Boggle Board Compression

Question

When working on Non-Palindromic Polyglot Boggle, I found it quite tedious to pack the codes as efficiently as possible onto the Boggle board, even with only two strings. But we're programmers, right? We know how to automate things.

Given a list of strings, you're to generate a Boggle board on which each of those strings can be found (independently of the others). The challenge is to make the Boggle board as small as possible. As this is (hopefully) a rather difficult task, this is a code-challenge: there is no requirement for optimality - the challenge is to do it as well as you can.

Rules

• The Boggle board will be rectangular and contain only upper case letters. Therefore, the input strings will also contain only upper case letters.
• The usual Boggle rules apply: a string is part of the board if, starting anywhere, you can find the string by repeatedly moving to adjacent characters (horizontally, vertically, or diagonally). To form a single string, you cannot use any cell of the board more than once. However, characters may be reused between different strings.
• You've got 30 minutes to process the test data, and your code must not use more than 4 GB of memory. I will give a little bit of leeway on the memory limit, but if your program consistently uses more than 4 GB or spikes significantly above it, I will (temporarily) disqualify it.
• I will test all submissions on my own machine, which is running Windows 8. I do have a Ubuntu VM, but if I have to test on that you won't be able to make as much use of the 30 minutes as otherwise. Please include a link to a free interpreter/compiler for your chosen language, as well as instructions on how to compile/run your code.
• Your score will be the size of the Boggle board for the test data below (not counting the newlines). In the case of a tie (e.g. because multiple people managed to produce an optimal solution), the winner will be the submission that produces this optimal solution faster.
• You must not optimise your code specifically towards the test data. If I suspect anyone of doing so, I reserve the right to generate new test data.

Example

Given the strings

FOO
BAR
BOOM

Once could trivially put them in a 4x3 Boggle board:

FOOX
BARX
BOOM

By making use of the fact that strings don't have to be straight, we can compress it to 5x2:

BORFO
OMABO

But we can make it even smaller by reusing characters between different strings, and fit the strings in 4x2:

FOOM
BARX

Now the B is used for both BOOM and BAR, and the OO is used for both BOOM and FOO.

Test Data

Your submission will be tested on the following 50 strings. For testing purposes you can simply use smaller subsets of this data which should then run more quickly. I believe that the absolute lower bound for this test data is a board with 120 characters, although this is not necessarily achievable.

T
WP
GVI
CIHM
EGWIV
QUTYFZ
LWJVPNG
XJMJQWSW
JLPNHFDUW
SWMHBBZWUG
XVDBMDQWDEV
TIUGAVZVUECC
IWDICFWBPSPQR
MMNWFBGMEXMSPY
YIHYXGJXKOUOIZA
BZSANEJNJWWNUJLJ
XTRMGOVPHVZYLLKKG
FLXFVVHNTWLMRRQYFQ
VZKJRAFQIYSBSXORTSH
FNQDIGCPALCHVLHDNZAV
GEAZYFSBSWCETXFKMSWLG
KWIZCEHVBDHEBGDGCJHOID
SKMQPHJAPDQKKHGTIPJCLMH
ZSFQDNYHALSUVWESQVVEUIQC
HXHBESUFCCECHNSTQGDUZPQRB
DSLXVHMOMLUXVHCNOJCBBRPVYB
DVTXKAOYYYRBVAVPSUAOYHIPPWN
PJAIYAWHMTNHTQDZDERPZYQEMLBZ
SYNSHJNOIWESMKWTBIANYUAUNRZOS
WADGUKIHUUFVRVUIBFUXQIOLAWIXAU
LGLXUFIXBEPSOFCKIAHXSHVKZPCXVPI
LIUYFHITTUYKDVQOZPNGZLWOZSRJTCTZ
IZDFTFFPNEBIYGVNTZHINICBXBXLBNBAL
BSKQNTPVUAVBXZGHVZCOUCRGCYISGFGYAS
DPGYYCIKDGCETXQOZGEQQLFQWACMVDTRYAT
RQDNIPGUHRYDRVHIPJLOWKBXMIBFAWCJGFMC
PFKOAGEQLXCMISSVEARWAPVYMRDCLSLPJOMQQ
EQPCNHQPTWABPFBVBXHQTFYELPNMNCWVKDDKGR
RAHTJMGIQJOJVWJBIHVRLJYVCSQJCKMEZRGRJMU
SZBJBPQYVYKDHAJHZMHBEWQEAQQKIEYCFACNLJBC
ANVDUCVXBPIZVRAXEBFEJOHSYKEKBIJELPIWEYXKH
DJUNPRLTISBFMGBEQNXSNUSOGDJNKESVKGAAMTIVXK
TZPUHDSHZFEURBNZTFBKXCDPYRELIAFMUWDIQTYWXGU
FJIKJROQSFSZUCGOOFJIEHBZREEUUSZWOLYFPCYHUSMR
TPMHJEAWVAJOCSDOPMQMHKRESBQSTRBXESYGCDVKLFOVS
ABJCCDJYMYDCYPZSGPGIAIKZQBYTZFDWYUZQBOESDSDGOY
IIHKTVPJNJDBCBOHCIYOPBKOVVKGNAKBDKEEKYIPRPHZOMF
IABGEPCSPNSMLVJBSGLRYNFSSYIALHWWAINTAVZAGJRVMDPW
GFMFVEFYJQJASVRIBLULUEHPMZPEXJMHIEMGJRMBLQLBDGTWT
YPWHLCVHQAVKVGHMLSOMPRERNHVYBECGCUUWTXNQBBTCMVTOVA

