +
    (ko == 0 ? "" : digits[ko]);
["Stack Exchange", "BB94 Channel!", "BB guns: all you wanted to know about them",
  "carrots", "A Grammar of Jbl", "69"], english_letters, english_num_callback =>
  ["A Grammar of Jbl", "BB guns: all you wanted to know about them", "BB94 Channel!",
    "carrots", "69", "Stack Exchange"]
["melidia", "diaklel", "44 miiko", "I Mess With the Librarians", "an isk ris tu lei",
  "lei e xion", "xagrisren et xep!", "on melkadren"], arka_letters, arka_num_callback =>
  ["xagrisren et xep!", "44 miiko", "melidia", "diaklel", "lei e xion",
    "an isk ris tu lei", "on melkadren", "I Mess With the Librarians"]

Will it be red sauce, brown sauce, or no sauce at all?

Unfortunately Danny Baker's Sausage Sandwich game has been taken off the air again, and I'm suffering from withdrawal already. Please alleviate my symptoms by writing a program or function to play the Sausage Sandwich game.

You need to guess how the target prefers to eat a sausage sandwich, given one of the three sauce options. Given an input positive integer n, then the following applies:

  • There is a uniformly 1 in n chance that the target chooses no sauce at all.
  • Otherwise, there are equal chances of red sauce or brown sauce.
  • However, if n is even, then there are n-1 remaining chances, which don't divide by 2. If you draw this n-1th chance you should consistently (for a given n) output the same answer. Please indicate how this is chosen in your answer.
  • Output must consist of the phrases "red sauce", "brown sauce", or "no sauce at all", in any of lower, sentence, title or upper case.

This is , so the shortest program or function that breaks no standard loopholes wins.


Print the Golfican sequence

The Golfican sequence is a sequence of integers, \$u\$, defined as follows:

  • If \$n\$ is a power of \$2\$, then \$u_n=\log_2n+1\$.
  • If \$n\$ in not a power of \$2\$, then \$u_n=u_p\$, where \$p\$ is the number of numbers from \$1\$ to \$n\$ inclusive which are not powers of \$2\$.

So, the first numbers in this sequence are \$1,2,1,3,2,1,3,4\$ and so on. Your job is to write a program which takes a number, \$x\$, as input, and outputs \$u_x\$. This is , so the shortest program in bytes wins.

I couldn't find a corresponding OEIS sequence for this. Thoughts?

  • \$\begingroup\$ So \$p = n - 1 - \lfloor \log_2 n\rfloor = n - u_{b(n)} = n - \ell(n)\$ where \$b(n)\$ is the largest power of 2 less than \$n\$ and \$\ell(n)\$ is the length of \$n\$ in base 2? \$\endgroup\$ Commented May 17, 2019 at 8:27
  • \$\begingroup\$ I suppose it would be. \$\endgroup\$
    – RamenChef
    Commented May 17, 2019 at 16:36


  • 2
    \$\begingroup\$ (1) IMO the multiple-box requirement is essential to make it an interesting question. Otherwise it's just the hundredth shortest path question, and probably should be closed as a dupe. (2) "If your solution can only retrieve the first box, 10x penalty" is a really bad idea. See Things to avoid when writing challenges: chameleon questions, bonuses in code golf. \$\endgroup\$ Commented May 17, 2019 at 10:34
  • \$\begingroup\$ @PeterTaylor: Thanks, I've removed the bonuses and outlawed brute force. \$\endgroup\$
    – Phil H
    Commented May 17, 2019 at 10:55
  • 2
    \$\begingroup\$ Banning brute force is also discouraged, as a non-observable requirement. Some people use a time limit to rule out the slowest kinds of brute force instead. For example, must process any of the test cases in under 1 minute (the time limit being chosen generous enough to make no difference to most approaches, so it doesn't even matter which computer it's measured on) \$\endgroup\$ Commented May 18, 2019 at 16:28
  • 1
    \$\begingroup\$ @trichoplax: I've modified to replace the brute force note with a time limit \$\endgroup\$
    – Phil H
    Commented May 20, 2019 at 13:42

Optimize Grocery Bagging

You've gone to the store and are at the checkout. Now since you want to be environmentally conscious (or are somewhere that charges for bags and want to save money), you've brought a few reusable bags and would like to minimize the number of plastic bags you use (There are no paper bags here).

Each reusable bag can carry 5 kg (5,000 grams) of groceries and each plastic bag can carry 2.5 kg (2,500 grams) of groceries. You will be given a list of weights of groceries (in whole grams) and a number of reusable bags and your task is to output a configuration of bagging that requires the fewest number of plastic bags.


  • Input: 1 reusable bag, groceries: [150, 500, 800, 1250, 2000, 300, 550, 400, 750, 900, 1000, 600]
    • Output: Reusable[2000, 1250, 1000, 750], Plastic[600, 900, 550, 150, 300], Plastic[500, 800, 400]

More TBD

Rules and Assumptions

  • Standard rules and loopholes apply
  • Any convenient input or output method is acceptable as long as which bags are reusable is unambiguous in the output (without calculating anything).
  • You may assume that there are no groceries larger than 2.5 kg. (No jugs of milk)
  • All groceries listed require a bag. Unbaggable groceries will not be listed. (No hypothetical 24-packs of toilet paper)
  • Any grocery item can be bagged with any other. (No need to worry about bread and cans of beans being in the same bag)
  • Groceries cannot be split into multiple bags.
  • There is no double-bagging.
  • Complexity doesn't matter. There may be inputs that would require computation past the heat death of the universe to complete.

Happy Golfing!


Sandbox: This could work as a problem or due to the naive solution being in \$O(n!)\$. Thoughts?

  • \$\begingroup\$ Related: codegolf.stackexchange.com/q/109797/25180 \$\endgroup\$
    – jimmy23013
    Commented May 28, 2019 at 22:39
  • \$\begingroup\$ This is NP-hard. \$\endgroup\$
    – jimmy23013
    Commented May 28, 2019 at 22:40
  • \$\begingroup\$ @jimmy23013 yes, indeed. If this were shifted to fastest-code it would necessarily drop the "guaranteed optimality" requirement. \$\endgroup\$
    – Beefster
    Commented May 28, 2019 at 23:13
  • \$\begingroup\$ @jimmy23013 I guess this is only different from that linked challenge in that there are two bag capacities instead of just one. \$\endgroup\$
    – Beefster
    Commented May 28, 2019 at 23:17
  • \$\begingroup\$ Do you mean requiring the minimum number of bags? If not required optimal, how do you decide whether an answer is valid? \$\endgroup\$
    – jimmy23013
    Commented May 28, 2019 at 23:18
  • \$\begingroup\$ As I remembered (USACO fence8), fast solutions to this problem may use strategies depending on the input data (e.g. try putting heavier things in bags first). You may have to specify exactly how it is scored if you change it to fastest-code. \$\endgroup\$
    – jimmy23013
    Commented May 28, 2019 at 23:22

Beziér me a curve!

Task: Create a program that will, given a number N, create a Beziér curve with N control points and output its equation.

The definition of a Beziér curve is $$ B(t) = \sum_{i=0}^n{{n\choose i}(1-t)^{n-i}t^iP_i} $$

Practically, this means that for a quadratic Beziér cuve: $$ B(2) = \sum_{i=0}^2{{2\choose i}(1-t)^{2-i}t^iP_i}\\ B(2) = {2\choose 0}(1-t)^2P_0 + {2\choose 1}(1-t)tP_1 + {2\choose 2}t^2P_2\\ B(2) = (t^2-2t+1)P_0 + 2(t^2-2t+1)P_1 + t^2P_2\\ B(2) = P_0t^2-2P_0t+P_0 + 2P_1t^2-4P_1t+2P_1 + P_2t^2 $$

where \$P_0\$ is the first control point, \$P_1\$ is the second control point and \$P_2\$ is the third control point.

Note that De Casteljau's algorithm may be useful.