Verifier

You can use the following Stack Snippet to verify whether a Boggle board contains all strings in a given list. I ported the Boggle search code from edc65's answer over here. Let me know if anything seems buggy.

function verify() {
  var strings = document.getElementById("strings").value;
  var board = document.getElementById("board").value;
  strings = strings.match(/[A-Z]+/g) || [];
  board = board.split('\n');
  var valid = true;
  var len = board[0].length;
  var result = 'Valid';
  for (var i = 0; i < board.length; ++i)
  {
    if (board[i].length != len)
    {
      valid = false;
      result = 'Invalid: Board not rectangular.';
    }
    if (/[^A-Z]/.test(board[i]))
    {
      valid = false;
      result = 'Invalid: Board contains invalid character on line '+i+'.';
    }
    if (!valid) break;
  }
  if (valid)
    for (i = 0; i < strings.length; ++i) {
      if (!findInBoard(board, strings[i]))
      {
        valid = false;
        result = 'Invalid: String "'+strings[i]+'"not found in board.';
        break;
      }
    }
  document.getElementById("output").innerHTML = result;
}

function findInBoard(b, w) {
  var l = b[0].length;
  var r = 0;
  var Q = function(p, n) {
    return (r |= !w[n]) || 
    (
      b[p] = 0,
      [1,l,l+1,l+2,-1,-l,-l-1,-l-2].map( function(q) { return b[q+=p]==w[n] && Q(q,n+1) }),
      b[p] = w[n-1]
    );
  }
  b = (b+'').split('');
  b.map(function(e, p) { return e==w[0] && Q(p,1) });
  return r;
}

Strings that should be part of the Boggle board:
<br>
<textarea id="strings" cols="60" rows="10"></textarea>
<br>
Boggle board:
<br>
<textarea id="board" cols="60" rows="10"></textarea>
<br>
<input type="button" value="Verify" onclick="verify()" />
<br>
<br>
<div id="output"></div>

Answer 1

C++11, score = 992 1024

The algorithm is really stupid, but so far no one else made a serious one either, so I'll post it. It more or less randomly sticks words on a square board, and then starts over if it doesn't manage to make them fit. It kind of tries to maximize overlap with the existing words, but in a really inefficient way.

Edit: Improved score by adding 1 to a side length and trying a rectangle after failing 50 times. Also sorted input strings by size instead of randomizing the order.

#include <iostream>
#include <cstring>
#include <string>
#include <random>
#include <algorithm>
using namespace std;

struct grid {
    char *g;
    int h,w;
    grid(int h, int w):h(h),w(w) {
        g = new char[h*w];
        memset(g,0,h*w*sizeof(*g));
    }
    grid(const grid &o) {
        h=o.h, w=o.w;
        g = new char[h*w];
        memcpy(g,o.g,h*w*sizeof(*g));
    }
    grid(grid &&o) {
        h=o.h, w=o.w;
        g = o.g;
        o.g = 0;
    }
    grid& operator=(const grid &o) {
        h=o.h, w=o.w;
        memcpy(g,o.g,h*w*sizeof(*g));
        return*this;
    }
    grid& operator=(grid &&o) {
        h=o.h, w=o.w;
        g = o.g;
        o.g = 0;
        return*this;
    }
    char& operator()(int i, int j) {
        return g[i*w+j];
    }
    ~grid() { delete []g; }
};
typedef struct { int n, i, j; grid g; } ng;


const int qmax = 140;
const bool sizesort = true;
const int maxtries = 50;

inline int sq(int x){return x*x;}
bool operator<(const ng &a, const ng& b) {return a.n < b.n;}
void search(vector<string>& s) {
    int tl = 0;
    for(auto&x: s) tl += x.size();
    int l = 0;
    while(l*l < tl) l++;
    vector<string*> v;
    for(size_t i = 0; i < s.size(); i++) v.push_back(&s[i]);
    struct{bool operator()(string*a,string*b){return a->size()>b->size();}} scmp;
    if(sizesort) sort(v.begin(), v.end(), scmp);
    mt19937 rng;
    for(;;l--) {
        int tries = 0;
        int side2 = l;
        retry:
        tries++;
        if(!sizesort) shuffle(v.begin(), v.end(), rng);

        if(tries == maxtries) cout<<"rectangle",side2++;
        grid g(l,side2);

        for(string* x: v) {
            string& z = *x;
            vector<ng> p;
            for(int i = 0; i < g.h; i++)
            for(int j = 0; j < g.w; j++) {
                if(g(i,j) && g(i,j) != z[0]) continue;
                p.push_back({!g(i,j), i,j, g});
                p.back().g(i,j) = z[0]|32;
            }
            for(size_t zi = 1; zi < z.size(); zi++) {
                vector<ng> p2;
                for(ng &gg: p) {
                    for(int i = max(gg.i-1,0); i <= min(gg.i+1,g.h-1); i++)
                    for(int j = max(gg.j-1,0); j <= min(gg.j+1,g.w-1); j++) {
                        if(!gg.g(i,j) || gg.g(i,j) == z[zi]) {
                            p2.push_back({gg.n+!g(i,j),i,j,gg.g});
                            p2.back().g(i,j) = z[zi]|32;
                        }
                    }
                }
                shuffle(p2.begin(), p2.end(), rng);
                sort(p2.begin(), p2.end());
                if(p2.size() > qmax) p2.erase(p2.begin() + qmax, p2.end());
                p = move(p2);
            }
            if(p.empty()) goto retry;
            g = p[0].g;
            for(int i = 0; i < g.h; i++)
            for(int j = 0; j < g.w; j++)
                g(i,j) &= ~32;
        }
        cout<<g.w*g.h;
        for(int i = 0; i < g.h; i++) {
            cout<<'\n';
            for(int j = 0; j < g.w; j++)
                cout<<(g(i,j)?g(i,j):'X');
        }
        cout<<endl;
    }
}

int main()
{
    vector<string> v = {"T","WP","GVI","CIHM","EGWIV","QUTYFZ","LWJVPNG","XJMJQWSW","JLPNHFDUW","SWMHBBZWUG","XVDBMDQWDEV","TIUGAVZVUECC","IWDICFWBPSPQR","MMNWFBGMEXMSPY","YIHYXGJXKOUOIZA","BZSANEJNJWWNUJLJ","XTRMGOVPHVZYLLKKG","FLXFVVHNTWLMRRQYFQ","VZKJRAFQIYSBSXORTSH","FNQDIGCPALCHVLHDNZAV","GEAZYFSBSWCETXFKMSWLG","KWIZCEHVBDHEBGDGCJHOID","SKMQPHJAPDQKKHGTIPJCLMH","ZSFQDNYHALSUVWESQVVEUIQC","HXHBESUFCCECHNSTQGDUZPQRB","DSLXVHMOMLUXVHCNOJCBBRPVYB","DVTXKAOYYYRBVAVPSUAOYHIPPWN","PJAIYAWHMTNHTQDZDERPZYQEMLBZ","SYNSHJNOIWESMKWTBIANYUAUNRZOS","WADGUKIHUUFVRVUIBFUXQIOLAWIXAU","LGLXUFIXBEPSOFCKIAHXSHVKZPCXVPI","LIUYFHITTUYKDVQOZPNGZLWOZSRJTCTZ","IZDFTFFPNEBIYGVNTZHINICBXBXLBNBAL","BSKQNTPVUAVBXZGHVZCOUCRGCYISGFGYAS","DPGYYCIKDGCETXQOZGEQQLFQWACMVDTRYAT","RQDNIPGUHRYDRVHIPJLOWKBXMIBFAWCJGFMC","PFKOAGEQLXCMISSVEARWAPVYMRDCLSLPJOMQQ","EQPCNHQPTWABPFBVBXHQTFYELPNMNCWVKDDKGR","RAHTJMGIQJOJVWJBIHVRLJYVCSQJCKMEZRGRJMU","SZBJBPQYVYKDHAJHZMHBEWQEAQQKIEYCFACNLJBC","ANVDUCVXBPIZVRAXEBFEJOHSYKEKBIJELPIWEYXKH","DJUNPRLTISBFMGBEQNXSNUSOGDJNKESVKGAAMTIVXK","TZPUHDSHZFEURBNZTFBKXCDPYRELIAFMUWDIQTYWXGU","FJIKJROQSFSZUCGOOFJIEHBZREEUUSZWOLYFPCYHUSMR","TPMHJEAWVAJOCSDOPMQMHKRESBQSTRBXESYGCDVKLFOVS","ABJCCDJYMYDCYPZSGPGIAIKZQBYTZFDWYUZQBOESDSDGOY","IIHKTVPJNJDBCBOHCIYOPBKOVVKGNAKBDKEEKYIPRPHZOMF","IABGEPCSPNSMLVJBSGLRYNFSSYIALHWWAINTAVZAGJRVMDPW","GFMFVEFYJQJASVRIBLULUEHPMZPEXJMHIEMGJRMBLQLBDGTWT","YPWHLCVHQAVKVGHMLSOMPRERNHVYBECGCUUWTXNQBBTCMVTOVA"};
    search(v);
    return 0;
}