  • The input will be a number (string or number).
  • The output will be a list of lists. Take an element \$l_n\$ of the top-level list. The elements of \$l_n\$ will be all the terms that, in the expanded form, are multiplied by \$P_n\$. -1t and t^1, as well as t^0 are all fine.
  • Standard loopholes are not allowed.

Test cases


  ["t^2", "-2t", "1"],
  ["2t^2", "-4t", "2"],


  ["-t^5", "5t^4", "-10t^3", "10t^2", "-5t", "1"],
  ["t^5", "-4t^4", "6t^3", "-4t^2", "t"],
  ["t^5", "2t^4", "t^3"],
  ["-t^5", "t^4"],

The program with the shortest length wins. Be creative!

  • \$\begingroup\$ It would be more accurate (and IMO less confusing) to say that the task is to output the Bernstein basis polynomials of degree n, $$b_{a,n}(t) = \sum_{k=a}^n (-1)^{a+k} \binom{n}{k} \binom{k}{a} t^k$$ In line with Avoid cumbersome I/O formats I would also suggest outputting the polynomials as lists of coefficients, i.e. as a double array / matrix with $$A_{ij} = (-1)^{i+j} \binom{n}{i,j}$$. However, all of this is moot because now that I've rewritten it like that I see it's a dupe of codegolf.stackexchange.com/q/69424/194 \$\endgroup\$ Commented May 29, 2019 at 8:22

Reading the text of the millitext font

There is a font here that fits each character into a single, 1x5 pixel block. It does this (assuming you're using an LCD screen) by using the RGB channels of each pixel to expand the one pixel into three subcolumns, one for each channel. Your task is to take in a string of text that describes this font and return the correspond text encoded.

Millitext Alphabet


I've abbreviated each color to a one character symbol (R=red, G=green, B=blue, C=cyan, Y=yellow, M=magenta, W=white).

Input Format

The input format for this is pretty open. You can have the input be an array containing each column, an array containing each row, a char[][], or anything like that. You can also choose to use the full words "red", "green", "blue", with uppercase/lowercase of your choice (but it must be consistent for each word! You cannot use "RED" and also do "green" or "Blue").

If your language happens to support it, you can also have input be colors (however that might work, I don't know off-hand).

You can assume that the input will ONLY contain encoded characters in the alphabet above (in particular, there will be no spaces or punctuation in your output).

Output Format

You can either output a string or some sort of character array. You can chose whether the letters are uppercase or lowercase, but they must all have the same case.




This is code golf, so shortest answer wins!

Test set


I Love U So Much!

Just print i love u, or I Love U.

But, because i love u so much, there is two rule.

  1. At least there must be one I, and one U. Capital or not is regardless.
  2. Every I and U must be stick together. No sole i and sole u.

This is , so shortest code with byte wins.

  • \$\begingroup\$ What is preventing me from just putting a comment or no-op at the end with #IU or something similar? \$\endgroup\$ Commented Jun 11, 2019 at 12:36
  • \$\begingroup\$ @AdmBorkBork Ah, I forgot that. Every I and U must be stick together. \$\endgroup\$
    – LegenDUST
    Commented Jun 11, 2019 at 12:37
  • \$\begingroup\$ That's better, but it still allows, for example, BrainFuck to just type out the necessary > and < and put IU at the end as a comment/no-op. If that is your intention, that's fine, just wanting to point it out. \$\endgroup\$ Commented Jun 11, 2019 at 12:42
  • 4
    \$\begingroup\$ IMO this is too boring to be worth posting. The string isn't compressible, so it's just a case of taking the string "ui love ", rotating by one character, and printing (finding a workaround in the languages where one would normally print with print). \$\endgroup\$ Commented Jun 12, 2019 at 10:54
  • \$\begingroup\$ @PeterTaylor Maybe I could make string more complicated? \$\endgroup\$
    – LegenDUST
    Commented Jun 12, 2019 at 11:11

Trace My Triangles

Triangular is a 2D programming language that reads its code in the shape of the smallest triangle that the code will fit into, ignoring all whitespace. Its IP starts at the very top point of the triangle, and traverses the code starting in a Southeast direction. The IP's direction can be changed with the following directional switches (Note that "." is a no-op):

  ` /
 < . >  
  , \


You will be given a single input string p which represents a valid Triangular program. Your task is to write a program or function, which will print or return the actual execution order of p. Instructions in p which are never reached by the IP can be ignored.

You may assume the following:

  • p will contain only characters which are necessary for a Triangular program to be valid; no need to account for whitespace.
  • p will only switch directions via the directional switches listed above. The o,e,c and z instructions will not be used.


Input:                    Output:
F\(?)1/%-<                F(1<-%/?\)

,%./                      ,%/

\@~/;<                    \~<;/@

123456                    136

This is code-golf, so the shortest byte count wins.


  • Is this a good idea for a question/not a dupe?
  • Is the requirement clear enough?
  • What other test cases should I be adding, if any?
  • Should jump-related instructions - (, ) and ] - also be among the things that can be disregarded?
  • Should any other sorts of instructions be disregarded?
  • Would any other assumptions make this more manageable/fun?

I'm not able to find a challenge which will return as an integer how many days until Christmas based on an input like DDDD-MM-YY. So...

How many days until Christmas?

Work out the number of full days until Christmas based on today's date based on the Western Christianity date source.

Write a program, method, function, routine or sub-routine that accepts an input of the current date according the the Gregorian calendar source. The date may be in any format as long as you use the full year (so 2019-09-11 rather than 19-09-11). Other than requiring the full year, the date format is up to you (DD/MM/YYYY or MM/DD/YYYY or even formally written dates like Wednesday 12th June 2019). Please state what date format you are using in your answer.

An integer is displayed according to your standard output.


input -> output:

12/06/2019 -> 196 days until Christmas*
12/25/2019 -> 0 days until Christmas**

* DD/MM/YYYY format
** MM/DD/YYYY format
  • 1
    \$\begingroup\$ Subset of this? \$\endgroup\$
    – Adám
    Commented Jun 12, 2019 at 13:18
  • 1
    \$\begingroup\$ @Adám Or really similar to Christmas Eve Eve, just accumulate the days rather than printing "Eve." \$\endgroup\$ Commented Jun 12, 2019 at 13:22
  • \$\begingroup\$ Yes, thanks. Those challenges are specifically for technologies that have DateTime functions built in, my suggested challenge would be more suitable for technologies that don't know what date it is or what timezone they are in, like the Sinclair and Commodore 8-bit machines, for instance. Because you would have to specify the date and the rest would have to be worked out with some programmings. \$\endgroup\$ Commented Jun 12, 2019 at 14:15

Is this set doubly orderable?

For input you are given a set of four or five sets of strings. Each string is at least two characters long; those first two characters are always digits but the remaining characters can also be letters.

The set of sets is doubly orderable if an ordered list can be constructed using one element from each member of the set such that:

  • The first characters of each element of the list are in nondescending order
  • The second characters of each element of the list are separately in nondescending order
  • If any elements of the list have more then two characters, then those elements are all wholly equal with each other and appear at the beginning of the list (while still satisfying the other conditions).

Although there is an ordering on the list itself, the original sets can be sampled in any order necessary to achieve a valid list.

Example valid lists:

13  24  25  36
13e 13e 13  13
33s 37  37  37

Example invalid lists:

13  33  15  35  (first character not in order)
13  15  33  35  (second character not in order)
13e 24s 25  36  (different overlong strings)
33  37s 37s 37s (overlong strings are not first)

For output you should have:

  • A consistent value indicating that the set is not doubly orderable
  • A consistent value indicating that the set is can be doubly ordered using only 2-character strings
  • A list of all overlong strings of which one is needed to make a valid list.

Full example:

1. 13 23 33 14 24 34 15 25 35 16 26 36 17 27 37
2. 33s 15e 16 17e
3. 34 25
4. 17e 37

In this example the 34 and 25 in set 3 force us to use either 33s or 15e from set 2 and therefore the 37 from set 4, with a number of options from set 1. Since all the valid lists begin with either 33s or 15e, these are the outputs.