The board:

TGBHXEXMPYTECWBSFYOXKXKXFSQJXKXX
UZBWMLJKSXXPIXSVYYXATVDLVOCVCXMT
WWPSJGBUFWPOUHPAKRZMXIPCHHXJYEAX
SQPBNXFQNMLAYSQVBGAESYGWQJOVXJZY
RJXWJJDWMNSGFUQXSEAWXBMIJAYWLRTR
XMXFEIWIHTIHIGMOJTKVARJNPSVFJVDG
XJHNCGCCQXTYLSJHVNOJXHTSCKERBHMR
XVCLACEDGTCCDGLPMCJXXMQMQPVGIAJC
DLZSFPURZYUGRKENTWSDPVLSHBFFHBMA
HNBUQYZPDOKNMYDDVBBOGJBQGKSMLUWQ
XXZRSEBQNCQDPVFNKSSQNCDLXERPMSLF
XKVAIHMHTGZVUXPTQUSXXGEBRXEZHOUQ
XRJUXYNLQFJLHAVQPNNXTCXYMRJPKEQH
FAPIHATDBSWXWGBHCQHPBWUVNMJGKGVP
XQVVWMLJRZZOVRZXESFQTAUFHIMLLYZV
XIWXUSOQTXTLSEABVBIPXGXSYEAEMGOX
WEYQCBIUXCNSJKGRFZXTBXXSMFLIRYPQ
XGSBIPFLPAMIBEEMVEJLXDWDUBHTYXDX
XQSUXIZQGFOCPLKSUOAREVYQDWDVXCTA
XXVEBKXLEKCIOFYYHCBJPCTKIAWKIMZE
ORVEUGVIXYEWFPCCDJCNUDANHIBODFCI
XTOFPUSHKQXAZYDYIYOBJZVZBFMXLGOU
SZNSCXHLWQQSGCTQMBPDGAXXTRACJJXO
HRUAUKIAXEUOPGUIKWRJMNKKDGUWPBGK
ZVEYNGDNBUEOJIAZQORVGBDSEEOYIYXX
VMHCCAIBHRHFAESSBXVJKQOSKYFZHVPB
ALEVJPTWLMZJHXXFJYXIZUEFLIGUSRRB
GZCBDHPKMXXBXDSBTZJDUYVXNQPPHDYC
UIPJQKEQSBNICKYPNEFHWVHFDFRUZOJX
KWTGKXGBEOIJHNVQFBTLNTNMYXLTXNMX
DIOHJCXDGWHZTSGYIFJPMRRQOMXVHCFX

Answer 2

Python 3, score = 1225

In this solution we use only 1 row. Starting from the empty string in every step we add the word which guarantees the largest possible overlap. (The words are checked in all 4 possible orientations.)

This gives a score of 1225 which is 50 lower than the score of 1275 for concatenating all the words together with no overlapping.