  • \$\begingroup\$ I find this spec quite confusing. (1) ISTM that the property "doubly ordered" should really be "doubly orderable". (2) The order of the elements in the list can be inferred to be arbitrary from the fact that the spec talks about a set of sets, but it would be much better to make this explicit. E.g. change the second paragraph to start "The set of sets is doubly orderable if a list can be constructed by taking one element from each member of the set and ordering them so that:". (3) On the third bullet, do the elements have to be equal or just the suffixes? \$\endgroup\$ Commented Jun 12, 2019 at 13:40
  • \$\begingroup\$ @PeterTaylor (3) Yes, the elements have to be wholly equal in this case. \$\endgroup\$
    – Neil
    Commented Jun 12, 2019 at 14:31

Proving that a Russian cryptographic standard is too structured

  • \$\begingroup\$ Nice challenge. I have some suggestions for usability for solvers. Please say what language your example code is in (C?), and an online runnable version -- the community standard is TIO which is maintained by a site mod (example in Python). A pseudocode version or explanation would also be appreciated. It would be useful to include a listing of the permutation as 256 numbers from 0 to 255 written in base 10. The references would be cleaner as inline hyperlinks. \$\endgroup\$
    – xnor
    Commented Jun 6, 2019 at 21:39
  • \$\begingroup\$ You should allow submissions to be programs or functions, which is default. Also, please change the scoring to bytes. I sympathize with wanting a fair comparison for 7-bit vs 8-bit languages, but past efforts have proved unpopular, and people now generally think of competitions as being within each language. You can leave the conversion to bits for your personal leaderboard. \$\endgroup\$
    – xnor
    Commented Jun 6, 2019 at 21:46
  • \$\begingroup\$ Also, please make the challenge as self-contained as possible, which means putting into the challenge body the parts of the linked papers that are important for solvers. Here, I think this is mainly how the function p works. In particular, what does it mean that the table k is an affine function, and how can it be implemented as such? Does s have any structure? \$\endgroup\$
    – xnor
    Commented Jun 6, 2019 at 21:52
  • \$\begingroup\$ Thanks a lot for your comments! I have updated the challenge and did my best to take them into account. Is it clearer now? \$\endgroup\$ Commented Jun 6, 2019 at 23:50
  • \$\begingroup\$ Yes, thanks for the edits. Now that you've explained that the core of the code is discrete log, I have some questions on that which other solver might have too. I take it the char declaration is to limit the values to 8 bits and discard bits? If I think of the field of size 256 as GF_2[X] modulo a certain irreducible polynomial, does the update a=(a<<1)^(a>>7)*29 correspond to multiplying by a polynomial? Could the initialization be l=0, a=1 instead? Note that LaTeX is enabled here, in case it's easier to write formulas. It also could help to say that ^ is xor, not power. \$\endgroup\$
    – xnor
    Commented Jun 7, 2019 at 0:21
  • \$\begingroup\$ The answer to your questions are yes---mostly. The unsigned char are indeed used to discard some bits. We could alternatively replace a=(a<<1)^(a>>7)*29 with a=(a<<1)^(a>>7)*285 (where 256^29==285, in which case we wouldn't need the unsigned chars. This code snippet indeed corresponds to a multiplication by a polynomial. However, we cannot start at l=0,a=1. For some reason, the designers of $\pi$ used a variant of the logarithm such that $\log_2(1)=255$ instead of $\log_2(1)=0$. Both are "correct" in the sense that $\alpha^{255} = \alpha^{0} = 1$ but this variant is less common. \$\endgroup\$ Commented Jun 7, 2019 at 0:28
  • \$\begingroup\$ I added these clarifications to the challenge. \$\endgroup\$ Commented Jun 7, 2019 at 0:45
  • \$\begingroup\$ The output of your C code seems to not match the table. \$\endgroup\$
    – xnor
    Commented Jun 7, 2019 at 1:02
  • \$\begingroup\$ The table was wrong rather than the C code. Thanks for catching it! I fixed it as well as the corresponding TIO link. To further simplify the verification, I added a link to the wikipedia page of Kuznyechik which contains the table of the function p. \$\endgroup\$ Commented Jun 7, 2019 at 1:18
  • \$\begingroup\$ Let us continue this discussion in chat. \$\endgroup\$
    – xnor
    Commented Jun 7, 2019 at 1:57

Continue a progression

The \$k\$th generalized mean \$M_k\$ of a set of numbers is defined as $$M_k(a_1,\dots,a_n)=\root^k\of{\frac{a_1^k+\dots+a_n^k}n}.$$ \$M_0\$, the geometric mean, is defined through a limit \$\lim_{k\to0} \root^k\of{\frac{a_1^k+\dots+a_n^k}{n}} = \root^n\of{a_1\cdot\dots\cdot a_n}\$. \$k\$ does not have to be integer.

An arithmetic progression is (according to Wikipedia) a sequence of numbers such that the difference between consecutive terms is constant. A geometric progression is a sequence of numbers such that the ratio between consecutive terms is constant.

There is a simple way to generalize these (that also explains why on earth is geometric progression named geometric): for a progression of order \$k\$ for any 3 consecutive elements \$a\$,\$b\$,\$c\$ \$M_k(a,c)=b\$. This way, the arithmetic progression is of order 1 and the geometric progression is of order 0. The order does not have to be integer.

The task is, given 3 real positive numbers \$a,b,c\$, to find the next term in the progression formed by them.


  • If multiple valid outputs exist, you can output any of them.
  • It is guaranteed that a valid output exists.
  • An output is valid if \$|o_{correct}-o_{program}| \le10^{-5} \cdot \max(o_{correct}, 1)\$.

Sandbox stuff

  • I still have to prepare some test cases, especially for non-integer orders.
  • Has this been asked before?
  • Do multiple correct outputs ever exist? I feel like very often there are infinitely many.
  • Does Mathematica have a built-in?
  • Should I add a time limit? I feel like brute-forcing all floating point numbers and checking if they are valid outputs might be shorter than actually calculating.
  • Should I restrict the order to be integer, or even just 0 or 1 (leaving only arithmetic and geometric progressions)?
  • Is there a simple and beautiful formula for the next term that I was unable to find and that all answers will have to copy from each other? I could neither get Mathematica nor myself to solve any related equations.
  • Is the grammar correct enough?
  • \$\begingroup\$ I think The task is, given 3 real positive numbers ... is a bit problematic, as receiving them sounds like a daunting task. \$\endgroup\$ Commented Jun 13, 2019 at 15:21
  • \$\begingroup\$ @Jonathan Surely reading three numbers or taking them as function arguments or whatever is your language's alternative is the least thing you have to do here? \$\endgroup\$ Commented Jun 13, 2019 at 15:25
  • \$\begingroup\$ I think proving that submissions will always meet your validity criterion will be rather daunting. This will probably further encourage solutions that just try every single floating point number? I'd recommend saying that floating point issues won't be counted against the submission, but you still have a bit of a problem. If you do add a time limit I'd recommend saying that it only need be tested on the cases you provide (and maybe remind users that hardcoding is a loophole ;) ) \$\endgroup\$ Commented Jun 13, 2019 at 17:30
  • \$\begingroup\$ The time limit only on the test cases thing makes sense to me. What does "not counting floating point issues mean"? I think most solutions will have to use binary or ternary search, and that needs a, uh, stopping precision or a iteration count to be close enough. I think we can remove the need to verify it on all numbers by preparing a lot of test cases! \$\endgroup\$ Commented Jun 14, 2019 at 3:58
  • \$\begingroup\$ Not every solution will use that, and even if they do they may still run into a problem where they can't get a solution well enough based on their native floating point type and have to do a bunch of extra work. I think it works well as a catch-all that prevents golfers from having to struggle with weird edge cases. But generally I think you are right that just having enough test cases will be good enough (also I didn't get notified by your comment because someone else has commented already, you'd need to use @). \$\endgroup\$ Commented Jun 14, 2019 at 18:08
  • \$\begingroup\$ @FryAmTheEggman double-precision floating points have around 14 digits of precision, and 1e-8 is a perfectly common epsilon for such tasks (and I use 1e-5). And if I assume floating point numbers are arbitrary precision, that implicitly disallows binary/ternary search, as it will never produce the exact answer. And since the only two solutions I know are Mathematica NSolve and alternatives and searching for the order and then for the next number, this disallows all remotely interesting solutions known to me. \$\endgroup\$ Commented Jun 15, 2019 at 2:46
  • \$\begingroup\$ Assuming the languages that people use to answer your challenge will use a reasonable convention is not a particularly good one! I didn't meant arbitrary precision but arbitrarily larger precision based on what you have specified. It is probably fine the way you have it, I just think it is better to think about nonstandard languages when you can. If you put no time limit, some solutions may try to iterate over all floating point numbers of a particular precision, for example. Sorry if this has gotten out of hand, maybe I'm not doing a good job explaining what I mean? \$\endgroup\$ Commented Jun 15, 2019 at 3:16
  • \$\begingroup\$ So in short, given inputs \$a,b,c\$ find a \$k\$ such that \$a^k + c^k = 2b^k\$ and then find and output a \$d\$ such that \$d^k = 2c^k - b^k\$? There is a significant problem in that \$k=0\$ is always a solution to the first equation, so I think that the geometric mean would have to be tested first as a special case. \$\endgroup\$ Commented Jun 18, 2019 at 11:03
  • \$\begingroup\$ @PeterTaylor I think so. I feel like there actually is a simple and beautiful solution that everyone will have to copy from each other here. \$\endgroup\$ Commented Jun 18, 2019 at 11:08

Nearest match with limited values

When writing my solution for this challenge, I found that I needed to generate a target value using nothing but multiplication and increments/decrements, using a restricted set of values to use in the multiplication. Given that my choice of values was restricted, I had to brute-force what the optimum solution was. gObviously, this seems like a programming challenge!

The challenge is this: Given a target value t and set of allowed multiplier/multiplicand values, find the solution t=x×y±c (x and y are from the set of allowed values), that has the smallest value of c. If c is zero, indicate the solution as an exact match.

As this is a challenge, smallest code size wins. The standard rules apply.

Example program (ungolfed): Try it online!

I don't think that I saw this particular challenge before, but I may have missed some. This is my first challenge, so please let me know how I can make it better!

  • \$\begingroup\$ Thanks for using the sandbox! This is mostly fine, except for two main things: "indicate that the solution as an exact match" is not really fleshed out enough, and that you don't have any test cases in the body. There are some other minor things, like specifying t and the allowable values to be integers (can they be negative?), and if returning all of the optimal x,y pairs is acceptable. As a challenge, I'm a bit unsure if this is particularly interesting due to its simplicity, but I don't think it is so simple that it is bad or anything. \$\endgroup\$ Commented Jun 16, 2019 at 17:56


Assume that there is a number a. In every conversion of that number, the number can either convert itself to its value plus the value of one of its digits, or convert itself to its value minus the value of one of its digits.

In this situation, find the minimum number of conversions one has to make in order to get to the number b. If b cannot be reached, output -1.


Input one line containing the integers a and b (where 1 <= a, and b <= 10^6).


Output a line containing one integer, which represents the minimum amount that one has to convert the number.

Example input and output (s)

Input:  1 10
Output: 5


-> 1+1=2
-> 2+2=4
-> 4+4=8
-> 8+8=16
-> 16-6=10
=> 10
  • 2
    \$\begingroup\$ At first sight this looks like just another shortest path question, but reachability is actually quite subtle. Do you have a proof of decidability? \$\endgroup\$ Commented Jun 22, 2019 at 18:33

Meta command-line flag parsing

The challenge

Write a program or function which determines whether or not a chosen command-line flag has been passed into the compiler or interpreter, but without accessing command-line parameters directly.


  • You must take in no input.
  • You are free to choose the flag that is being detected, but it must:
    • be a flag, meaning a boolean command-line parameter that does not take any arguments (e.g. -O3 in gcc)
    • not print anything by virtue of being present (e.g. flags that print a version number or usage help are disallowed)
  • The output must be 1 if the flag is present, and 0 otherwise.
  • The program must successfully run without errors, whether or not the flag is present.
  • You may not make use of the arguments passed to the compiler/interpreter, whether directly by value or by a property such as string length, array length.
  • Separately from the chosen flag, other command line parameters can be used, so long as they remain the same when running the code with and without the flag in question.
  • This is ; shortest answer in bytes wins!

Sandbox Questions

  • Are there any other edge cases or loopholes that should be accounted for here?
  • Is this a , despite the challenge not being related to input?

Feedback welcome!

  • 1
    \$\begingroup\$ I think this may have quite a few problems with determining what counts as a valid flag, what counts as accessing them, and what counts as an error. For example, Pyth has a flag -d which turns on debug printing. So any program with constant output would work. But that doesn't seem like what you have in mind, would it be acceptable? \$\endgroup\$ Commented Jun 21, 2019 at 19:57
  • \$\begingroup\$ @FryAmTheEggman Indeed, that's somewhat problematic. Maybe restricting output to either 1 or 0 would stop such trivial solutions from passing, though at the cost of punishing some other languages. I don't want to completely bar short-but-clever solutions, though, as I find that they can often be surprisingly creative; the example you posted is just something that would end up being used in too many solutions. \$\endgroup\$ Commented Jun 21, 2019 at 20:11
  • \$\begingroup\$ That may help, but you are right that it will restrict solutions a lot. I think this is a challenge that is worth running past quite a few people so you can get an idea of which flags are actually worth (and possible) to rule out. \$\endgroup\$ Commented Jun 21, 2019 at 20:23
  • 1
    \$\begingroup\$ Thinking of Japt here which takes its flags via an HTML input: if I'm understanding correctly, I wouldn't be allowed to simply read the value of that input, right? What about accessing Japt's internal variable for storing flags, would that be allowed? Also, should our solutions run no matter which flag is used or need we only support our chosen flag and no flag? Again thinking of Japt which has a flag that negates it's output; obviously that's going to be problematic! \$\endgroup\$
    – Shaggy
    Commented Jun 22, 2019 at 22:14
  • \$\begingroup\$ @Shaggy Both the input and internal variable would classify as "accessing the list of arguments", which is disallowed. Only two sets of command line parameters need to be supported: [some set of starting parameters not including the chosen flag, if any] and [the same set of starting parameters] + [the chosen flag]. I'll try to clarify both of these things. \$\endgroup\$ Commented Jun 23, 2019 at 5:09

GPA calculator

A GPA is sort of weighted average based on the grades obtained in certain courses. There are a number of different systems, but for this challenge we'll use the following grades and their numeric values:

A: 4
B: 3
C: 2
D: 1
F: 0
P: —

The grade is P is a special case. For some courses, students can sign up to take a course as Pass/Fail. In the case of a Pass (P), the course has no effect on their GPA whatsoever, but in the case of a Fail (F), the course does affect the GPA.

All courses are weighted by the number of hours/credits they are taken for. So give the following transcript:

Course            Hours   Grade
------            -----   -----
Next Gen Perl       6       A
Advanced C-minar    1       C
Starting with Java  3       B

We would calculate the total points as 35 (6*4 + 1*2 + 3*3), and then divide it by the number of hours, or 10 (6+1+3) for a final GPA of 3.5. If Starting with Java had been given an F, the total points would be 26 with 10 hours, for a GPA of 2.6. However, if it had been a P, the total points would still be 26, but with only 7 hours, the GPA would be 3.71428571429…


An arbitrarily long list/array/sequence/etc of transcript entries. The entries are simply a two item list/array/sequence/etc with one item being an integer value (doesn't have to be an actual Int value, floats are fine) of hours, and the other being the string ('A','B'...). The above example would have been:

  [6, 'A'],
  [1, 'C'],
  [3, 'B']

You may assume that the list will have at least one computable grade (so no empty transcripts, nor transcripts consisting only of P). If your language does not have strings, then you may use numbers for A/B/C/D/F/P, but their values must be similarly spaced (e.g. 10,11,12,13,14,25)


A number representing the GPA (float or other non-Int value)

Scoring and Restrictions

Codegolf, so shortest code in bytes wins. Standard rules apply.