The result:

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

The code:

import sys

def com_len(a,b):
    for i in range(min(len(a),len(b)),-1,-1):
        if (a[-i:] if i else '')==b[:i]:
            return i

def try_s(a,b,sn,mv):
    v=com_len(a,b)
    if v>mv:
        return a+b[v:],v
    return sn,mv

ws=[w.rstrip() for w in sys.stdin.readlines()]
s=''
while ws:
    mv=-1
    sn=None
    mi=None
    for i in range(len(ws)):
        mvo=mv
        for a in [s,s[::-1]]:
            for b in [ws[i],ws[i][::-1]]:
                sn,mv=try_s(a,b,sn,mv)
        if mvo<mv:
            mi=i
    s=sn
    del ws[mi]
print(s)

Answer 3

C, score 1154

AVOTVMCTBBQNXTWUUCGCEBYVHNRERPMOSLMHGVKVAQHVCLHWPYGOSUNSXNQEBGMFBSITLRPNUJDXJMHIEMGJRMBLQLBDGTWTWVJOJQIGMJTHARYPDCXKBFTZNBRUEFZHSDHUPZTCUZSFSQORJKIJFYNDQFSZOWLZGNPZOQVDKYUTTIHFYUILHWWAINTAVZAGJRVMDPWPQRBEAWVAJOCSDOPMQMHKRESBQSTRBXESYGCDVKLFOVSOYNUAUYNAIBTWKMSEWIONJHSNYSKYVYQPBJBZSOHCIYOPBKOVVKGNAKBDKEEKYIPRPHZOMFGJCWAFBIMXBKWOLJPIHVRDYRHUGPINDQRQPSPBWFCIDWIVPAWRAEVSSIMCXLQEGAOKFPFFTFDZIWLMRRQYFQSXORTSHLUXVHCNOJCBBRPVYBMKSNTPVUAVBXZGHVZCOUCRGCYISGFGYASOZGEQQLFQWACMVDTRYATIAJPVUECCYFSBSWCETXFKMSWLGFUUHIKUGDAWIYXCPZKVHSXHAIKCFOSPEBXIFUXLGLJNHFDUWSWQJMJXWSMXEMGBFWNMMGOVPHVZY
NDUCVXBPIZVRAEBFEJOHSYKEKBIJLPIWEYXKKXITMAGKSEKNJDFMFVEFYJQJASVRILULUEHMZPEUMRGRZKCJQSCVYJRVHIBJUGXWYTDWUFAILERMSUHYCPYLOWZSUERBEIJFOOGQIEVVQEWVUSLAHZTCTJRIABGEPCSNSMLJBSGLRYNFSSYAXHBESUFCCECHNSTQGUZTMHJABJCCDYMYDCYZSGPGIAIKZBYZFDWYUZQBOESDSDGZRCBJLNCAFCYEIQQAQEBHMZJAHDIIHKTVJNDCBNWPPHAUSPVABRYYYOAKXTVDWQDMBDVXCRGKDDKVWCNNPLEYFTQHXBBFPBAWTPQHCPEQMOJLSLCDRMYLABNBLXBXBCNIHZTNVYIBENLXVVHNTVZKJAFISBDLVHMOMCJPITGHKKQDPAJHPQBSQKWIZCEHDHEBDGCJHIDDPYYCKDCETXQZBLMYZPREDZDQTHNMHWYUGVZIWGAZUAXIWALOIQUBIUVRVAZIOUOXJXYHPVAZNDHLCLAPCGDQNBZANJNJWWNUJLPGPVJWLGUZBBHMYPHICQUTYZXTRVIXGKKLL

Use two lines so newly added words can reuse the letters from the top line.