Sandbox questions

I thought about allowing for either +/- grades (A+, A, A-, B+...) which would definitely prevent using ASCII values as easily, but then that led me to the thought of having a second input which is a list of possible grades with their equivalent value, but I think that would start over complicating it and way over benefit languages with built-in hashing. Thoughts?

  • \$\begingroup\$ Dupe \$\endgroup\$ Commented Jun 21, 2019 at 20:01
  • \$\begingroup\$ @FryAmTheEggman not a dupe. This one adds the P which makes it a good bit more complciated. \$\endgroup\$ Commented Jun 21, 2019 at 20:03
  • \$\begingroup\$ I disagree, solutions can just filter out all the grades with P and then exactly reuse the result from the question I linked. Unless I've missed something that definitely fits in the criteria for being a duplicate. \$\endgroup\$ Commented Jun 21, 2019 at 20:06
  • \$\begingroup\$ @FryAmTheEggman Between filtering, vastly different input/output (all of which could/would significantly modify entries — the input/output in particular was a strong criticism of the one you linked to), I think it's sufficiently different and golfed answers wouldn't be trivially different. But in any case, I asked about feeding in +/- or arbitrary grades which absolutely would be a totally different question. Perhaps instead of being dismissive, you could be constructive. \$\endgroup\$ Commented Jun 21, 2019 at 20:24
  • \$\begingroup\$ It is possible that I'm being too harsh, but I don't think I am. The only real problem with the other question compared to this one is the rounding, which could basically just be dropped from the answers. While I agree that this isn't identical, I don't think it has enough reason to exist as even with the +/- grades it's still just looking up values then applying a weighted average. It is certainly possible that I am in the minority, but I think my skepticism should encourage you to find other people's opinions before posting. \$\endgroup\$ Commented Jun 21, 2019 at 21:03
  • \$\begingroup\$ I agree with @FryAmTheEggman; the core of both challenges is identical. The only real difference here is having to filter the Ps which is a trivial modification for most languages. Having said that, though, I wouldn't swing my hammer at this until a few solutions had been posted, in the hope that I would be proven wrong. \$\endgroup\$
    – Shaggy
    Commented Jun 22, 2019 at 22:26



I originally wrote it in Chinese and am still working on translating it. Sorry for the inconvenience.

Given an array containing n integers A1,A2,…,An, where all of their initial values are 0. Please implement the four operations to the array described below:


Given the positive integer x and y, let every number in the range Ax~Ay (including Ax and Ay) be added by the given positive integer c;
Given the positive integer x and y, let every number in the range Ax~Ay be (similarly to the previous spec; the rest of the spec also conforms that rule) multiplied by a given positive integer c;
Given the positive integer x and y, let every number in the range Ax~Ay be changed to a given positive integer c;
Given the positive integer x and y, with regard to a given positive integer p, evaluate the value of the following expression :```[pow(Ax, p) + pow(A(x + 1), p) + … + pow(Ay, p)]7```


In the first line of the test case, there are two positive integers n(where 1<=n), which represents the length of the array.

Then, input operations sequentially, where each operation occupies one line, and the input will be 4 positive integers as in: type x y num, where type is either 1, 2, 3, or 4; these identifiers correspond to the corresponding operation as described above. The meaning of x and y corresponf to the description (where 1<=x<=y<=n). When type is not 4, num will be the positive integer given c(where 1<=c<=10000); when type is 4, num will be the given positive integer p(where 1<=p<=3)。


For each test case, for every 4 operation for the array, output the value of the corresponding expression.

Example input

3 3 5 7
1 2 4 4
4 1 5 2
2 2 5 8
4 3 5 3

Example output

  • 1
    \$\begingroup\$ Is it supposed to have restricted-complexity? Or pure golf? \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 11:37
  • \$\begingroup\$ I will add the category later after I translate the problem. \$\endgroup\$
    – user85052
    Commented Jun 27, 2019 at 11:46
  • 1
    \$\begingroup\$ In that case it's restricted-time. But you have to specify exactly how it is tested. 8 seconds doesn't mean much without the test cases and the the exact computer it will run on. restricted-complexity is much simpler. And just to confirm, you wrote this question, and it's not a homework problem from elsewhere, is it? \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 11:56
  • \$\begingroup\$ Yes, I wrote the question; I will accept your suggestion. \$\endgroup\$
    – user85052
    Commented Jun 27, 2019 at 11:58
  • 1
    \$\begingroup\$ Because, I think it is designed to be O(n+qlogn) in some algorithm competition. If more than that, any straightforward algorithm will probably work, and it's better to just removed the restriction. (Also note that on CG.se, in a challenge that there is a known best algorithm, coming up with the algorithm is generally not supposed to be the main challenge. If you go for O(n+qlogn), they will simply ask for a reference algorithm.) \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 12:10
  • 1
    \$\begingroup\$ Don't use multiple test cases. Make the list format flexible and q implicit. I'm not sure what's the best way to input multiple operations, but at least you could make operation types consistent and distinct value. If there is no time or complexity limit you don't need mod 10007. Even if there is a limit, it is probably better to make this an input. \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 12:18
  • 1
    \$\begingroup\$ I think if you wrote this you probably know that O(n+qlogn) and O(n^3) (or simply without restriction) makes a lot of differences. The code would be much longer for O(n+qlogn). And for reference, this question asking for suffix tree or alternatives had the first and only non-deleted answer about 2 years later. Not that you can't post it in either case, but... just make sure you know what you are doing. \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 12:29
  • 1
    \$\begingroup\$ I mean at least you could make operation types any consistent and distinct values, instead of 1, 2, 3, 4. \$\endgroup\$
    – jimmy23013
    Commented Jun 27, 2019 at 12:31

Answer chain: decreasing character scores

Assign a non-negative integer value to each letter in the alphabet, and write a function or program which takes as input a character string and outputs the sum of of the values of the letters in the string. Characters outside of [a-z] and [A-Z] have value 0.

Now run your code, taking as input the code of each of the previous answers in the order they were posted (including your own). The resulting sequence of scores should be strictly decreasing.

For example, if you are 4th to answer and your code defines a function \$f\$, then the output should verify \$f(c_1)>f(c_2)>f(c_3)>f(c_4)\$ where \$c_1\$ is the code of the 1st answer, \$c_2\$ is the code of the 2nd answer, ..., \$c_4\$ is the code of your answer.

When there will have been one week with no new answer, the 2nd-to-last answer will be declared the winner.