char l[2][2000];
char s[2][2000];
char w[100][200];
int d[200];
void pri() {
    puts(l[0]);
    puts(l[1]);
}
void sav() { memcpy(s,l,sizeof(l)); }
void res() { memcpy(l,s,sizeof(l)); }
int fit(char *t, int l0, int l1) {
    if (!*t) return 0;
    if (l[0][l0] == *t && (l0 <= l1 || l[1][l1])) return 1+fit(t+1,l0+1,l1+(l1<l0));
    if (l[1][l1] == *t && (l1 <= l0 || l[0][l0])) return 1+fit(t+1,l0+(l0<l1),l1+1);
    if (!l[0][l0]) {
    strcpy(l[0]+l0,t);
    return 0;
    }
    if (!l[0][l1]) {
    strcpy(l[0]+l1,t);
    return 0;
    }
    if (!l[1][l1]) {
    l[1][l1] = *t;
    return fit(t+1,l0+(l0<l1),l1+1);
    }
    return 1000;
}
int main(){
    int j,i,n,best,besti,bestk,c,tot = 0;
    for (i = 0; scanf("%s ",w[i])>0; i+=2) {
    int j = strlen(w[i]);
    for (c = 0; c < j; c++) w[i+1][c] = w[i][j-c-1];
    }
    n = i;
    pri();
    for (j = 0; j < n/2; j++) {
    int k = -1;
    best = -1;
    for (k = 0; k <= strlen(l[1]); k++) {
    for (i = 0; i<n; i++) if (!d[i/2]) {
    sav();
    c = fit(w[i],k,k);
    if (c < 1000 && c >= best) best = c, besti = i, bestk = k;
    res();
    }
    }
    fit(w[besti],bestk,bestk);
    d[besti/2] = 1; tot += best;
    }
    pri();
    printf("%d - %d\n",tot, strlen(l[0])+strlen(l[1]));
    return 0;
}

Answer 4

CJam, 1122 1089 letters

qN%{,}$__W%Wf%+:A;s[,mQ)__2#_S*:B;+2/:P;:DX1$W*W]:M;1:L;{BQCelt:B;}:T;
{A,,{B:G;P:H;D:I;L:J;A=:Z{:C;PL+D-:QB=C=
{T}{QD+:QB=C={T}{BPCelt:B;PD+:PL+B=' ={MD#)M=:D;ML#)M=:L;}&}?}?}/BS-,Z,-G:B;H:P;I:D;J:L;}$
0=:KA={:C;PL+D-:QB=C={T}{QD+:QB=C=
{T}{BPCelt:B;PD+:PL+B=' ={MD#)M=:D;ML#)M=:L;}&}?}?}/Beu:B;AKWtK~WtW-:A}g
{PL+B=' -PD-L+B=' -e&}{BP'Zt:B;PD+:P;}w
BM0=/Sf-Oa-N*

The program constructs the rectangle from its center, spiraling outwards. This is a rather simple approach so far. There's still room for improvement.

The code is a big mess right now. I'll clean it up when I'm satisfied with my score.

Try it online in the CJam interpreter.

Board

GXVDBDQDEVGPVJWLWSWQJMJMHCFNQDIGC
UIWESMKWBIANYUANRZOLGLXUFIXBEPSOP
WOPZUDGQTSHCCCUSEBHSKMQHJADKKHGFA
ZNQDKVWCMNPLEYFTQHXBVBPBATPQHNTCL
BJRDOWLZNPZQVDKYUTTIHFYILWADGCIKH
BHBKZXZGHZCOUCGCIGFGYSUGXYQIUPJAV
MSNGSBSMLVJSGLRYNSSIALHWWNTDKQCHL
WYWPRVNDSEBQZUYWFZTYBQZIAIAWIELXD
GSPAJAPSDIOHJDGHDBECZIWKSGVUHQMSN
VTHITUCGAWUUCGEBYVHNERPMZPZMUQHVZ
EIOYCVPONTFOOCUZSSQORJKOBGAFUMDKA
WGUAZTEYVXJYGDVKLFOSFMISJSJIVOPZV
IASWRNGTDNISBCBDJJPVTOJLBZRLRJGCS
IVPHDQBWUQHEOGOSUNNQKZFMPYVEVPYXF
WZVMNKATCBBXHDTROXSEHHTHQCMRULYVQ
DUATISIGVTZBCJSVZFBGIPMGYDPYISCPD
IEVNPBSDXCERINHKTYSMIRHVYMWDBLIYN
FCBHGRXLBMETYKDJQUIFKPJKDJGCFCKHA
WCRTUAVQPVUSOEURAFQBXIEVHDFXUDGYL
BQYQHTHLITUQPSNPLTISVYAQACMKQRCXS
PFYDRJMBZOZSBVKGAAMTIKWHJCFBIMEGU
SQYZYGOMRVWEKOVNKBDKEEVCHJVFOYTJV
PROEDILJXAORHMQMPOSCOJALZBETLVXKW
QRAPRQUGECLYFPCYHUMRYPWHMAFZAPQOE
WMKZVJXMFBJNCAFEIKQQAEQEBHYNWROUS
ULXYHOVIEJOHSYKKBJELPIWYXKJBIAZIQ
DWTQIJCHMXEPZMPHEULUBRVSAJQRXEGAV
FNVEPVNOJCBBRPVYBTZPHDSHZFEUAVQGV
PHDMJWJBIHVRLJYCSQJCKMEZRGRJMSQKE
LVMBLOKXMBFAWCGFMCPFOAGQLXCMISLKU
JVNZGEAZYFSBSETXKSWLGTYRTDVAWQFLI
WFWFBMXMSPYZANJNJWNUJLJXMGOPHVZYQ
PXLLANBLXBXBCIIHZTVGYIBENPFFTFDIC

Answer 5

Python 3, score = 1014

Starting from a null sized area we add words letter by letter to the area in a way that the area will always be a rectangular spiral:

  .
32.
41.
567

We keep 10 boards candidates throughout the computation. At every step to every candidate we try to add every word that board has left in every possible legal way keeping the spiral layout. We score each of the resulting new boards based on the total used word length / total area used fraction. We keep the best 10 boards and repeat the step until there are no words left.

To minimize unused space at the end of the spiral we try taking some steps ahead before spiraling to create non-square spirals. E.g.:

98765
.1234
..

The word list is given in stdin. The (quite ugly) code takes about 30 minutes on my slow laptop. (If you want to wait less, you can choose nonspiral_length=13 rather than looping through the possible values.)

The resulting board:

JFOOGCUZSFSQORKIJFYGDSDSEOQZUYWDFZTYBQZ
ISDHUZTIIHKTJNJDBCBOHIYOPBKVVGNAKBDKEEK
EHOJPSLCDRMYVPAWESSIMCLQEGAOKFPJLCAFCKI
BZMFYLPNMNKDKGRAVMCTBBXTWUUCGCBYHNREYPA
RFQTXESYGCDVKLFOVSSFQNYHALSVWESQVEQRIRI
EELQBZNGZLWOZSRJTCTZWDGUIHUUFVRVUICPKPG
UUAHROPSCPGBAVXCPZVHXAIKCOSPBXIFUBTMQHG
SRBXTQNSXYEIPLJIBKKYSHOJEFBEXRIPXFAOQZS
WZNBSVMAHKWZCEHVDHEGDGHIDTIGAVZBLQYSAOZ
OTLVQDLYXMTIVXKBYVRBCJONHVXULMVXGIRLEMP
LFXFBKVGHAFBGEXMSPSPBWFCIDWIMOUCLOTMQFY
YKBPSYJFBAWNTMRRQYNQDIGPALCHXLEDTADHWTC
FDXBEUBGEGMHJWLSMKFXTECBSFYVDSCVXWVGEWD
HPCARTLSSKMVVGUWZBBHMWSGEAZLBJCNKIMKBTY
UYIWKTRIUVZFPNQTYFXJJQWIVNDHMLPAOXCVHGM
SRNTHINYFSKXLFYIHXGKOUOZAEWQDJNIYAWAMDY
MEIPMHFGCEJRASBSXORTSHBSJNJWNUHTYUQHZBJ
XLZHQYSRCKNDGOUNNQEBGMFITLRPUDFGYXFVJLD
XITNMNSUECHSTQGDUZPSKQPHJAPDQKKHRTLCAQC
XAVCPHYOCZVGZXBVAVTNQNWPYOUSPVAVBMQHDLC
XFGQOJIALHWWAINTAGJRVMDGKKLLYZHPOGQWKBJ
XMYEDNOWESMKTBAYUUNZOSYYCIDGCETXQZEPYMA
XUICSCJAVJHPJAIWHMTHTQDZDERPZYQEMLBYVRA
XWBMFGWFBIMXBKWOLJPIVRYRHUGINDRSZBJPQJH
XDENPFTDZGFFVEFYJQASIBLULPMZPEXJMHIEMGT
XIQTYWXGUMJRGRZEMKCJQSCVYJLRVHIBWVJOJQI