  • The values you assign may be the same as or different from those used by previous answers.
  • The value 0 is allowed.
  • Different letters may be assigned the same value.
  • Upper and lower case versions of the same letter may have different values.
  • Letters with diacritics, such as é or Ä, have value 0.
  • Your code need not handle letters which have not occurred in any of the previous answers (so if no answer so far has used the letter W, it's OK if your code fails on strings with a W).
  • 1
    \$\begingroup\$ Using characters like this seems to have some problems. Do answers get to pick encodings? Or is it really more like \$ f(c_{1},e_{1}) \$ where \$ e_{i} \$ is the encoding used by the ith answer? I think this may work better if you say it should work on bytes, then answerers can choose to interpret those bytes as characters in whatever encoding they want. \$\endgroup\$ Commented Jun 24, 2019 at 3:45
  • \$\begingroup\$ @FryAmTheEggman I have been thinking about this as well. I want it to be easy to set up on TIO, which means that solvers should be able to just copy-paste the code of previous answers. Forcing solvers to take into account the esoteric codepoints of golfing languages would be a hassle. I'd be more inclined to let answerers pick the encoding, but I suppose that might create potential loopholes. \$\endgroup\$ Commented Jun 24, 2019 at 14:34
  • \$\begingroup\$ Wouldn't asking for a hexdump of the code to be in each answer ensure that the solutions are fairly accessible? TIO also has problems with displaying particular characters (null bytes especially) and other things like that each text box has one fixed encoding. \$\endgroup\$ Commented Jun 24, 2019 at 14:39
  • \$\begingroup\$ @FryAmTheEggman How about restricting to letters (without diacritics)? Anything not in [a-z] or [A-Z] must have value 0. I think this circumvents the problem. Or maybe extend to letters+digits, or even all ASCII characters. \$\endgroup\$ Commented Jun 24, 2019 at 16:35
  • \$\begingroup\$ I don't think that works with your setup at all because a submission with no characters in whatever range you pick must always score exactly zero so it necessarily ends the chain. \$\endgroup\$ Commented Jun 24, 2019 at 16:49
  • \$\begingroup\$ @FryAmTheEggman Yes, and then whoever submitted that is guaranteed to lose, since the second-to-last answer wins. \$\endgroup\$ Commented Jun 24, 2019 at 20:41
  • \$\begingroup\$ Ah, I missed that, sorry. It still seems to have trouble with languages that could write all of their code using bytes outside the range you select and then throw in a small number of other characters, but I don't know if that is "broken" enough to make it unfun. \$\endgroup\$ Commented Jun 24, 2019 at 22:55
  • 1
    \$\begingroup\$ @FryAmTheEggman Potentially, an answerer could use only non-letters, add a single a, and then all following answers would need to assign a large value to a and avoid that letter in their own code. That seems OK to me, but I am a fan of lipograms... Anyway, I have updated the challenge following this helpful discussion; let's see if there are further comments. Thanks for your feedback! \$\endgroup\$ Commented Jun 25, 2019 at 22:05
  • \$\begingroup\$ Can a letter be assigned the value 0? That would be a non-negative integer \$\endgroup\$ Commented Jun 27, 2019 at 14:51
  • \$\begingroup\$ @NickKennedy Yes, 0 is allowed. \$\endgroup\$ Commented Jun 27, 2019 at 15:02

Enable 2-char Jsfuck

Provide a shortest code that makes JavaScript able to do everything JavaScript is supposed to do, i.e. able to access(so no deleting [].prototype.toString unless you have a backup, even if you can simulate one [].prototype.toString) and exec anything, with only [ and ]. You can choose your environment(FF/node/etc). Answering in 6-char JsFuck or something similar is welcomed.

E.g. If you run

Array.prototype[''] = 'a';
Array.prototype['a'] = 'b';
Array.prototype['b'] = function() { console.log(this[0]); };

then [][[]] = [][''] = 'a', and [][[][[]]] = []['a'] = 'b', so [[][[]]][[][[][[]]]] = ['a']['b'] = function(){console.log('a')}.

Of course, this is an invalid answer, because it can't do prompt(web browser) or fs(node) or anything similar, or even calling the console.log.

  • 1
    \$\begingroup\$ This is not clear as is. Are we making a transpiler from JS to JSFuck with the winning criterion code-golf? Do we need to support all JS input, or just some subset? \$\endgroup\$
    – lirtosiast
    Commented Jun 25, 2019 at 17:13
  • \$\begingroup\$ So a solution like translating "[[]"→( "[]["→) "[]]"→+ "][["→[ "][]"→] "]]["→! is valid? \$\endgroup\$
    – Adám
    Commented Jun 25, 2019 at 18:18
  • \$\begingroup\$ @lirtosiast I said "makes JS able to do everything in2c", not "make a language that do everything" \$\endgroup\$
    – l4m2
    Commented Jun 26, 2019 at 9:28
  • \$\begingroup\$ @Adám I don't think it's possible. +[] turns into []]][[][] which is not even valid JS code \$\endgroup\$
    – l4m2
    Commented Jun 26, 2019 at 9:29
  • 2
    \$\begingroup\$ @l4m2 Please include your not-golfed reference implementation so we can understand what you mean. \$\endgroup\$
    – Adám
    Commented Jun 26, 2019 at 9:34
  • \$\begingroup\$ @l4m2 If I understand this right, you're saying that we should write code that modifies the JavaScript environment such that all JavaScript code can be converted into JavaScript code that only uses characters [ and ]. Basically, add something to Array.prototype that makes the [] characters JavaScript-complete. \$\endgroup\$
    – wizzwizz4
    Commented Jul 24, 2019 at 13:28
  • \$\begingroup\$ @wizzwizz4 right \$\endgroup\$
    – l4m2
    Commented Aug 8, 2019 at 15:11

Tips for golfing in H

As for H as described here, what general tips do you have for golfing in H? I'm looking for ideas which can be applied to code-golf problems. Tips have to be specific to H (e.g. "remove comments" is an answer).

Tips in Standard H are also on-topic, although its spec is in a bad condition. You can use the H description provided in the first link to solve anything you don't understand about.

Please post one tip per answer. Also, please specify which H implementation you are using if your code runs only under a specific H implementation, as different H implementations can have different behavior.

  • 2
    \$\begingroup\$ Why do you want to ask this when H doesn't even exist? \$\endgroup\$
    – Adám
    Commented Jun 28, 2019 at 12:40
  • 2
    \$\begingroup\$ What's the rush? Also, H as linked, isn't a programming language, and it has a very strict definition, so there's no much room for golfing. The only thing I can think of is to omit quotes in the rare instances where you want a string that looks like one of the 1999 integers in the range -999–999. \$\endgroup\$
    – Adám
    Commented Jun 28, 2019 at 12:57
  • 1
    \$\begingroup\$ That would be a duplicate. \$\endgroup\$
    – Adám
    Commented Jun 28, 2019 at 13:10
  • \$\begingroup\$ Tips for golfing in Turing Machine but Way Worse also only has 1 tip. \$\endgroup\$
    – user85052
    Commented Jun 28, 2019 at 13:14
  • \$\begingroup\$ I can also think of a lot of those examples, as in: codegolf.stackexchange.com/questions/19423/… codegolf.stackexchange.com/questions/101557/… codegolf.stackexchange.com/questions/96682/… \$\endgroup\$
    – user85052
    Commented Jun 28, 2019 at 13:42
  • \$\begingroup\$ It does not matter whether it is a programming language or not; this question only cares whether H can be golfed or not. \$\endgroup\$
    – user85052
    Commented Jun 28, 2019 at 13:47

Find the closest hex colour shorthand

In CSS, colours can be specified by a "hex triplet" - a three byte (six digit) hexadecimal number where each byte represents the red, green, or blue components of the colour. For instance, #FF0000 is completely red, and is equivalent to rgb(255, 0, 0).

Colours can also be represented by the shorthand notation which uses three hexadecimal digits. The shorthand expands to the six digit form by duplicating each digit. For instance, #ABC becomes #AABBCC.

Since there are fewer digits in the hex shorthand, fewer colours can be represented.

The challenge

Write a program or function that takes a six digit hexadecimal color code and outputs the closest three-digit color code, where closeness is measured by adding together the difference between each component of the full color code and the corresponding component of the shorthand color code.

Here's an example:

  • Input hex code: #28a086
  • red component
    • 0x28 = 40 (decimal)
    • 0x22 = 30
    • 0x33 = 51
    • 0x22 is closer, so the first digit of the shortened color code is 2
  • green component
    • 0xa0 = 160
    • 0x99 = 153
    • 0xaa = 170
    • 0x99 is closer, so the second digit is 9
  • blue component
    • 0x86 = 134
    • 0x77 = 119
    • 0x88 = 136
    • 0x88 is closer, so the third digit is 8
  • The shortened color code is #298

Your program or function must accept as input a six digit hexadecimal color code prepended with # and output a three digit color code prepended with #.


  • #FF0000 → #F00
  • #00FF00 → #0F0
  • #D913C4 → #D1C
  • #C0DD39 → #BD3
  • #28A086 → #298
  • #C0CF6F → #BC7

Golf an H quine [duplicate]

Your task is to write a quine program in H (described in my esolangs.org userpage). (It is unclear whether it is possible.) Referring to Computer Science stack exchange, which seems unappropriate.