The generating code:

import sys

class b:
    def __init__(s,l,ws):
        s.d=[' ']*(l*l)
        s.x,s.y=l//2,l//2
        s.l=l
        s.dir=1j
        s.ws=ws[:]
        # last char pos, remaining word part, used positions
        s.wx,s.wy,s.wp,s.wu=None,None,None,None

    def copy(s):
        #print('copy',s.x,s.y,s.wp)
        c=b(s.l,s.ws)
        c.d=s.d[:]
        c.x,c.y=s.x,s.y
        c.dir=s.dir*1
        c.wx,c.wy=s.wx,s.wy
        c.wp=s.wp[:] if s.wp!=None else None
        c.wu=s.wu[:] if s.wu!=None else None
        return c#copy.deepcopy(s) is very slow

    def score(s):
        placed_chars=allwlen-sum([len(w) for w in s.ws])
        used_cells=0
        for i in range(s.l):
            for j in range(s.l):
                if s.d[i*s.l+j]!=' ':
                    used_cells+=1
        return placed_chars/used_cells

    def get_next_bs(s):
        bl=[]
        for wi in range(len(s.ws)):
            bl+=s.get_b_for_w(wi)
        return bl

    def get_b_for_w(s,wi):
        w=s.ws[wi]        
        bl=[]
        for i in range(1,s.l-1,3):
            for j in range(1,s.l-1,3):
                for reversed in True,False:
                    if abs(i-s.x)+abs(j-s.y)>5:
                        continue
                    bn=s.copy()
                    bn.wx,bn.wy=i,j
                    bn.wp=w if not reversed else w[::-1]
                    del bn.ws[wi]
                    bn.wu=[]
                    bnr=bn.get_bs_for_wp()
                    bl+=bnr        
        # only use the best for a given word
        best_b=max(bl,key=lambda b:b.score())
        return [best_b]

    def get_bs_for_wp(s):
        if len(s.wp)==0:
            return [s]
        bl=[]
        for ir in -1,0,1:
            for jr in -1,0,1:
                i=s.wx+ir
                j=s.wy+jr
                if (i,j) not in s.wu and (s.d[i*s.l+j]==s.wp[0] or (i==s.x and j==s.y)):
                    bn=s.copy()
                    assert bn.d[i*bn.l+j] in (bn.wp[0],' ')
                    #add/owerwrite char
                    bn.d[i*bn.l+j]=bn.wp[0]
                    bn.wp=bn.wp[1:]
                    bn.wu+=[(i,j)]
                    bn.wx,bn.wy=i,j
                    if (i==bn.x and j==bn.y):              
                        spiraling=not (bn.x==bn.l//2 and bn.l//2+nonspiral_length>bn.y>=bn.l//2 )
                        #turn
                        nd=bn.dir*1j
                        if bn.d[int(bn.x+nd.real)*bn.l+int(bn.y+nd.imag)]==' ' and spiraling:
                            bn.dir=nd
                        #move
                        bn.x+=bn.dir.real
                        bn.y+=bn.dir.imag

                    #add bs from new state
                    bl+=bn.get_bs_for_wp()
        return bl        

    def __repr__(s):
        #borders
        x1,x2,y1,y2=s.l,0,s.l,0
        for i in range(s.l):
            for j in range(s.l):
                if s.d[i*s.l+j]!=' ':
                    x1=min(i,x1)
                    x2=max(i,x2)
                    y1=min(j,y1)
                    y2=max(j,y2)
        r=''
        for i in range(x1,x2+1):
            for j in range(y1,y2+1):
                r+=s.d[i*s.l+j] if s.d[i*s.l+j]!=' ' else 'X'
            r+='\n'
        return r

progress_info=False # toggle to print progress info

allws=[w.rstrip() for w in sys.stdin.readlines()]
allws=allws[:]
allwlen=sum([len(w) for w in allws])
max_nonspiral_length=16
best_score=allwlen*2+1 # maxint
best_b=None

for nonspiral_length in range(1,max_nonspiral_length+1,3):

    length=int(allwlen**0.5)+nonspiral_length*2+5 #size with generous padding

    bl=[b(length,allws)]

    for wc in range(len(allws)):
        bln=[]
        for be in bl:
            bln+=be.get_next_bs()

        bln.sort(key=lambda b:b.score(),reverse=True)
        bl=bln[:10]
        if progress_info:
            print(wc,end=' ')
            sys.stdout.flush()
        #print(bl[0].score(),wc)

    real_score=len(repr(bl[0]))-repr(bl[0]).count('\n')
    #print(bl[0])
    if progress_info:
        print()
        print(nonspiral_length,'score =',real_score)

    if real_score<best_score:
        best_b=bl[0]
        best_score=real_score

if progress_info:
    print()
print(best_b)
if progress_info:
    print('score =',best_score)