  • \$\begingroup\$ Can you explain your comment? I cannot understand it. \$\endgroup\$
    – user85052
    Commented Jul 1, 2019 at 11:19
  • 5
    \$\begingroup\$ Note that challenges that require the answers to be in a specific language are generally discouraged.* Is there any reason someone wanting to golf a quine in H would not just answer this? \$\endgroup\$
    – Adám
    Commented Jul 1, 2019 at 11:26
  • 1
    \$\begingroup\$ Now your espolangs userpage says Programming Puzzles and Code Golf. This is just what PPCG stands for. We are not called that any more. \$\endgroup\$
    – Adám
    Commented Jul 1, 2019 at 11:27
  • \$\begingroup\$ I am trying to retain the style of the rest of the esolangs pages. (e.g. see this page) \$\endgroup\$
    – user85052
    Commented Jul 1, 2019 at 11:30
  • 1
    \$\begingroup\$ Information contained in public Wikis does occasionally get stale. You should update all those references to PPCG. \$\endgroup\$
    – Adám
    Commented Jul 1, 2019 at 11:31
  • \$\begingroup\$ To clarify: is the current name "Code golf & Coding challenges"? \$\endgroup\$
    – user85052
    Commented Jul 1, 2019 at 11:34
  • 1
    \$\begingroup\$ The tour page seems to say Code Golf Stack Exchange. \$\endgroup\$
    – Adám
    Commented Jul 1, 2019 at 11:35
  • 1
    \$\begingroup\$ @Adám Wait, we're now known as CCSE? :S I thought the new name was indeed "Code Golf & Coding Challenges" (CGCC).. \$\endgroup\$ Commented Jul 1, 2019 at 11:37
  • 1
    \$\begingroup\$ As for the 'challenge', we already have a general quine challenge as linked by @Adám in his first comment. When you have more rep you could offer a bounty on that challenge stating you want to see if someone could come up with a quine in H. But, is H even released yet? If yes, maybe the first step might be to ask Dennis in the Nineteenth Byte Chat to add H to TIO, and also have a github with the source code/compiler of H. \$\endgroup\$ Commented Jul 1, 2019 at 11:40
  • 1
    \$\begingroup\$ The new site name is "Code Golf", and the subtitle is "& coding challenges". You can refer to the site with or without the subtitle \$\endgroup\$ Commented Jul 1, 2019 at 11:44
  • 1
    \$\begingroup\$ (referring to it off site should probably include "Stack Exchange" as other golfing sites are available) \$\endgroup\$ Commented Jul 1, 2019 at 11:49
  • 1
    \$\begingroup\$ Relevant \$\endgroup\$
    – Adám
    Commented Jul 1, 2019 at 11:56

Machine Learning Golf: Fashion MNIST

The first instance of Machine Learning Golf received a lot of intention but also revealed some problems (mostly caused by people in this community being too clever ;-)).

I intend to address those issues in this second installment:

Fashion MNIST is a dataset of 60,000 labelled 28x28 pixel grayscale images of fashion items (T-shirts/tops, Trousers, Pullovers, Dresses, Coats, Sandals, Shirts, Sneakers, Bags and Ankle boots).

Your task is to design and train a neural network that correctly identifies these images. Here are the rules:

Performance Goal

To qualify, your model must achieve at least 95% accuracy on the training set (i.e. you must identify at least 57,000 images correctly).


You may use any language and framework of your choice.

Your dataset must be taken as the current version of Fashion MNIST found here or in the references listed in that repo's README (for convenience).

You may reshape, permutate, rescale and offset entries in the dataset. However, modifications made to one entry must be performed on all entries.

You may transform the labels any way you like. (But you may, obviously, not change which labels correspond to which images.)

During training, you're allowed to use any dataset you want (in fact, you are allowed to come up with your weights however you like). Hence the above limitations are only relevant to assess whether your model meets the stated performance goal.

Your model

  • must be a 'traditional' feed forward neural network, i.e.a node's value is calculated as a weighted linear combination of some of the nodes in previous layers (which may include a bias node with constant value 1) followed by an activation function. Note that this allows you to skip layers and use convolutional/residual layers,
  • may only use the following standard activation functions:
    1. \$\mathrm{linear}(x) = x\$,
    2. \$\mathrm{softmax}(\vec{x})_i = \frac{e^{x_i}}{\sum_j e^{x_j}}\$,
    3. \$\mathrm{selu}_{\alpha, \beta}(x) = \begin{cases} \beta \cdot x & \text{, if } x > 0 \\ \alpha \cdot \beta (e^x -1 ) & \text{, otherwise} \end{cases}\$,
    4. \$\mathrm{softplus}(x) = \ln(e^x+1)\$,
    5. \$\mathrm{leaky-relu}_\alpha(x) = \begin{cases} x & \text{, if } x < 0 \\ \alpha \cdot x & \text{, otherwise} \end{cases}\$,
    6. \$\tanh(x)\$,
    7. \$\mathrm{sigmoid}(x) = \frac{e^x}{e^x+1}\$,
    8. \$\mathrm{hard-sigmoid}(x) = \begin{cases} 0 & \text{, if } x < -2.5 \\ 1 & \text{, if } x > 2.5 \\ 0.2 \cdot x + 0.5 & \text{, otherwise} \end{cases}\$,
    9. \$e^x\$
  • must take a single entry of the (preprocessed) training set as its only input and
  • return the predicted label, in the format you've specified, as its only output,
  • if a given weight occurs multiple times in your model, you may reuse it to lower your overall score,

Your answer must include (or link to) all code necessary to check your results -- including the trained weights of your model. In particular, you must include all preprocessing steps of your dataset.


The neural network with the smallest number of weights (including bias weights) wins.

Parameters used to preprocess your data don't count as weights.



There is a tutorial for the Fashion MNIST dataset available on Tensorflow which serves as an excellent starting point if you decide to use TF as your framework.

This challenge had already been posted on the main site and got closed there. The purpose of this post is threefold:

  1. Is the community interested in this particular (and this kind of) challenge (in general)?
  2. How can the user experience for the target audience be improved? Specifically:
    • What are likely causes of confusion?
    • Can we streamline the structure?
    • Are some of the rules too restrictive or not restrictive enough?
    • Are there any loopholes I've overlooked? (For instance, preprocessing the training data opens up the possibility of encoding labels into the feature set. Some of the rules are already designed to prohibit this but it's entirely possible that they don't suffice.)
  3. Any other constructive feedback?
  • 2
    \$\begingroup\$ I think a lot of community members don't know anything about machine learning. You might try saving this one for later and working to think of some entry-level challenges that work as a sort of tutorial when done together. \$\endgroup\$
    – mbomb007
    Commented Jul 4, 2019 at 16:59
  • \$\begingroup\$ @mbomb007 While I like the idea, it seems pretty tough to come up with challenges that are both simple enough to solve without knowing machine learning and yet won't just get solved without using ML at all. That's what happened with my entry-level challenge. Still, I'm sure there are a few people here who would enjoy this kind of challenge and seeing them might spark the interest of others. \$\endgroup\$ Commented Jul 4, 2019 at 17:21
  • 4
    \$\begingroup\$ I think it's worth considering the questions and requests for clarifications you were asked in the comments of your previous challenge and making sure you've clearly addressed all of them here. \$\endgroup\$
    – xnor
    Commented Jul 4, 2019 at 20:17
  • 2
    \$\begingroup\$ I think there's a good chance the best strategy for this challenge isn't anything that looks like training a neural net, but some form of massive over-fitting like looking up a few specifically-chosen pixels with a lookup table, implemented with neural net primitives. I'd suggest you make a good effort at trying to optimize both strategies yourself to see how that turns out. \$\endgroup\$
    – xnor
    Commented Jul 4, 2019 at 20:22
  • \$\begingroup\$ @xnor The previous issues have been addressed, as far as I can tell, and I'm not concerned about overfitting. If you can identify a few relevant bits of information to pass the test while staying within the rules, that's fine with me. \$\endgroup\$ Commented Jul 5, 2019 at 4:36
  • \$\begingroup\$ I should make clear that the intention of this challenge is to provoke people to come up with tricky ways to lower their score -- it's not meant to result in neural networks you would actually use in production. Hence I'd like the rules only to guard against approaches that trivialize this task (like encoding labels into the feature set via preprocessing). And testing indicates that you might be able to come up with a solution using \$\ll 1000\$ weights -- beating every traditional approach by a long shot. (Which, to me, is pretty exciting.) \$\endgroup\$ Commented Jul 5, 2019 at 4:55
  • \$\begingroup\$ If doing hardcoding-style stuff is fine, then it's confusing to mention training as in "Your task is to design and train a neural network" if training is actually optional. And "I intend to address those issues in this second installment" makes it sound to me like you're trying to make the challenge watertight against unexpected clever approaches, which I think isn't what you're trying to convey. \$\endgroup\$
    – xnor
    Commented Jul 6, 2019 at 2:05
  • 2
    \$\begingroup\$ Let me also reiterate what mbomb007 said about many community members not knowing anything about machine learning. I think it would be good to write up a friendly introduction with everything that's needed for the challenge while keeping unfamiliar terminology to a minimum. Separately, scoring by distinct weights strikes me as a really bad idea. I wouldn't be surprised if the optimal score is 1, achieved by an enormous network all with only weight 1 that encodes the whole dataset using its topology or its choice of gates. Even convolution layers seem very exploitable combined with copying. \$\endgroup\$
    – xnor
    Commented Jul 6, 2019 at 2:10
  • \$\begingroup\$ @xnor 'if training is actually optional'.. Hardcoding weights, as far as I'm concerned, is a form of training. That's fine -- you may come up with your weights any way you like. But I think you're right about the distinct weights bit. I'll add a mathematically precise definition of what I have in mind that will also serve as a rough introduction to neural networks. I didn't intially want to do that because I think it distracts from the actual challenge but at this point I'm convinced it needs to be done to avoid disputes. \$\endgroup\$ Commented Jul 6, 2019 at 3:57
  • 1
    \$\begingroup\$ @xnor, yes: it's quite easy to show that \$\textrm{selu}_{0,1}\$ and weight -1 gives a Turing-complete system. \$\endgroup\$ Commented Jul 6, 2019 at 7:23
  • \$\begingroup\$ @PeterTaylor That can't be right. You have to, at least, allow recurrent layers or some similar feature. Otherwise you won't be able to model the Ackermann function. \$\endgroup\$ Commented Jul 6, 2019 at 22:20

Calculate a constant from its characteristic function

You are given in some language-dependent way (e.g. a lambda parameter) a characteristic function which is basically another way of saying a function. This function will take an integer input and return true or false, or 1 or 0, or any other two 1-byte values of your choice. You can decide whether you need the function to be 0-indexed or 1-indexed.

Your job is then to take this function and calculate the floating-point value obtained by interpreting the values of this function as a binary fraction. You must use as many significant digits as possible. I will assume IEEE double precision by default but you many use single precision if your language specifically supports it.

For example, if a function returns the sequence 0, 1, 1, 0, 1, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0 (this is OEIS A010051), then your program or function will output 0.41468250985111166 as any further terms will be less than the precision of a double-precision value.

You can decide whether you want to round the 54th significant bit up or not.

This is , so the shortest program or function that breaks no standard loopholes wins!

  • \$\begingroup\$ I'm missing an explicit statement of where the binary point goes. Also, some tests which cover corner cases (minimum, maximum, denormal). \$\endgroup\$ Commented Jul 8, 2019 at 10:57
  • \$\begingroup\$ @PeterTaylor The binary point goes before the first value as retrieved, so 0, 1, 1, 0, 1... becomes .01101.... For corner cases would it help if I said that at least one of the first 1000 elements will be "true"? \$\endgroup\$
    – Neil
    Commented Jul 8, 2019 at 11:30
  • \$\begingroup\$ That would eliminate 0 as a possible output, whereas I think is itself one of the corner cases. It would also eliminate denormal numbers. The smallest positive double value is \$2^{-1074}\$. \$\endgroup\$ Commented Jul 9, 2019 at 6:19

Make a bigger number

Each answer on this question must be a complete program (not a function) that outputs a positive integer in decimal (trailing newlines are permitted). Output should be either passed to STDOUT or displayed on the screen. The output number must also be larger than all the numbers output by all the valid answers at the time of posting. One person may not post two consecutive answers.


Your answer will be scored with the following formula


Where \$B\$ is the number of bytes in your answer and \$B'\$ is the number of bytes in the longest answer before your answer posted by someone other than yourself. Lower scores will be better.

  • \$\begingroup\$ Could you please clarify some more rules? E.g, in these sorts of challenges, using the same languages twice is prohibited. Even if you're fine with it, just clarify that to avoid confusion. Also, state when it will end and whether a lower/higher score is better. Further, some languages don't have an 'integer' type, while others have many - how could they get around this? \$\endgroup\$ Commented Jul 9, 2019 at 20:19
  • \$\begingroup\$ @GezaKerecsenyi Prohibiting double language use is only used in specific scenarios it is in general it tends to be very bad rule to have as it only causes confusion. I definitely do not see it as the default and I don't want to give the impression that it is. Since programs have to be complete for this challenge (we are viewing programs as black boxes) types should not be a problem. Maybe I need to make this clearer in the post. And on the lower vs higher, that is definitely something important I was missing. Thanks! \$\endgroup\$
    – Wheat Wizard Mod
    Commented Jul 9, 2019 at 21:08
  • \$\begingroup\$ Answer chaining questions should have some inherent mechanism whereby it gets harder to answer over time. As it stands, this can run forever. \$\endgroup\$ Commented Jul 10, 2019 at 13:43
  • \$\begingroup\$ @PeterTaylor I was under the impression that later answers would get harder because those challenges have scoring mechanisms that benefit later answers. \$\endgroup\$
    – Wheat Wizard Mod
    Commented Jul 10, 2019 at 13:51
  • \$\begingroup\$ I'm not certain that I'm interpreting that comment correctly. The answer chaining challenges I've participated in didn't have scoring per se, but "the penultimate answer wins". \$\endgroup\$ Commented Jul 10, 2019 at 15:01
  • \$\begingroup\$ @PeterTaylor I was being a bit more general in my statement but that is the specific mechanism that I see a lot. \$\endgroup\$
    – Wheat Wizard Mod
    Commented Jul 10, 2019 at 15:01
  • \$\begingroup\$ Hm... actually, Peter Taylor's point isn't entirely wrong. An issue I can see is that the scoring formula will probably end up making each answer at least exponentially longer than the previous one, and that two or more people could actually duel just by adding more bytes to their code, e.g. A: 50000 9s, B: 5E50 9s, A: 5*10^99E99 9s, etc. In my opinion, the previous scoring formula (\$\frac{B_{n-1}+1}{B_n+1}\$) was much better, but with the highest score being the leader. \$\endgroup\$ Commented Jul 10, 2019 at 17:24
  • \$\begingroup\$ I think I can parse that response, but I don't get the subtext at all, so I'll rephrase my point as a question in the hope of eliciting a concrete response: what prevents two people from spinning this out to infinity just by copying the previous answer and adding +1? \$\endgroup\$ Commented Jul 12, 2019 at 10:49
  • \$\begingroup\$ @PeterTaylor In my mind two things 1) it probably won't get them better scores 2) it wouldn't be very fun. I am a little concerned though since these are not excellent motivators. I am going to think about other scoring mechanisms that might further disincentivise this. Or perhaps some restrictions on answers. \$\endgroup\$
    – Wheat Wizard Mod
    Commented Jul 12, 2019 at 13:34

Output Ordinal Numbers up to n

Moved to Output Ordinal Numbers up to n.

118 119
121 122

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .