From 1192345b396c5b26713c1420d6825997cb67cc21 Mon Sep 17 00:00:00 2001
From: Anton <anton.cheshkov@huawei.com>
Date: Thu, 30 Jul 2020 21:11:16 +0300
Subject: [PATCH] white paper update (#581)

---
 wp/how_it_works_diagram_5.jpg        | Bin 0 -> 71972 bytes
 wp/references.bib                    | 421 ++++++++++++++++++++++++--
 wp/sections/appendix.tex             | 425 +++++++++++++++++++++++++++
 wp/sections/available_data.tex       |  36 ---
 wp/sections/conclusion.tex           |  28 ++
 wp/sections/considered.tex           | 126 --------
 wp/sections/empirical_results.tex    |  13 -
 wp/sections/how_aibolit_works.tex    | 235 +++++++++++++++
 wp/sections/implementation.tex       |   0
 wp/sections/introduction.tex         |  91 ++++--
 wp/sections/motivation.tex           |   1 +
 wp/sections/notes.tex                |  39 ---
 wp/sections/pattern_emp_analysis.tex | 124 ++++++++
 wp/sections/related_work.tex         | 158 +++++-----
 wp/sections/risks.tex                |  26 --
 wp/sections/saved.tex                |  63 ----
 wp/sections/selected_task.tex        |  84 ------
 wp/sections/selection_criteria.tex   |  83 ------
 wp/sections/threats_to_validity.tex  |   4 +
 wp/sections/usage_scenarios.tex      |  68 +++++
 wp/wp.tex                            | 146 +++++----
 21 files changed, 1508 insertions(+), 663 deletions(-)
 create mode 100644 wp/how_it_works_diagram_5.jpg
 create mode 100644 wp/sections/appendix.tex
 delete mode 100644 wp/sections/available_data.tex
 create mode 100644 wp/sections/conclusion.tex
 delete mode 100644 wp/sections/considered.tex
 delete mode 100644 wp/sections/empirical_results.tex
 create mode 100644 wp/sections/how_aibolit_works.tex
 delete mode 100644 wp/sections/implementation.tex
 create mode 100644 wp/sections/motivation.tex
 delete mode 100644 wp/sections/notes.tex
 create mode 100644 wp/sections/pattern_emp_analysis.tex
 delete mode 100644 wp/sections/risks.tex
 delete mode 100644 wp/sections/saved.tex
 delete mode 100644 wp/sections/selected_task.tex
 delete mode 100644 wp/sections/selection_criteria.tex
 create mode 100644 wp/sections/threats_to_validity.tex
 create mode 100644 wp/sections/usage_scenarios.tex

diff --git a/wp/how_it_works_diagram_5.jpg b/wp/how_it_works_diagram_5.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..6d82d145d831ee1e7b3179e7cbc34f787d93d4d7
GIT binary patch
literal 71972
zcmd41cT|&I(=QymfS`bM0fB@b5JGP%C6I&`s&u6X2oQSrR_UD(K)OIeFVZ{QD!mh`
zROub1i?qXi&Ux4Kyx;e|=dACKv(}kwt*h;|_xxth-gC`0Ggp&WKLC&5P#6?&;|2h5
z<N5$xO#>7Fw{HGJ|D11K)9t&r|DpSL@7}pfa-Wox<UR=r>4S$4NXaP3NJz*j$tfN_
zdPMn%^Z^w$)g$Wb`$zu_a^s(#w{G9R?)Zp|gzWn5|4+L58$fyg=Hpv3w{EZlZc^U3
zMS0_@4ZwU2^5!kTt!t?NL3i)nzH|TPEs}qltsVgYcW>OfdE*u-1v%-R8+Y#kZrr?e
z`_A2alvLDg_aCzh(a_S-zeMW2N*Lh~7MIX}<KdZ@R6R@_BN3CwxWB{36Dq4}Mu8Af
zZ5>^=xZDa6Ef-ulh*JS-p7q7TDliEDb>W)LlYi#x-oIf0T!Z?D%l&^qs89l~0o=HK
zi;VR4t^2qB!TApeO13+9sf1o4<L*&Ere)WDgK;Sz78ViJ(s}njtKtg{L`>c_{?~mv
z4)bi{!lEwr>(vzC!L^(>DQ{5%o&)~m{)fE(`6@D6ZzcCqOqDn9iC2#B^y*#>3XL+H
zfjL9|IPUP(dmK90>^ds)7Vh&ke6U2la|~Yo8b+CCT!0IX>X?bO7qD}{Dk@f;cKEVf
zME*^UEVT!AW@hL*m`>D-!nH`HEI-YPas?>0#rXGH%+F1BwE69PbDo+RL89N7r)YR{
zh?I_Rcs^X|F4cC<(JPmZ@8`9P#J*Jk(Y>gJXkjhW&%zFE3kz!Y-fvVEv-vV&)9Q3E
zLk+HjU|at?M|P>%%Q3oHhskfdjm=DCG>?}}AhKtCuIx)2yYh6oQxjpEI!)Nyw{+c_
zTtVrP0y3RO2yAA!ifA1X2DN*6MCRmKzN;unbrglbhke}O^ASz$OK>mY?!M49FKM3;
z)v3qKDs+1TEf+FXYT*H~QP3P1p6F9SWEii{@Ko#wS&7;d*?DDhLGik<)8g+Sy<~^_
zG09vMEGGoVdYh`RUqu*}qB4SHQhBq7xImQ&Z)v=l&PTN9taTULR*jt+S5M@TsoR*1
zZiVM=F;=-utM4<S)Z3HA?xa<|UK}u&8VUlpP%w5y&cKp@v2l|5Lxaf0jTxyh;n71b
zP^)EM=o$+z(c0urUcuNV%k(j{4XzQeR(&IVvuV<l$JRx!%2UUBY)O~4N;p@cZAGTH
ztm*7-2!kOad`cLlJopy+4ViBF>XKoxCSJNjM$YbLyiR+YYWR5OgV;p}yj&tIgldzz
zazjACyTiZs%$(&HAxg6N;sPI3v4re=^v!L0EupP)V?@gTXWWD~xUgthIYO?7-zHZu
z&|}t{Tf1i_!jdv$Q?{>>u0??#Gn*@IX0qrjR5@;Px3^MRU_~#H_3hnik7utHfFNdB
zqUA@pLK`8auj{P9Bx1nqP0}arvD|{Chmv5=nz(Asg+wo`_Xs2Y*XMcug*C=fDF3ui
z;aX0|=gJ6%@Aey==rwGROi5SWr^plJuGM7oK23W|180O`ewE{d#{T`wdSm~>#xW)q
z1(0f+P7;Z<V8JR@jd{q_b845{AKH#e*%L}O-h#ffmil1iZDR89QKq(D<NYCexI`BU
z8U)O4oPZb_BjfN<Q20yM^yZ0Kh4PX;>Rdg|sw6D|zF#DI8sGa0JAl2>n^xlmHGe3c
zMhP&0zk)o$kqJLZe$^TvGPK_&m6ge_C|HfSL<X4ANFuU;zg<3$R@=RbPy_o=WF)**
zy6da!>iD|z;}w7f=V{3%P!_jstYuTbYiViZo6}=8E9#f~bg(9bQ8k|ZO_CX$b|ZOP
z5tqE{f9Am|bFQ<L++px>jh2RyC$<*Dt7P23$JoTYK?I2<E;$|~Dpu;~KT=DP0za+!
z;hVUn71-I9OXIRz3+ve{k?e8h>wd(SBWi*QP7jg_tAKDsSte2C-!&e$+l8tFM}2qW
zxfAe)(N_I_a=<CyySM~7-a1#c{bgc|IwpoM6D-=P?5J9wUmDvcP4+4-6X*8|wMGnT
z@ZG6U8FXh^Rhph>_z--q|Kr%h@EaEhKhl&0u20aQU~;c?&Z){Xqjv5}E>{ROrg2=X
zXH<>EK}*j*_0}$dPdG%D6T|XwfvHj0+8-s2iI=+_JeyjYk(L=QhClhU(9ncegEIEI
zy-ymw4_40fCFg28zaDtFUAC(4Z|8oW!q0$3;+9Gtbc*AK8T;2#fA-&-488awAroE!
zcH?8po6g%`&iMeDRZs`Da|Ohmc0en3p^7%^vbB#}Fa${DPpXyzSWQ27;nM5A&#8<|
zbX!B)F@kRWK~3{k$-r*o2Pf>pvVk_z9pzH98r@Ul>7Xw6o2BlL=_PQv<!fbPR=H}z
z>;)Y|Wt_1~TUz~V4JafJr{i{1T+%m-?ddKg0jj3+)6NxXtQC*uSCZteiFZL+P)0aO
z@R#Bjr~NA2N)p=DAlz|IC@M1b;uQJirbHqknE|UZY6<nFJU)}B%a`hhy*K&0ZNXxq
z-W2w_CznR_w#PiKO(Yq=B0FmMXvFyVB_j?>VCe>7^jX|LwE-=(o_^Uw5X8TqhkP~+
z|C2YJ<I0ABFruyGc+*e<)DIiBqB2dsDcA#Pe>!}}#BMM##Dtu2)A>}w5{y`ML$xMg
z$@&Judw)(p(+R<?kxB}stU!cA*jEs6bgI|CMyoJ!%V4TQmqz^x0Bj2FBo#F}a8c{d
z=dUm3O8A;#sv&(&$Q3&Cxv*;dxFymjJ7Z`xWfKR(cU$SdG|ANByJNy`v2K++AH7u&
z*M(d3#-NRyjw3a81PV476;8y4euypBvALKbU|P+Kq707aF?-bM-;%-?DRa!f7bgn~
zKk$62{ivKzGCYPn^{%PcoEW{K5r#YqistSfn5I(gtBwBsQToALCzaUeB(qLSVt+DY
z|Amw%TS{^?)03xZ%2e<Vqc2^w=HO)UC^#;fQyGS8{53^7rW4|=?M|J9+y4}=@leul
z0$bgI$&pRar{K5m+TfEe^m(<Ogt4a}qu`DW0k$M!2@H;k&>VKiWK8W88B>FuS63<;
zP4U?CF-)O?6=ACtS<B9WjAPh_TEqZ1&S}<<!(Uk+c@{6%t@3LkO|0^?E4<+Fp>C<Z
zkHuB?k$X&FDm{C&u{XXJ{JYy7!dTk<EQ+sde_PmE5Tl{Dh1?tcA?(tVzXPt(${%bG
zCq{i>N0*W<2Z3Z2DSx(%_-R*1!JXZgNAtA6h3G8`IhMy2tAwu=OKn-1eDucGy5bj)
zE{!iV!qgfeO2B;bUAP3`2?xg^1AUa8<0m3Y@1AQu1DSBf2Q~V6ZAzongDnWKX2fc;
z9&86Ea|OobM?sb96XjyNYm;)lrjtGzdEu%is)mVGFrYm8PJ0!I2g2F=UdwXsEGK|#
zmXo)PDZyK$gvDd3mPb*FxBD|oQ~v(r5`D_^7NJA!r-L}hZUYy%PQcqKdUQYQ-7+c3
zK{9)uyzZqG^z-`a(Vozz;v_vLPxrW`I02U@@*aY6FZpK+(>bMoDu1_zE!Pc54DYjp
z=x8$9vN<MC*k>Q!t=Zr(pYCT{1+x70pIxB;gHMmv@K!pfU<;Qrrq4=EohdrEmEe>t
z!bXN+qeUCF0$^-CLtIK<XoSloN$Aj`gq+a(aw{gfl6|qA?d6$1fx?;1zSvUr`Zb1~
zuCqNb)^y>~Jm5r-o?ROo5$^L0aO>-}zY3sm>pQo2@wI|K_i;a8H{N1Wt;w+QJS?Gd
zIgaEq^A1+cMEOBOhKp9bND}klARBk@Ic(VcaY~LS7jrsWvtR%p^`nqS%KY<Ic`DQM
z!{t(nzGod=Ub1Q;Ygd4nZ~@k@rRSL(D;A8U8IK#>PLab$xyG+0#n##>%yJFui`w4?
zFMC57BH(x<?CU2kj?s`kCh>>waS07}hW*w`tyJo(D1k+jG=nwz`LJ0&T}Og(YvdZ1
ze>&mCk^>GJ$5P@bW-F|$-57|F9reGQctiIA=DIn!lIB5e+EqeZHwA0&mh*0VJT(&k
zZKKngBJhUt>Qqv~gjdMlzYKJ|!AdlN^>Iss=Bl6<4X#}}dxY^pTp+bEty1kxQBOIg
z&)?VPo;IUxc2S8mxu<tsV_%jzj*k^4aby+#?nM%AFGbg3H;cjw!2ZJVYjkh!d8c+r
z>A}cs^poz9YJ|%ek5ly;wjh$sax#J%ZncxH29{X}C@Po#QE@GGU%GRytW)Avk)u$l
zidA!8mO&ShUIDmNVPe8Xl&?o%+R?x+Y<FE=LQP*5w_z<Pjzo!M_%F$Tt<DTGiaSar
zv|PO7{E?=O;97X;7Mn^*gqWm?ounwcJo%{!tIX8PV^Eb}4Te*=+NSMp@#rQr@hkB{
zB-ua0Dn7$VIHe-+tJ6nakpdWs=dBG#%HzNWJAD?#?z|HtdS?DxsgCZ`I^{ctwfQjw
z1Db>GFDNJDjzxHQ;^vG_>f>I+l{$QdBZDY2$xHrzKh+^)&Bb`m@;(04K)v#u29$q3
zBzNgoMc9M3I7U)7eO34XPhGtL*ZRQW%;GR*Mt|`r>cIPk2O!&cPni{5T6|11lJ-F&
z(?V4rYal;as%ES!wRN9|3|qD5t)UznScBEORNvcG#O=p)jH<42{o2SrL{^}F9mpDn
z=-%1tkg`12Nuy4d)A9xmsUOmNb07GE+ZgCx-$MYh{9qGy0nZT{Sk!U#y!qaU&C3|c
z8!7A5Om2iJ2u(My3xYzz@o?GWpl)|cwYQ=6&~7KOegjG#)2N%TS(4pemmlKfZgV52
z>a%1L^5c3|uI7xjrm9R8Q<_fBfd^do@Ws6sQBaf6CGS;Hi;$JWy&Y7^M;%lgkjqVK
zsu>ut!P)>d>FSD%j129V(Pu0#`F$DjUGsT{yblGR+tQwuWp33eNHws=9wi)lEW$R&
zD^F`2k#7r&bKwA}VvNmeDYBxXb)H)us-%TgPfm7aK0W0g^~;Q{ZfIHcbwO?$Q939p
z8oi+=_ANT%e^h(-^}%}?)p~6>Mcp&0KF+v8Hh7k~T+nvV9udb<nb63SnebNtKyH&x
zV%8{1PAQ+!yHL+7vQIvIVDx*K8L74hh1ejns&NG@fr6`53O*xRBV<n|YkFnn#$CRc
z59xdTmC?wc2TPJ>OIN<3{puZ&Lt-x#U8M|S3e(8gQui*E+{KbVHG09P@h~|WnTq<e
z?UD?hlq8+&DP-<NgYdJ8Q9K<6sUsRGA28fMJ1zOUW?q8G;0VwBlz9Ez&Zy^}<SSph
zoEDchaY1zV9lJq<pgQFlVqREHqYpU|xs64}QUhp|s{4cMH0@Xz0%uLlwJ}AOWBQh{
z>GD1`55)R{cq&Bl6NxNxAM%k5f*-n0j^k_xK4s)@5Lj%_E!$JU3FTG{OnXtUzlB>H
z@j(7qW*yJI<c{oC#0o<=H~?xdY%7hwF@F)NuNTmW9V31X@0nQ9(~4t$j|BFGj#sju
zJgm9=5%}B2LhvnguPINaPtfV7OW}HLW%O8$;XW7s+0+C>;uNYSRRx*Rtq-DIYlta-
zMEmi7{h7FxhOVwP-ma#T?RaJ<Ln-<SJuQ;&@x42U?MR^E$v1;|Qxv7Q8PyeFxV3?$
zdye}GFe_znz7w-bZ#}*n)F@4{>%k<y`>g1RH5j(UBiaLYK_ygqxQ}`9;udfh%F2MX
zTch)jGgkBxt^kr2ry*H}af?lZE)fK!<A|{v)@8H4t{WEbcMzkQR{(riuf<%8^B3aZ
zlFz~8^E9ubqy)@LiJinE(%(<+7t0FgLg-_NrXRdwMs|H$-q|@Vj$D|=T>%6J4<0mn
zdM4TYRLg1pUb*-_J|X}Ye<Gq~Jgrlmc}F*-uNI)=`@8B2ppSpsTY3JaHA3UaNBO+*
zd7ogrq$42zWDj*&rg8;HvN`?mmNs+qFy_?b?Z+Gcx3vVm)jW9q_>cQv%MI@=Zovs{
z9d?|?a^D1A#YfjNCrIIeeKpILAK2$>RspR53q6<%+5r}gg*knmE(HF$KZ=D+?|3pi
z{a>#K0K4w*aTC&`GtUZmG7oN#DIK09)g29=NtQp)3BMmv&iYUz&C;gjY#OoIU?rBy
zAj{zvIkL*|vo>!?L5)fKmn1g*12sqZd9e6q)%TRds2Gcx&{}ZJ=W>&u&~Kkh_<W0c
zG}F90blFn|ig=nzSYdLs)teyhmsJ-JaDR1vkOJd$RD;)$2|wG77@W$t1L9%@AM+6#
z2#kk70ts+LgD0;LRl;Z=59|>j??8|P0Mr`S5=l}#4{_Mi6}_itC+yJTPMbb!HH;s~
zky1QpMCb46sf!B1L2Q-UmLGn^|55Vfm`AaE{{5myRG?GZ@q^Kdg%0EUd@R^5JgR`R
zUTA2Fs<gZ==&&Em<<T3K{71*CK_mBt&Tn1Ck?4<yg34d$-Y$7!(rnZSU$KZ_FzQ=r
zk^aZLq|yiz3YY#Su4V6FOo}0-*lEmMDrA|bD?f)QQcuMc`PBd@4z`y|V%Kox?CXuY
z$l)(R{@N)@?#M;1{cdSd>R2+j!@Roi*%MoRLhZq$>>FR`FKY@D%}mP_PjZ4TeQMW;
z^U4{PF}#%Hb4sEDuQZ$qE#PdVV!+s0qWT8A9MC9vm%Fs4ojC(nYS+OigK;j#SFa6{
z*smWxra4K}dm>y{M860Y08zks5YXy2y)yl;)n4o==?f=@2YtsOq6K>oDH0^UWgxX_
zUioVIWymj{O7Aib$a`Z?C9+|O-D~AO_dB4lg%Qn!e49ckr&1Ro#B}(_g!<>aVr$T9
zV;6Quc?P2}1O+8yS4&!G!nBigQn1C6r~GGI&~vuhBytLdtW0is#?7KR4bbjeCv(|Q
zbvJuM+v?n0rSa!?I?RO=pP<I15Qr{!PJ98UqDhV56#)6tzFAsszeI|5EQgCff5A$|
zZX6R5X``n(^)sC;St|#vC%mXG)J7eF=&_fP5u_hra*{gW7XcCj32;#cO3LK3RtKFj
zTlVx(1BR2*Z1%U}>Q>7c3hUEPl*Hp*H{YI0o&P>-lDOP{#Sq{W{=<E*M0h&Uelk@|
z*1^=$&Nxh?ICcb_1++69&DqAqWm-Ic$q@CUK0aEP@JpGCB1&_PBk_$3i@K<WlE<pX
zdvVN)Fy8`%q`v&Ya!r1OHoq8j3U^?HkBL5y^1NG!l5KfiEIfa+duz{khN9m?lmcEL
zIV9%7H2lPBvdcY)kvy4#%1bJm6DGH8#Ej!Gpdh@KNUh$sOh#L|CRU~v2Zqc8Q}+{n
z+3-mXIl_;Rc6WACbaS)k`Kv0Uy#{5dd58rDi)lsyLD=W8T)JYbT7wn@?n}D;+i<<5
zN>Up0#m^c_E*9ajQ?v(#a6MNs&VIPa;bppUqyF!cV80XfW~BszVJB6JtCzbTvtQ#A
zhrCM3hGib1T&M$HEWd(EL*<Wq%W!<y@>CR4`o>mx#KlkQQ{qEo9kZWanFFz(E}vHf
zjA>i}ngW2P<VKrv9Y~CyfaPykyAxE@+bhRu=<m+bjAZC6E;$OCz^Q=~CP)NN##Rp`
z)fdz<3DjCQjqdDrA7ZZn>LMQy31PJmzUE=87w*CApMx{%B4?E3KL(0b?;*mY4@%i&
z1_$M-Tm9QRpG$n(qN6D|1^H5_xlKk+&BMSaFjBMITtos!eyrT{J7>Uqp$${tDEu*x
z0N+l$@)cm0d-tT(b6aRZyecQdpn&<Uu%&~eO);va!~Tex42*0m+Dt8XDd(=uZ2~nu
z9&OH2H@AVQhyscB>&_6ljg^+pTIv}4<D^ZB;-w*>G;x_v9odjK6g}p}<rBp--jmqL
zgeR$DtIC`s{ZWZfgeb@)>aI~*>-Sqvt^meYfCH#Zgj~lJV9Mmuww-bQVmt|)rF`CX
zuF@J}a$4^Jk!Fa&w&#SgKC3CItdoC0K$QMfhfLR7E6)mH@m2xXfTPAdIb_eaHD&FG
z@Bi6Maj+0<b*C4xN#b3M7BrO}-^D&g66;l)J}7`o6gO)B?nj_=mKEbch~|K$w>#=@
zMc!<kFkJ!cTu*Z-z0Mlu&x2SGIQCe-?VisxGH7m!$u3fp{#qeZ7g-Xn>^h>ESQyb=
zQPck%A3kLQ!!=T2m+<&_a0-h%iQ#CTBwIG*!0975Ziah3ZP6H;7^?vWmGk)0aR^qo
zEJDzQUZUh~5fP0YpI(N4z?=#P1dqT3-Qjv|in4~j#Zsuk;U}4S$$orYJ2JVn%N9ti
zF%TQYRZr&vqPzmomsQL~g;kk?1vUnCnv&Z$PX+wu<yXC>$?9H?CP;FZIz430RWE#V
zufPYDfckFmCw(g=!$c#|&1X(y0;P5&;i4fYQ@hkvJB{uoiZG7kNzP?2V1%A<J%1NT
z+2yc3AxzllpHzgaCeSGM<?o6G6@}#wEie!`Gv#Bv@1hG0QMoP3+hO_iyF!~G+TF7H
zmDa2{x8DZ~V@Q{!ex{xE<@5`n49Z<!)kVhI(LQFAQ`{<MoTp8#I?VQ4IR{S_$exP@
z9(8{)2&7z}*p?01Hb2)Q7tiT0?{h0Si0HE(P_!YbQ;yM;M$*bCW|1h!-UGp|gUV$R
ziP%LHI}8KSA#2B1!-tFdcq^&9o%SDO^Zw9;C>z$!#8@@>rNHs%;b7G!bd!cumGcH8
z)*2s2mdpz$B4E34aM#5?qu&V6yAN+pk_knE&7{E%O1R)jc5%HCa6voO{l3sq%lK(m
z_W9YTcl=H6JsncJVJ)gBB>t=?TY0mUf<UTOBCz^=Kw-L}d5E~uvQ%U^vHEm02eJrk
zfaLWMcOW@mmtl$p!v@9?Jlykv5%*@a_vc`&M1oV@X16pY5T|%e3YM1`eO3+q0<sw3
zR%)Jfjx9E$Rxx>VZ`!{evr}{Cw0o8#PT9?7{i;OH#BiivvGK3=5BI%7(-4j|2pD&0
zUzj2sd<+3!i~uaw=mVtFEvWbUGM^MT1%Z;ty0*>5*%J7NETOi}QPL-JXTI4F7XqdR
ztV7M+49MkP8u&(zDdyB!Z<6)NxLSM;l){6*PAOPtLlFH1DtZ?<Yp9Yv;5zX)^(Q4L
z2wq<qLxN*dFX&8!!8Nh7aLDtQls>Ac@@Dms^RJ0w%8^zH$GY*7hoEO`K*@INAcwao
z7&73L4BavHy#|YO6kxkY8^4&Q$7BFoS)UpcNGOmK8(U+M_)<L)@?P_n?p7Z26~MRS
zQuAC-KPro}dZrz&Wt5z9MoDVMaLhW*^EUZm=r6<TZ{9J);g<3(i8k=l&dTDhABYez
z{yFopDX3=4j$NdLSD*(L7Z;ag!NGC`Fp>CqHYdIxQs!L)G-Y1lzXF6k4Ey=sSa91v
za_X-!pN-D9b9=2gmIloHNt4scLm8XNw`jtpr5slVnc1~{M!&FYc3W!w={4W;@PCf=
zp~FAqc*<;5VNr&*b@ZQM-c8uIL>sMkkNA`*=Oo8phdy*2;Goui60O8|_Cp>Xcx2aE
z;#UhaTK}cK!2W)RsK47?I!@zy0YdI8)NMPX2*b!Q9(E20L^d?kWPO}~xub={W&hFc
z*-;d=)%`>)qN6<;vrunV{rpJP!c;_Atik5Vx9YbKnrdPZVy`Dmp|NLyfhh^?Rs5sA
zAWoZd4|d4O-JkTMPacZJ$q6u}Am^Uz>~*Hix_{83p&>#z6as@7vW(&qJIjx$VTym<
zji>C#`0ZA2IcP1U-rvi8ZuWEtnotf(NMW{W;|^0X)PX`1C|Pr6gqc1AZkt-?Z5%#0
zrcFZh+NbvNTRiC+ySuw@{5s5LbkQ{Mf_z}D=A7my$&3HtNK0B!|0&+en_V{B-I4eh
z>cMqnFrm@d)jB(cSt_h=aAu^}`?>7)m=bRQp-@X-yQJW=hYw==W^F_Jh`7Wp_F1Kk
zLkfo7dbD)&Swoh_3Zj`X*so9|SZ+BeCWGT=<Jvf>u)bOMO_aSC9O8T_^qjSSNl{Vj
zo$HHDrm{G;i6;uh1<(V+r8*2fp_5Qx%>12OiwM(+Q4^Ma(5Wt@+VDMpXP^oH!3so4
z#8)MmzM<fcg+qj4Jj&$^$vZWx0dXa}HE3TYE^QVB>}aJFl=IEdolu%KBH&!au*A$+
zWN-yQJ-8TN%1q3nD0`Asd<Ccq4p`qC?9~j>k8ZdE*bHv?%r6ZdYCqdqJRS8<`1u;3
zAT+DnSz%jU`Bb4CO`9w_z}+n~b$>*ldZVwU+M)70fw%hLW#Xo=wl6_CKfala1agDy
zi@RvrZrQ{GRdJ3|d^|~`R3i^s8PoVGLPJFA8)nXZdH-dW;&gx20=6scNf=B$JJalQ
z!HIzhc`w!8hL^ijkr;ZiE4bFMY|);F0>W@&8Xf<Vs+;Q3NO1i2hwqB9{GVygWz9<^
zJ)A3^NON|Yl$pyUhJOjBbv2AVEr0S|)iYh6i&+3T)lET};ti6HonMrEFN}km>)s)d
z<8QrGEvj695=`QoYA34fKCY7x#}Jp*j=r|Rl=#15|9Yo3){x&5>+Y_XOj5Xa9zZSn
zprRUvYqOF#WovIzOJQ=a<pA4>P-|q_y2jk$FOMq`+D7g5%2;UZN4|%uQK<eNV&lE#
zSj!?ch3?c@4pV!!<d<RpA+lkS8BScXGqOfJ<9|&<0V?(M2IrchH$3zjHc_9Sj*86s
zsv%OAj>_|R1liVx@{52yFh<L)K#nkwI#a|3hbxES>+IZcod--@6}2s?Qn%b5jgV_E
z2{5KB=CQ6k{ipzLAR1`5ZPbp&G0U`s#{*t{;u{%P(f;%zDg5;*6n@^bK3=L=V#jM7
zL9=gAQ~M<LOf@<4)&Ud79?8JNqoomh4y|1rP<P7HcXmURo5y)mRuJu{tK1KviX6)e
za*n5nu;?5FG<Y6VRn680=DGRphS|717~HmA{CCICBs0g_$*gK^%rMp7af<thBU{0u
zVHo<^R#a0JThb$};S(*L_~+2SHpO5?hOTtTvhuZpg=K-lo_9$ft@4Zs{yZLJ(bQUa
z%DP-*WeCKf+p=PV08HQ~d!lE`Emu?W>MXSB^6RfkC9A~`TXthHbrFNxf&i{j)^*;W
zzoBop*)pDSKOfSa{}p)!m{gGY-~1WBaF;$me0V|ec3(cZ!U_pt)Hidy@!7FqR76>s
z<UYjml<@*s<)GX$;{j2|yEO-iS`}pJQ9jzMt>;R-sAUZGaeA*Wvz>CT*8myJ*sbdY
z1_`!YVv<}hg$AZqW_<0PSbmwHeSHp>NDF|3_RQGqgU}|If#jYMeAo7IJCm;$5@)Jk
zRL0eK@1nTv;^+!6XW=LNY@y>Fvyi!nnn=`deDj_9UD<D5yT=0s<;O~;gT*5Wyv`r&
z<1yiC#fP@9O6>~A@{lg*mT>^Ul;r7u7x!LQvvb!E^Glu|4aayeC6}=w`#^=xIH?FQ
zSGYfCshp?9ZZW{6F*)(u)~AB~hoo{$6^-@YoEq76=_U9zsS2yU67-shehYy+a&Ycq
zA<~KS@5DLH_gEwxwRY+CtJwC`b=_O#GrBqA{7r|jS9p|4Gu=sNkSOj_qi5Y4LG&vg
zlhKxF81Li$VTVwyww%owT`hd|Gq;+=DkF<dV9@=3?|?q>EF<pF{)>an!jNfu!qZIe
zOm=i<ScO9xlFiZ%m}I%@e=N{mSMz<Cnb};GA&j_f;9~!%W<hQTK_SV&%MPNp*HRsV
z!eA^NMsbf4^~ZEaA<IX)!*8lL7MO80nWw<eS2digv)fyZh?Jw(WA}R`BB{{bg9P9y
z%yq!fDb=B3yLqCqs;;8M6fc2Z;9h(de%)-0&*b4>lf=5Lan33yH#BNI{^J^B#6M8l
z%Fx{UzNe9`7djilAmI7Os+L9PJ-fiN2KU9Vw%6xTHH{HwC8@rL$R2V%Z+M?{t)+at
zkOY&h*O2$!on%68;#A_qJt>vVD)oJltepl3vTC_9ue|Y<IFUyHW8_itT0Oegks&WU
zH2@N3;N34n91o&oU|^-8jKc?y>)m?apWrQK&7gQ5nIg<)(y-&pYm(;nbSM~~KhoF*
zEPw&aT-cF`)AW<UE@<8#*%cKadlyTXq8`&5K^3=I_pfl3da%tbT((w})m667xi)+z
zuK2mOb3~1(ypmCw)mZErQNaFwxceq{;o3$Q$n*}k%`C&%!OgOr?)v)rBNH731s_-*
z)2n97*bYx2(85G2nEa!_I4Z4w+(9GCM}-Fu$y0nYj7tv$+A*mG*3iJ`R;qnTn)4jU
zr-=&;dmPA&xv0L2y~4)aZE#kNN!m7g)=F)35%vYejc@%(+#c=rw6QgE*xWC0`IWNS
z-Sd4m!y@DNI~&S2QS*TqI=$Q8w+eH$`6R8iaDhzfU_SVRzA1>I%85L=To6O}3C~!%
zO8K{64=UD>bX4ViZy-BMM3XPYbw02LOBWw4O<V7lRQPokTXkY^crvD`=={0)m@vb2
z@)$w>`xj|JlEt{#)N~+x8Hq(rc4rIEzw>T9nS71-^(7B7U1I0K715<6?4=Kyg3@Y^
zw1Sg41hW#Cl`A}U71X~w3hWz6X%<v?<){IxvyztkVYO|kE`9FkrI6td0(ni-!cTab
z2<wM3hi)5_E}WI3CW)4Ye3`H@#CbsB#=Xl|$&?G(LcF(1aoE)$P_6N*iry+}vMaQ9
zSLSHqxg*v@fbC^rJum2G`>^BaxS;*kWZIXMSS{w*ExY|BJ#<_=ngFVWa2$k&Sk2o)
zpX0_7IMt|f=jw0?IM~>_e+KN2_K)U~df)WMhpL(6QbtaUHDXF(!;`hD-yw$6)k348
zYKd(y%f$GNq?C;A$tL1Spf9BU@?f;UiL4}J6vcy2TmQ9C4;SRw6RWE;97bhSy<}ed
z{#nz-dtu<?Q43)lv+Wn5<1S?#GuAfR_{4L9lZ|uySjqRn5r~l_>Ni!aE!x_CJ6j{Q
ze%tiN(BEU*BkUs=?1XN2-9ES+l6N5@GB%dTig^Sqs4?Ku%OICn)H13INZ=asP0?u1
zTGz|am%Oq0=ha?XVPw9uPM@_V#-hkkO~f)r=w7=tL7L9|B=|`gg`tKxs}ognm~Jby
zH}}lFRKHV4N5W1k%5pg89rq}K&%bsS8n6+OCY_8v`gYF>5iE>0*V(U}uO1haSjr_P
z>Ys~L4ZU-+?jM$D;yThlpKY3u%L%X6*WBMMm^95;O36M<k#u+El8xt_yW=~a&h-Z^
zoa|VY>l2kkZSbkqwPiWI&wRp0a;J(}?yI}FVa}3+dKg;HQB7NpUA-GQzahjx<sk-x
z!A}Lvk(uN%76&0t-@ROjz9}kR{VgDUsqDs>xQVW^=w;h8N_+#l;4aPqYH392@$B=j
z75wq2kdVEqda0GjEqJ#6b2H|yJ_gD-Mn)}ip2^6r8uo2#*}R!v$FD(Ba(6U%y6deo
zX$w&>D+@WTeayGzVhhU7^10@Z**RbLE6Yd+)mt+K41HoUeM+RYE3W{g5Mh!(xW2-a
z2!(?9uLMFK$D2kmhA947WO9xbSMo0i{CM(WggIxSlcsy`?qp%F>s=Rvj#^`$5}k5m
z#o<q3;KSqtUxxrs{K=WTx467B!|#0|Nl_tF0-C%kd@WnJu0X*WIt9zoxCu<rLV+6@
z5l%1Whu=`<uVwpcr#TPB!+=AICcSTvmEvSR=;H)zdWnQ4!DX|S4xv(eH&IKgXCV>O
z?_7!u*O5keg9Z85oHSJTM3;{UQJBlBO+`q%zaL2kle_GOvqj4iy?=dG9wR)ea$i{*
z-`D6vHvIGc&+w?{U%@is)*dxFkH#-3I9$c*BBQ_%ko)e_xs5pN;{~n?y=+k~*F~%P
zh_B;?rN<Q};8_PF($noynS5+AqPT_F)r~G8<$z-?t)c&r$oJBejdA>VXYQyf8FLTc
zqLfBwgzKxeaYJV;3PhkFC?;Sa;I7)qz7SQ$^I6>jzLIcTBzFV%;q@<GJ=@(q>Z2Q@
znf206+tDOY)(;ClzN%@SH``E{2E0U`_TB7TnoHQR!ye8lxOV+nUM*&0R24C?djtFW
z(rnqZ&dzG=R70a+XsassbZ@2}-RoZqK0b`SNp5vi06Y83Z25qS{NA6?jVr)RyV<c`
zheO)K9udN<@R3H0)?e%Al{qcbo$E(lPUn7}XGUi^Z{97jR5(YzrP!b6KO(IFJ6#7v
z+#A547C%op>AiXMAo_w65a^Xu-?PUIDj$`Rv)}3YY6<P;grfIRkS%kmB(6@wDBeG|
zK^)Q3{*O{m#5rsUyRfub8>tUgau`kZI^IafTa6%`!uxo1Hw+fXpLm#{!XfG*B0NYn
zJ!qV&y_YuftV1{Ejb!7ioDpWuyjMPH+N|7maO|c9HW4SXw`*)-Xbks|oz`2a>Adr`
zyjkK~!Dkuy`bVhzH`R9^yCrAdjZKemmD@!AQB$r@{)hhJGoy~w>u!UkI?U;T9uxg`
zEZ}z1k*Dk5flCEi_i?_!6M?gBg~$z23y#bmSw%fE=gr;<vEPIV!Uy*8bVZKn8po$C
zN%5xdg~20rf4DyWptsJ__lZeeau$5b(ve@8!eyY>0X2>-6FC;<0QZ*1Mof4)4cUc4
z_;}q?>4PY~s1_P~9rf4D1c{WYF@?b>`G;~K>Sx<F^$sGz2cD2ACMWxesb_98R)$fp
z3dDQT9)lR(eoN|^P_4$hjU|_M&hbf;Nf?a>rsV(+5l`k)nzk~#7N@1r#%)V>rDUx7
zr0xVOUH(Z=@!lBn8jz|{XxxawQ*6&;mG5|#<se4iKdNt*BzZRFf;O$6uDXBpsf=!S
zv7E@P43b2lF}Nz>nY#52nV#-y7XctX1()7yE;L6g#X}HVGcPix%0*3A&G_Dm{UW)7
z7PCfY%^4k0#Ur=jGZC)SMA^E_n2iiBW1ZhMOePvPc(r=?@1Cb;BTpbdq9R;Rd>>#=
zIHsGIlpG89w6R!6sg7q_E6Viok(m^<PbxD|Gk(tH3zx=q6`9<nFq^!*hZ?Xi^iT!F
zyWhs@3bza@n8arY+2*QVm02o2agK-6`yM1PxP?={+JW1sEs&})<w3NpZV}@LGIh~H
zeurJZmBn9tR$~J$bE5HkGkyu9><P?OY)Y$ish$11rs<tSuUJ^z<O*pL+^v%=;@FMx
zahBg6X->Op8EtC-L&LrvBpgm(2%Rl;+zU}7D=JObFd)1dRiU+2&^9l>QS%)en4F@u
zct12z(eAoTie);pg^yz<$qu8Jt_yWZSLqj<)6&Z0tRjptDChASE*bY3EmT%=@w%X^
zj*wfXnF?m!Z6mH|{o(C0!L0HD>s4WkjhuQ^shbBj*ddOC&<L)=Vd>eC`9r=Jg7}aI
zxvZ&$KXmb%*5Hx~cUE?ngyiTBtsEYjpHhq3E99LDejgPSH%D`4I5;?u<$%hy9<=cr
z!^VG>__gQ?JE^;58pBSN<6YK<H=TO(dIy$Trcb6ZrJsh>7~A1?7=Q8Xd_Ex7`tkiR
zi=J!w)t#r4y?YLSZanuxwk(~Q<JmVm=3_)=KJPK)F9~_&e9TQuqaScLWH*aCCKuq<
z8b~Hec&)GH&G=pBj4{7yD?cCM3`-Hh{T&ifXyG8~dvjdJm%|_L4TnQuVoQWt+Df#d
zrOET$pY@1m5e^Md=iSsFrfhRPjCE~e)+>F%i<70fzD_mowWy+E)#{RH4!I;5oaf`B
zM?2lfph6FMkMeS>L|TmBJL(tw8euKP&YPRv6i_tEv^eSW#iWxH`FVsXLr_PeR8}dv
zPm3%bo4bv)Qk!Ww8aetp{rs5jITK?)7G-)sX2sQ;!4lU=<wn?NM88=yXsco`XrZOj
zTEOpL0ajwxidIe!)53l@-5KDkY}^i?M{<0B;l$SxB}}3TedDqQ8e?-MKRn$KW%RuE
zlN5dF%-QBNg!_UM+k#`a)Pck*Fc6;~Lxd`Cu(5U?ZjrRJV9uSG!whTj(%^Yc*rzoA
zZxGBFrH9t2wY8nxs!e?u{nT09=bGrRHDwy821o7u<;JP6$c0F)07cOrpM{MmpK{qB
zQ?XMgrj%-LEEw?(f5N$52>rQZHT5Hehp(Tdb?$XHGOiY;bJT|Sk;8TMMK<57kI_q7
zlBiKG)MX6H#f=@-34o&D9X#nF%=|RNGKcO=QIwv>V%ju7VqXnL9c(+>!e=rxClpmN
zR`g)$DBtmDtl2iO{pI7e$5TdH^Wc{79Q6H{DYX!|mg#$TN`C!s<L`dd4p*dcDSqY;
z*kpaqRbJG|){Y$%VWj&qJ`Li+KE`Ks&fqNlBdrOM(LXkiAZ>Gmp)sj|-2P9Wnm_hv
zxQSbDt|}Xz`ST{R08P5ilG!)~Xc)sdQ5qcV+eSO}p#kZrsb{*&V%ma06XO-yfvL$s
zf4rDYgVr=Z8@HfnZi+k;j3N10lp#64PxZx<qXq1@rwyaCn1()#n}z4#a&;7}P_a#Q
zy(&BuPdv}M4KouK<uhkrS{EA1uEG=3Lg>a-W030*Yj}jvzW#3hsvdRe<NUp*^jh<7
zM5q`R56ZB`Q0$^Jsg*fsuZvs$q5l*Pj5XGRxa`=M$<z$g*x5YADy<c88+v6<9SmN|
zGb_g}jY+kZB1@d?vl@ywd5JS&Drjh?QT3gAs>m30SMS>@bL7_agWo2ID?qFLVF<bB
zYl3s__&$NVb$g6!^;6`5g~aa%)cyg-j=`Da9XSghgIHK86tMK!B~gYW1C&%V3Sdx$
zR?SX5R%P(b(!URt0(k%%Jn&YNZ#^8ay#ibT-jwi6&A|mCdAf5)zDBH8M#ms_1ROOr
z!m5<rGmm2jXIUgq?H+CJb^htaE=9lYFJZW29>zc+MPN{MDyQj#Kn~9Wp*F`6m{FE4
zuY2Dj->!%~SMgwVMi?HaR_cjA!Z=o?^w{&gPZ?-tl>0_LN7CU5YhUX&JyyAcQpn-$
zblfs~_feG>UW}dRF)v3NiV9d^bL-gz<N`1?>XTiDR9ILXHA<8_^b~#(`|Du{{S~0c
zAt2V5?FTD`&=mkTdBN6*prkDre)TNZv_IJmda>fxa9ut#p<IGHRS^~ks$g(fGb|J<
z()b0K03HD*$wu5Md1x)Fw9VtFH{D*|$+@&+vA2+oEm!S=XorbH;d3bU{NE*QDRp^P
zY`3$KrB2%+fB8vtUTRw$e)#?n@Ea3zp4+(r;Im5?=yqLm;Sz-^)yBRWlg|(BkF+Bu
zCUJDR$XoGH_OJKK*;yW&?v$t%@J0tx##fd|jf9uTgh<xif1k9Si6D~Raec<Y6%OM;
zIEWZ=l{2K^?2rY@4RzQ@*I_A3nOh4T&-YVV@&7ilH?r|W__mKMLX=LMUe~}_RrHs9
z)R{e3)F~|~h%!#FaJ!mXsS>op%l?vz3Apa^@S#ZJX!Y+VvIOOvvaFDp&azZW>eRqK
zOR3;x1#u0WH#8)Q*Vmm#afEgAd}F?X<;Jfi)>E|@;=_Nj2ZWm0`tem_wD>>u3F@r9
zg0M`b7wj?dK&KL+J$+&7Aav$EPY$>-Mw*pF=qzg8Du7vms*J8XSASAx$bhJ0P)o|I
zf5$abHDm=8So3kydDc$qcdCi()Ff7sfF;cRYuU*Dn}To28)}Aca7RU0R@!bDmj&r_
zKGuSGBh9hcX*5jKone+#gvY8em1W(KuA)~>e_`-eZhu^&zOGJ9j{srW^6Msp3v9Am
zyR@Mip|@Lb_iR@V$B=pF7m!gikPY(oU()^i<WfastF^r|Q)Y!!S?4j8#hXR6TjgRK
z-n&TUh;7JNI!bIk0ILaCisZ*uACZsoef-SBV~1(5rQf;Z;Pa=Uu0%eF0%AEWG>N($
zJS~r?(8A$ELmMe6$!-3>=QWnJxBcGuSP?6RE_2e9YW`oK_?)O2<|qZfMHecCB_o^8
zCQYW%HM0@ChNI4=>@Z3`SBPAjBA_P!-$IzLH5r)6QE|J<l$mGuZT|Z2Ch0$&MD<3t
zjnXL!3xs+U@pb!Yeo-)VZvX9^MJP09wk?i!WdneIf%1zNz5=|}_QU@C9lNjEdvT$*
zk7!+d!Sso-F!`85Jz3Z8<LbFi&A!R&C>~56xn=yKwRQY#HD%)3I+N!A;P<lHNt+B|
zYd)L)yiiV?JIC$Mg@7~xzS2tLbl4KrJPdDK!f@A_7c@&~5f@U7eaGb4<?9%`rQK>F
zldTah)Ecy^O_yY^5)F-l^}=Ma@sKNkNO=cIn~N>+>0yTdsK{A#U1eA~el)?+X1-;$
zq-FdHV2eT@&h?j9c(fyo;4{yVY66dsP+Li~r-wE_P#3#%=|vSa%?NwD|Jj7`Sqp0M
zXZX$f(k)(_Oef#UB;k3%oErDiHC9t25bsLs<DJ8~sOB${k3#~-O)YL$BKpCV(n<_e
zSgY$gIPsqmO%5gF&2>}KEOVP1tN5J6Uy1epWcMeH%Dv5O9m46vH6S>bIjRZ4tTzAv
z-TwK1Xo9gXo+GcV0B=c0jFzEqQkx*D7K{N=A^KqqnXZD)$nVNA8V@A=zKhoSyjSfF
ztugXWP;}>or^9jcBfv%m;NaVrzKxFZzKg7zFa6JkuK;BZev^ce>gK}lyRf*W;mnf8
zcV>5~2i<12MAS-zjh5;Qcpw|QaCVS`X?k$&T>pMq<MhjE5X&8yn!gQ~b_kguE0-X>
z6ZDjxcHhb2nC5(4;|VD?<qt7FKPxCHCO+AMe0`py*2V1mwPgq$3^RzejQJKX$zyci
zIYmGPxt)-ix%%m%ozGl0rqzP-!Thn>AqUC8>xFoZSA6*+rLSOyvF`)@MZj2*BmsYU
zGYbxIX!HDw1^yFQ#VUkvZR1(@$?v0c6-;V`QJnaz8R#p^ly!-zUaP$F0-5VNf)iE5
z{yBob1N}_8=^h32#Q@lZv%s0ge8!V%Y6^8<A34_pj5lI!@J(`3AHVN((eu-wh(YqJ
zM!Ez+-~oQx=>ye1Ml|>wKh@*WW3=&YPNE3vhzDNka)ab3xO3;}ZfWfci3aZhr~1qs
zyZ07?*tLHxTJ#!~7me608YllfLTR+UvHfA&Ph;npve)c8a0CvekGCgn;dA*>Y4v2b
zQdr!g!B?y*3E}QEU+y^>c{~}38GjpTS9xj`bj%Q_T&1oMo&ae^+!oPf1xZvcE6;wF
z%c8{W)O(5~VRlDKn0gkie(#jRf|=qrf8Ez9!l2N*k<0nr<6WVBmvL|oq<1dVs(u&z
zd7iv*-AfHI5c#XK86zI0-;K7<?pY~Rtsa^4oxl65VWQqxQZpi_ZB8^=!pKxFf-kW^
z7BVT`TA`3z*E8j~4>s10Bd4B()GTW#8`crf$zqlj*Bh(1ej6I_9ztgtx;4s-l2jib
z6=mT=rsDd7t#N^JL75gXtV+#v%SGpar-;w0SVeW5;#XK9<GHA=?k=?ourp~wHa5^-
zI3SQf#}2MRCSDvf!_JuL+mG2oY3?c;F7x~<q_;Xwv@GT0qM63365qC23!Jq!ok34*
zr!~`iQ{s)0b}2oW)XT}-I%7TeHGb5H^W=29yQjB3%tUfew4JtKlfW25AE|9DAhbBP
z1+3cG;NLPj%v0l!*GRWM8l#?xSSe}uR~~ndhQ=E{Qnaq%;Fu3%X@lC7dooRW*YH@_
zP-Qe60}Hf^WM4%l()A0<FXQ6Zk=}(bQ%1_;PG$ndpybY1fKUd>{f%jpHw)MC=n*69
z>{yr4#T4MrzEGkHM7A!YecpwpMd-cVT$*@Z(%6(&(MCdp^BhR*mROvD4%1CB+60+e
zTA0LH4Zgr3wFq2`uxDDO^{7S>7#6)YaL&_eOU4%T-+DLJH>RvUq;2LZ-AuaPq(@CY
zYm|-U#bE-9NtucXqGA@5(=Jk48NAaaU8gDjx1JkRM$T)z(YB<YHF+`>H1(ylMs+Pi
z5yz?1!6BBv9GIeG>BK#lob;(?M)#-rpS)-<_Yt&s4E{u4KSBAWK>s6tKB^hfm+6cO
z{X9-R4v7Rd0+zf52AOsT#$W^{V2tk)H^!7a#B;s_&q&4-JY38<-L>t@6cP9LX{UQ!
z3^fS_<i+ts?Xa1qmFpgIVxt^}IKLzoNu^p)vCHSV22`4?lq{eE(sTtN8+l>}wa8}x
zGBT=^ii{lOHmFrhrDMRKa5;yL)xg%6);Fyj*U^vJ$I4+~d$7!~n6NLHKgu<Bdh+iu
z{r4t)5i5yW3E(P1u2y$Th$(b=6V~F3;h3q}@y%;cIi9Zl^a`J5q<VM(xbfpZtm*&u
z`Y>fGGkWQ`1$ac*lHxs$=HPsfEBT%8+|{g5A{#oYyqlc)GdaqvxLEv0g5c$svSBK<
z7k`$H%679KF7tjVsIJS;4lZJ9eC58UWts3Cd|2_F;h0^NyQ@~HS9;CYmwBO{0L8cH
zg^7>W16{>V=1i5HW{3nWGd{)CO=Ijcf2%jGw+r4LlLv(Q^n42-(YgX`GhR@}Jp^?+
z*rhj@wU)#iegUiZS!>(Bsd(vFsU7I2f;Aa*-8PMjH?cCokQDJI@8!06MNS1k3NY2z
znddz%)vmL+BquKm;g+dqy$P^l8P_DaQolY&{YH5Pmjcm{XqUCIEs=v?4`_|O`g*=`
z&!Y7oM-`Bj*%GS+ra0{35z!q9Y<94?V_{_M&$o70fZ&*jmgE2u{qvzkn=3$n`z2G>
zdF6e7PUd?v71i)F3hg2;u6a<A_j-&$i|@zr0y7J^uI2LCa<$X*X_ktL<FenjzZ*A_
zek8o@zgPYCn{xZSYU}NyOIx@*E3XXe%`V}NNu7`_zPSK?O9pok445p$Tp@cZv6UM<
zgyj$gqhOTDainB|qvQc!(L3v!k(K;VE;Yt|e}bxv3x{AIhN1-BqMiF}$vudu1p|xZ
zop(O(C09X`V;$s2|4ZgC1URkp_il*uL?y%CndTN?I-eqRY}iLRd7soY#8;$16a?B`
zS_E|k2+)JnI%pzCiltCR)0WnBTh4;Pcdf5oRz{Zh>=_?;^TtJ2<(RX-;?l%}oV;~l
z;3j;2U6m!iIFk6<sk!KqQBL5G)o+m5!OI=9U&DKzBZ6OeuK+2GGSuKvvh)35g?f}m
zZflgg?MJnVijQ%d2G~G(eO;hqsdgS)rNXTI8_c9zSb~J9rm*XF>$A>k=%?tf$6^95
zybDrCzM@)uQvt?hTnIgIL)9cmy`&@63ZpU<#GSao_^1NlaPeRAe}o`msDZ?8-hSfB
z)#u1BY&;vxp;T&~Ae<$ZVjGjcB%qD{i~~khfi;dMhAqSHYxih}6tqE}jIbv~ZC0Mj
zG#0phtZ`?i(X|Ub<g3;<tYo;Z1rJv*af){+LD=Afz*S6Wu^=$vhV9vPM#RtLu=f|d
zb@jtk>q5@r&s?+jy4OA%pM$-=yGt)~lXl$pkFGc)1B?;MXZCThW<8cj^h*v;{(0~h
zeqUShAl)Q(!5gvamoj5vppIASbAg|M71Qlaf^BFfKk#>#1%~i{Pc||W>RbiqA{$Ez
z+g0qliC<y-Ir&LCcF1=^|K%!B5LPo9<8Bxhc%O5(hb1S!3$-p)(hQ$f%|9><U)rA)
zU*>F^vBk^&ol})NiqVN*wr)vEOk0tP^A&DoYaE3Qwl|JGXr*#2Dg3S089{oxH+V@R
z!A)&U4qIkf&ohvW@54<cH1v_g{QJ6>{Wqn^v*vPVMtcdgEkAcW8tMLOx6G!8$&tW;
zkPIS~U&z@=(YO7yyLpjubRh1tsQQ_;9e8SL*C3(vu}puYkL)dHc6O?TX3IW^_Lf)z
z$O)KfNUWJdu~-0e$J)QSkq`bmU;O|M8A&f6{IAo336UwvJZJu=ZzhDzdm|#gpw@q~
zQx_^km4K;Y@%f?Mik6m^PzC(|qd1~&Un`t+lnsc>WdbTHl~X#ieWri>uY!sHx3MMm
zUesSJ`mV-h&C^+O{?_RPvtO21fGu-?kgH4gY^jd1cL#?^O}b7&gzPwx%4mDae~dNY
za&^8**3i`2i~PSa_ntv*F5S8y&H*qcn`}S=1PDydU`&)mFgb`|G6DofWH8`ulQR;C
z3^pKgFd_%TwuvGdBr+JAoNX{U*u%Ns%&mRr)S2^5)!eH2qw11g_3G7c*Xr)o&-1iM
z#;)V1D=@?R_J(J3W$Cld`q+|PAPC}qwWXMfipbL8g1!wi?8;)A`q4l@&-mO9FFfV2
z<KBHrQIgjvp!=<lV`6~iJudnjRrj0XuS>U@mVdSJ&i7;vJ1(UKj_fNKX|mEjN0ES=
z_EJ$(B%@#?Mx*WuB*{%kmg%)obi`#oguu$6y2TA$b-Ww8?#nlBp;gtUqGFFD*Z<+~
z1OGinrH5mb8bobBe43rvWZy@3dz$=@pnZ(_q5Ut1w-M7Nxk_Ly2}W(4w1wZ9{`2c+
zO+{sW0y!luZ%BO}J0yp(dH$)Z#Lg0z2#be?cXRvtr?SxaH%h-%s{ln#s*Y|BAgw;6
zP=kwd;SfJ~z`G=BpyrGss#kDJH;!WD9?9)3#otfN&wW310OgEVcs`*ju;&QB_q2op
zX(Cj`1x(uSi<^K0PIC8o@s$nXI{bBZGQgp3sYJM?+fl-&;)Zg&wjLBLJH^{<D<thS
z=l<994Wa}7rHk;=>*+sa`zrjB_rf~$gm}2M`eVN!KOuPkkgcoaNrdFT>6HAk<6IK$
zXX25%;v=$L&*im~p@O2_#XRWQWPzA5>kZidXwhmENnO0tb>xa;&bmWh^uEVtJX;bd
z<Ai}^FetpOs<hYR?n8v*2g0!9EemGc!`uqdJWliq@h+J^3=8rc1YeZkTE+RH)D1XB
zo~%<5)6UK-luPA<L>D@qyXZl6IM)FUoQr)7d29U{>?BgyH8gfGl7b{)aYL)7EUS1g
zin;=6kr{jE0W{BT-W_n2pdw*nJl;h2(qWIS7Mh>0zhVJ;iG40Xt-076RqZxNym?39
z=C#eL1Uil#bN#{$L>~z)^8?*C)F2f+8rF^zbp`jwxuXWsgeoc(1PpvWah#I>RyjbY
zoYAPAcgZYu3t7TElMuUa*l<|}UV^!I%JjI4w2W6s?}gibIY|Jp#10Jm4sgs<6;7j-
zTC?DtGCs8<&wSsPGD@OmHkzrxQ-_%u2}mhM$B`<dVw*EE&Wj#@QVVl@P4*vL-@v(e
zwv~phLCJdjt1NxAy0b#VT;iO__CJQ?zfpX(N}{S#GXK5#`1i94YF1KFCIZ^*#-mYI
z`BDn)q8v;)bR`9#r2cz7&E?J{M`ftb!%$9R3cY^#ib>lKFljFvcBCQvBUtBGn~7MD
zr||T{px%;~>%VF(H5)6D#f+47?~YtV<ohSwN&$-vQt>*$A{o@Q+Id;TWy!!`p@#l&
z-+Dg@W9^aj3tUwd)H;S3YnGYyY|IfVdO6mLGOL91rqQ5Z_wrko)mZ@YXk9HTYy<%X
z@d^QxCB0>q{3-tb<I9b(59kE(!?j&`jz45lj3@4q`-7YZ>93|kFXpVjw{Q$Ri5#=}
z5={90ChY`$th&H9pcr;lx)f2Mef+t`L1de=23ROd+)x=pkLgI(HO=%THUcWI0BmDW
z2$LTd^T3<{Wxt?H<-P?JI}=R%)xd<``*vR7jk%<%0&;V0Z2=5250JsW4ml5#dMb&)
zYaUkRzF@4i!nb~^IlqTuj|Z-;7(dxcP<WK0H>aP15i?L-%BBk*9kTG9>ltoIK<7jy
zmm$7=c5LT=(d91gEt$TDR)sj=A=I3BK(Rfx%L3MDV{X-@yh*V{DBa&M7Io~AH+E1F
zW$4n3$M@%O`zS4srcUl-UM@3J9+4uB13)kc$xv+oW>9(wt8Pwo?$LGMwU*S9B0eyd
zZ*V%xCF(Ziz8!!S>)UHxt^U9*smu@^+zX2&P#J$xr;*R|@A=}6-#=g<!R$%i!Q2o)
zvgfnlt~A@@y9mL}gQE}|_OC>?vb+>M(a!~1M&|Z9Zw8B55w*xM<;pscrhnX+y(+}4
z+<w>gukN3fhCOwYG#=XD=8PSEWkuX|3HVK6x5D)DecDV^GjBzjiGY2ZMh$IuBaI$I
zXeqeaCrO9srXpS$x4ffhv!Y#9lhBybqq@18`KmBl2hClL87}q)-CxF)G(Nf2X&1I}
z+tMzGX5o5|yTechTh<7u&_%fe7)7uM1?gvrnrAsI|CGqs<CQKP2#(m1HfD}K#71?M
z8c3!RbS4OVl(Zkx0E5e=seALa-h*cm7;dYz0Z*+O6<2s!M%l}_z?t@bqjO@Kgb|uK
z5-{I8!^6rur=h%@lJB1wlcH(uK+TlQ<a1QswO{wb(9D}ghspK+bn~D#a>I{+VApBd
zNM36{4p1!>M8SHB8a;{6q_6pyEK1{GVoT#o^iO8$vs-0$dc$TBX(Emoi%1BF%X6HW
z#<AYLKpeX^6&Y{w%6Ob+yX@U$o4#N+3f`B1mU2?pQ#XiJXf%k=_HKYF3`Q3wvlBQO
zdtaVRg}kd}VXMv_GK}HXfe*J|yh#QL*J;8>WO1P0c%JY3=J77`>W)DfQS98$o6JND
z#_h~31WV1rE5l01?6kG$NkaV^T(Dt}N11Rax0ZJX-<*jhcsJPY`bKh`oqqM=3aVza
zx9S_dY1Az{Y6kbD|93{6tiupUVZ=K*`B9Iu&7t2zkqBE8^$*D_$4U8(Y#=c`X>=M;
zj-jg-<>_TY<Xh$&34zkPgYR%_FW;OnwiiK74Qn0FD%y$a-G;ks1Z$yL{Vp;gryA%i
z$6hg;y2_8!(3G({e?>9oNkivx4(36D9QAuay69=GNLevn$K{w@_1WrV$)$dA#i~{d
z^v2^j?02cvYXOEfO%Y1aHdG{iab>BetLID3Pl)t>{SlvxhA_0&;CaT*+$Fd8W*Mo!
zpsWE@*L(}CDQVEP0K|lCU}wF6g(=X3eJ&-){bIE>7h~Nd?b;FB&bJ)Y6?XIpYWqqJ
z#c6pX#1mpoF(IXBk<M~_9B``KlmfC3n|n-F(e^wTX<l<o9dH@`-7)5$30yMW|1}>Q
z`mVg=JJTPs4-x+VDO%rvBTTsLqiWxc>_hZqaQp_rs$lY2G+oXBN)e>kqpq%EPev(>
zcC`Mn>4CQVLq@|6$q4NzZHM?>XwH+}fSZ_bu|^vd>+Uz}p8%n-U~$fS3o$e~Bep_N
z#Z^Z~JNy63;undi59N;F9U-?J)`-pA2vTYTwg9Y(5QCaOli{Eh<+kP)_OcTbo3i>k
zzT}f=F-O?<kMrD=`2@XUhX@h4l^{vsK`+F!m$Oxu`whZ^_%OGX(YnYZF@T$O4?oD<
zq>`QgCDMtR=Wz2d#e@agKNh^ZcooQ<9zf)MHcuHIQwG!TSiwxkhs5*vqOKUbS=n?s
zM`GOD`P>J|@vkw`(CloU50lD9I)wtthDBF%gtkj&FdrFNSlqu{C^@{HB|9iluAQ&^
znKUdCZr(>&TNdTTYqu?ytvyxcu`3)Jm+AMEcGcs&iH(aQ6>u?pBfFkC+$OY`QG}F|
zuNpbLLG_`5C)n8I(MM@5Gf?5rXA8>j;U!6`@neW>t3PC(s6<+G8v9f*z-gWPiXX)9
zA!2_Xl-Yns6$nZd&*`_+ez%G=*-sD9?kV52l-<G(e0gtRCE2BkXHmup*Ue;@y`N9X
zmq`9GclmSZmB7;Up0Rpgy@^2@{C=5)>dXXQh#EPIkT+cJ!UQ#Km2NMh5i38aV?W%4
zepCy~<I0!Q*}X3kx7Rs<4HwDrH~Eae2yZKj6dn30#~s%Fs&|hik5lfpsV;+jF`AKN
zfcT|f@T=)aOcb!=VQf-kV98J=I&bh~T7|N7lCnEfAI&8JvLppDdMU1oYODcv9r_xD
zbwi&<7y-po^0?ICW#iu;8Db3s=ry7$l!OWEHSjq8_||d?4S}jY6H`<&+jN|$D=Z$r
zlxri`-6@V90*UGL6#i&7OCdU%lAe9|P#RKYn7-3?z&M9=S2IVe7h#J?g&>HSs2DV!
z$6;!)H4ZPT=Z@j#KGsD15WNUu!RMDW*JuR<fZvpDv512#13)ns1?tP{kTVLC@s<W*
zG32o*_v>P*$MV`e1VHW8G3mSGZCmc#%^t?0ACsR60JB!vabcMUlt!wpGVTqN*L}tu
z1>B{z-_1Lx)ItF9T@2?Xjn`jX2Z<RzDcj8U-Rv2kO2(Itmb>(;YGnBkbkXq**s&xS
z!pb&^@L89Z8|W4sR$Q{cZmGH*vU~ZO{i(Q2l^yw=-kiQwBMWnJVqD|$_(Ky({j&rh
zE1DQo3egpUiXH&@m0o&KI&;lKxsHR<Z}i=XuDquU1Ou#2^K8odMKf8YnCb^na&)RC
zxQK((F|Bx2BKYw{y&L+9`jS{fopE-zkqr6zsgO@ZD+@7jNUzV6BD1B;rWfT!rJ|8)
z(BpkhNS~GtmdpaDY!s4^g@YtjhzZlaVQOK@RcC_x2Ie2%S*L)xN|VOXl|^3bDh?ja
ze>a-YVb^Mcil<CMrgG$k);9BgPjzS-&6-+T@$x@J!Acpn=BsUL?ce(uKk+b1i^KZg
z9}|vZy>1baU%Qzmg-zl0?`~KPV%!Z6cTiL?@-By1w1_alcRj%)G5RhFu?tggKE@B#
z6)=DCM6IL_#`GelcYwT3db*Lh?4B6F?bw@x8}+o5MZ+~;J4(|&%;wNmE!>fXfr$jU
z+}_6KrbFA)UAgqs{dIs0XNB+RSwO<x^c~<8(+YQ|0|lx6E6+oQU)akl!q<;(UbCN4
zS{|mOE8Tgp_N*<LEDllhercxq?YPSW(263?qM^|)0A?xWkL!MX%Ih^dtbAUeo8;<F
zA8!FIz+a&=KvDTBIg<AvbA5?5`^RxwaLJv?oU)?xWfuJe32k}tOWkJ)$Ij;4>-=dx
zbX;F)J(w{+>0{(o5u1b2yiRjR!uz(DoQ`NrS897#@OhMwTb1D6!ndVdv&S_vSU4>9
z@6FPmOVR;70YQmhII^(k?(j3;TOd9ms2+9Ds%=WO(o^T3*^^8pE#q>@&TXP%<|aSS
zrzGvtTAx(xhRLmDkX%}my}H;=Rhe@c&w_=i2u(|a9Qq328pMf(LrPus&ch(wVwc};
zJKt3&uP{~2*9v?fz0%K$c!Pe6&(E=l7Ti@M*`Umk!tdfW=e0AVnhQdUbj~Z}Bzwd*
zRhLb1`A#NfhE4WAw6!mp1tYoJa|O7RBql6$&fQ-5af^zI-3YwVm96{(KNiHWw(v#V
zzrlt|(DJh_X-=6R&g;r*6nCpfD^(IA7Vy^kbV3(P@Kkh(OZ0ep<JFBWY$_AlP~Bl;
zs`tC>SZcQrVEhO5F<HRt7}6%2_G+--Ya<gg%^<df0jx2AK*s=(RjpIqWNtLL1acCe
z>D#A(Zx;w5xG#v8>O~|;q5JK(Qr#{BM>m6@TRitlsqG)5@pDn6R8FKcvBcaEq0b8W
z(Z2i?`gaI*az-+fFUPxo%>}2GrQ1%!C&?3$BH1Xcw}-_~bp2oxihm4WP`M4!^1C|_
zU3k1~IXK`xn_ErriTtJxKGFI^cEV}5%we*FJg<s8y<TEZj_W{yiL<1MvGzt(xq+FY
zOhvmz3fiLOfydX4FL3u2XdHZTb?z+95PcET`Oj)jms`4bZ5}m#zdAYj7XI$K-+Am@
zB7WeaKkxX_+N;K2)H`{XE6RS{CR<#p0scHszE`cLf^jlpv!!EoF-63$)NW97eT>o^
zI_^lJL2>EuaFMvU^Ve@t_Df@#per!LC&SCJFCj)Tis3QjL*vsy?}0Y>7vHt3c(SwW
zk;N5Ruh3G7US>T(JjFNZ^h*xR&T6)hza0d&wz`YEXK!5CE1Xp*1pC*Ww7v+9i|jCp
zrDl$gkxN&2Q_J;f%XEvUki|%Uk~&hmuuZEO49@nQm&Ar+TwbPiU+pjT{lXU60YB$r
z9|dtYlyy{?qk`HT#d%8HiZrn+ucifkm=I0#8Zr_rW%<wOw3(V_QVF;nO*ig4lasN!
zu?=@l`hpHk1JCvld%l1FA!DKnDW}Pwq@d6Cqix4<4eM6(hs_&{=&3o_Z;kpt!8K`%
zJ^BizT@gI?Y#12<k74FwN4O^~1-on3PaUv2&2(dB&{NuSffZAH%($dG7d$3rHf)w3
zYUu3wO23`wYO)a}Q~mf4S0k&d8S)t2sP>YMH){~fr$#&+=EiB&?Gf1TqC~xX1tFDP
zCMJyAhkoO&^2Jnv7@C5`RQ6fw3`CL3#EskZua1SR1IoE1B=uaE3Gll>W`^w)@tIWm
zIxxRVg1n}R3k2tg;eLFR@xSzz7M9~c++=YA6`M^-A9yz`7uyiE)0kOw9%CczH62oG
z*$R~RH)ZPz>?@9^@6X!)&R?WPc`u3zTB{y{D19DKBy#hnHa=Ptj>O;&sS62jU*?GF
zYhmzFgZH=oOQS5=ja@=i%26xLKv6d#tM0wi=^I)_4R#{i({a-v4F<ChNW}ywlHU{a
zGMNPAuz8$SZYp6bkme-WPvd5nf|n5|n(s?e45GKcYBMIi8{+3O%uQy(saivWuHF9U
z=#;k{dS%i+HWk{5C0lX_e5n)G%MmZ-fip3&S))WGQY+2^s035P`|3Uj@JWi4IaG=h
zjUI>-C_gMwv_8p@-l^>BPtXKPTcm92!%I%$svWcd`?}`wy4e@I9-?WFO0nB&?fndm
zc18Nkl5}VK#0WP`#P^qao^x@p>4zM(7bDO&=ncOzd9U8&BYYclZV6k`7S9Xd(>9}l
z?NC$XSzCb>g{C-`IDd@Wc@%zq?1|Lx&FGLE*DPGDXVN;UolfaY%E(z!%^6npO<Gbd
z@t|Q-&J%r>+I>%ww0d$FOQtR*%%6M(p99=kQpd1s3$F6kZAYR9y|b@SRDPb_IKDHE
zoc5Jw_ZN3f{yEUBPr?T;bfI4qpz*=)P7Lq&da{IRrEw_vs?<zLyQAk+G%VhDLUdL&
zmvhzODDW{GQ{>vCuYUIpw^0_Ur#B12Mdh1>Kc`av&Sc&A$o}l%FR0`^;xq6ny!+H+
zH}tg@+5f?1yB?H~naEKgQHU;x7+%(B@BO>gUGm2Dx0koB_6K)IF2ltB{#TUzuRd>*
zTeKG${~>E1IV@o{xi&2_&o`uNMgY0+=7Z{vl`D>|&QWIuD+9qFNBnFKXcjZ^zdz2D
z`$uD_cUt242BWXucA(<)!?^PY8`Ymo7Y3|1DDv!(Wh{!3K_$5C#$p55SUqv<<L!>G
zmBiNB#+b#hD<VO`ic=;k)M%sneR>Pip=jyl@l{^WG&hbv&}A^>V_BJhbUby=i8w|K
zu<Piwaqeg*ZtCHxzPgHY-cQNe2@nIPxeBl_OV$RwjI-Sv00G~TUDGo8pB{QzM@_zz
zceQd{^-D9H#tl4q!E>8ZcKDURNfqA3Q<N`8arxLF%^RcE+NW0rvnRY)C?TU_Ah|v>
zMu=xdCE6XN&CnUw378H%;risTMeEZJ{kjq(5-r4N;Dh?%@-}&7v?GK)zI9jlC7Fu|
zmEj02P7Y^Q)=4_~ZS3ptsEB1={MS~BWjb$w$3Ve>BK+qsU)JQsopb-QNEvf6hh+Nl
z^zQ{m=Do(BT3Hx@)LHA$!G`*ctkM{VmNWaK?q=?gVX|8<Kf=axBsyOL#`}SG8S*pR
zUd)y-a%^X8(%XFx8N42_07|tU@d`HPG-s&z<X}mS3PfZx9(VKQEu|LidP)!i90pRm
z&XwU^lu|%<YPADgYiLOdQ7Luq_Uf`JWUQ8{AHPQ!Ti@Uq*_{Hj`BBEKUQa(lL9sr6
z4Q6Itw{feE{iTnI&<4cK1DcMtS}TG$6<~2MT9BEpR8gfZcBL=*E2c{DeXKDvm1UM#
zTvwv)Aa<Z%n}Un8Ev&S0498UJY)-JNeR{2%c`0rwvjBHyRwcVp{By}vW#F+|A1{W+
z)FzV-h7IfYsn3(HnTKWuL;QyA^A;DQ#vB1mJS+uFyVmba6S7Ba-7^Q(O^asUH+a7v
zdS(u38p}ZhH7%IV0SmA__Jt+!JcW7i*Eix{nJ#8><j=6hKeu)m5_47YW7}5nOS|o5
zr)t#Z1Q20^>VbMp6AJ4rnwgo9HwzTvHH1jaInsNKm4WKpuTldZFbOdAJ2Dx}f}bD*
zJtpdj0a_)|tLS>>CZ?UCx*zR58pPLRr}H^hBs<w4T&mf?y|zveigoP0gv@QNmL@7L
z`q)v|tiYgtQFz6Na9VAJ4sqQ0{^}USATIP#ca847bfSOo&aEoZi9(ZkdtJS`vIz>!
z`z^>{2m<WaysB>kzi~ZvK*cXT8b6V-mW=%ImRLVMtjZwjl?Hc_&57R5-YK%~?0vhl
z!6-H|%M;gttLr~6NCp*d=L@L@JnD&kaWZkwt9ZZ`*+!A0pW6{KtiO=#a)Bk#um=QT
z`(H*i@(}Lo+CF=(s`}KUw%K$}knwR=8aV8Wm=S0ID6fvBdxlWoD2Z(@jyT`u&SlTH
zdin(^7mg9*EO$Psha13?;Iff1V)k{_q4U(c8b>`^mmFQY;ztEq9srAJeZx~uB+uxf
zl88uLd>z3%1s1?lcm>8{!ol6_9j3JVqN1W=ZH1p+S1PftgI_I_HaLzRoo-ath&8u_
zqI#&fYRE0!y82xVHSz+MS$I^jvO2u6k#kXlsdED#&h)tC1Kea7;FN6xrQ*`F(Lf+V
z3Btr-$B8=(fBTBE;8dY{l!>8t*090nBb1(Nff4H86G7kBT5XNA8f`6*t~zRlILR)5
zn}3*?)m{4DRXWx1ASGM$#XHr2w@5rxeiAat;P~-)Zou|jW!-MD78u@}G{%M?F|#5O
zSGGi_`ufh9n1bHwd?gR7t81CygoeK(K=faTtE*eR<zAF}V>~_cWyW&DAlBmEuYh2V
z>%~6DfeoL1R?IyyCB%JWnw}8A9sPYib2Z$UE>%TExJW~0agV^mex|q;%q9<~1`HvN
z<CDTpGf@|!6EvEsnXr>uKk<KmV0KeGwX)=4)!#pj;a-uni*UtozKR>cR8{s;c`0rw
zE_rw+=-WS-RPeHzt2Z3f4iokAf}v=_IfXnAm<q3QI>%E-dT#dWa83>VL+B7(LtaP6
z%LBJWeXx<C1jB5KcHTzEBr=sHs>KtP%POPP(|>;JRiO}9cPs!%z5x`*4UW_VYW1@C
zC%8Td5^GndmCOZ4roJP!YF4$6){lYv(6dn{;FIm3osDfd3&g6;RBBur&q(*CeV_L+
zy!si^RqsqL1G~!Kpf~|nY}+!{(ScO5HzVbjd!&kOjAoAx*2mDXaZPQwlmpjC=ObBG
zEZ;WSLUO+J4zpMCBwrV`)o@|ihS?!cd6b(LqoHTk&kP!RXTP3Zt308&)EN^@BrmFw
zJrDR0m?{=HEUqVLChE(*T6D3KbdD4pXlCNI15#(OUwjZ`O+<Af?hag`^>;#|dP!PB
z>Y>pC-L|U%5BO0>(25d(WY@&#F75%C9N=!*$hB<sJ^7e@C%9DCS-(Ih+x(lh%dhOE
zZ?O4~3)MB{bCZ?Q?PtU;W13#Zu2Oli1-PrkP-E0=6YsrPm(0!knx&z(hzN67?O7ap
zLNh?}0mQ~4F`3Sgp?h9Or^NlBvHJnG6-19m?!nBiWTGVEq`p`7azVsPQf3GOx{PMW
zzPmYz?aRO>N0zvaT`)ZSdp*U3f#HYrJ+As@G>z3pMn)-AOiUkJAOTW$(aXq4pC@NZ
z%~KFJXFxB0WgIoZ^;w2|n0Wqo)v#-6UO+`|dozQld5gIniZ?T$0W!`B(YYX8NtZ)_
zrvCAYZ>{WQ?Adq-%X0ZgO}DaSI=`&RaJ?_xLdn{-%?^73V0^vu3yYNmP=Y%zp2GJj
znT(&7D08JjA8@@)La`X`{qc-3FV!8%iTVb0I4SIsG2qHHO)9?UA)O0;Pb!k}p>)KX
zzi(%#iNE#PleLwxIMKkMTtW?7Yh&0<?mil`ghw}R8jzcCIWLtlqub?}4c~Gr^tzDU
zNNtmM4oTN@R=-zvdTYK-vZ8L7(_u3YwfWs#Eu=Xd>=jt1Bq@6j;oLW-Er|oP%}2d6
zSIJ%fmC3s(MK(6~H9=1;6i`!IT~!O)<!U4y>8H(b>6PP1Vq9jv%RwGZyD!K*hpFFm
zj<r&#Z8LXrxpeZ4ZbeVh%+;jY@O<{xuh>VBt5h+8S1(HU)#rS9r5Ol-WD`XSma(=G
z;}5JoH`~~&ubBA`HDIh*htUq6oOjwOySsP6mlQ#KGT|!bwcbD1Mil+=s+KW-XfNoM
z81EYc#E*9!aGOUHg|wr&ZnK`OX#4zme$lzOxW;YUmQNRQ6Sj?{X&zW<z^b1T^J<yI
zm1vlc!~pz(hG*CNh#6QFxY+ou=c1>odKFJDZ23-gr{0rcx%AGYw`xSDU3K3LPaYji
zg09b0&&Nx^%ekn+4Bv0gkEU9zk{1W$(sb0zCZ23c`wn<qOq<bVR7uW!?DQ^ZN@Ze1
zsl;b&Lv@aTzBdvK9VgEIvLIo>9{Dv4T^P^^Z}R9|=F907BJVvO#3}^H?8`021#Wv{
zDc|dTq);n<FTK$NX>TzF$p8_Am4Hj&!HHV2#S|drd&RK#b;*Hb54HL5W2MXs6+TnJ
zJx_mIGdQSZSjgQe<QFs&VoFg{Q^`|-8qc~&umlf|S+vsJt)iy;RM&csn6|sRSnnYv
z2guqCuiNml&JwcgLPG-++v}SU&ot3tsKqziXfB!$|62LI)KNYRnOJC94^85aXytAC
zQ0jP}KfgCb>$D{5P10i0j2|?`MGV4pM!h~e&%*k4^RlwT=gS|m$3~QnW--Pqdp0I>
zYSBV;0Z}Oj2)(gix@i4;Kq>#uh8_n>>FI11RqTn8&;$R=>7iNT*NLANehZlHb8h3d
z_Fb+@*R=@~E=vQ7fe@R)jmZz4VSAZ5J_sP-ixwXV4LR~G5CfD#)n^~{(Wlxa{TP;I
zR{d<ADLFaRm<cQex6d?yXFYGGJq+^bMzG_1-~|<q9oKEk{%Ww68fVKqpmT~Evo9>A
z%#_}B#4KaD=Yj1jExO#SMJO!GPW!pLl8`|yI}dyVpT?>{(u8DWQ0J}&{YQu}K^_U<
zB#g8@6-3t0KPo1u#ne@<Xg3MUS9lGJR2i3>R7)kOE86`bdsBLjdv{Sz7fjBfAjWY)
zuR9OEvtnAXZaSH@@@qYae1Y5RSmcbS_$2Y*@9I3_`49VQ%X(tI$=`wIz=3@k{b`ib
zqr%>M)X|-BXCrSUlMp~O3*uvY1XRI2Z$o00rzw7FRyjCxyOp=kF6*uts0Um$n1Y?q
zm=<tbWu~~IA0>BOHC0QU-{*Sf;sc-!vpA_SLtU%n=Hf#r@(xS?-JEHSnx8wX`#+9a
z<20vMF}mJ8j`Z;25BBD@d_OB>OgsBk9MS0SMb~@`nomprka3yV{a8zD1HmsjhhE0>
zd#LX*%;8%b$714T0tsimbCo`xkJ8wvINf$Vhj_u`%`3N4G8qVU?w|*ccg^P=Y=7Kt
zB;7snRfDTB#1Cu^$W(NBmTTG*S|ZI5r96v%8n<Xs_rfy-W9>79BU2;l{(g(Lc-|Oh
zzrVONQB%5uoK}4nPQ5{)7EsTj|IEr$-cekGh;)IV8zGFe&shey=(3V!66Xy1Y+5bM
zNju-TAQ>6~-Kxpd&fa6{`MOYek$UbjUO9JJ9W$!+sg~4!*+!9Bq=uU}GlExnOL%8O
zt&+Bc)L~G_o@@VthIUXX3B%+5Uei2hFs<CErrK^oBG#U!+}?~{)Po6;$f@<a246&x
zvZ1sJ<_Ah^v%m|-U=pVV(^+ONZSuJhlJ^oxIt9VCa?!MmwS&Ur{3L2gOVwSgXDhXN
z(CYjrgrGLo8`pMt4Ez(EqD^LLu6Bn+1M1#nf)-F>g__?B8LhF3PPFB$iH>nd&0t+I
zwEP?oK{#gxxUOfe?(@t-3%HY6?8CDl64F=l93W??ew8rq{YGOZ!VW6i`cZz9b~lT<
z9=#1rRvbaaa&r-)BTIBvM(Hv1;3XKnk9MUz@d7U9EJ536^3D_{V4cTH)YU1rh&z3K
zDFqT<QslPB_Y^)7uV)HVWXPXLuJ?&DyOIyEDKvgsAS^d>Y2(nVJW5Pd;BF4(u*x*c
zeaW~P=(6BwZwE-Rz}X?97@Eel-&ep5eCNd0VGFF7m{zqHQZAWB+=q$M6%GW*L6FYi
zB*)ZZj7v&Qxew0LaivEjK-bGK-SO}yKBlyPzn9ddc;8{AS5I-1A*{TzD<#!S+%3qp
z$hBVH`Ga|Ugnezb!$R(pBPFIM7cKPB##OzAXK9=weQl}tUSiWyFud>|bAzuh>&MI*
z6kD1so+Mhxbl*1ry2EskRdvXV;azA++cZB?QuZ<U!G5M0$eko$uwL61F;TWN82EeU
zD_h|~`{i3|9x;)+r!nCIxO)wwz=(;Wj*%i37f<=Yc)1MOJ)Lq3--x&5gdaPi(tb3v
zk=Gu6DapkR#M5}`R-RHeq&8gC6%6U49cQ!nE{6S?q@v7Yf`a1I;!s1`f5n9a@5l#K
zJwy1$_aLESszwm4hA*zql5`Y_;Eck)w0WKvh!0d$CM3EEBbf&p*E9rlYX+IpI7RC{
zWmBgz9r)53E}vz8-AO*$C~5ZN1{t*21qIkRq%0dhXKJZUPI#~N89Gj%ZY0T{IdgXs
zKe0)=KuesKav4AX?&08(n_DS-siPK+ox;;Kw(nzSPmtL_p_q^3@+mO117Pe5qG!*n
z|2g+10T+;!ob=V$1u9gV<;GJ^*$3c?@HCm8L()IwmMsy~Qg$yUT&-+i&Sxx8@#>JP
z3r7V1{`L7uz@uF45~-09_h({S@mW1^C?zl3-ID6p$dZwbxbbVSxsSbfKXx*6&~d<x
zD}uYMLT&ub^6`7hhxIPKb82fxjIU`osSfD<F1VGB9*+D!W8?m3isC;C*V#6)4&;1g
z1sJ5ZG!BzUK%|p9xb-~qm6BoUzq}HV1Gnyr9RQBf<8SfEU%<9rVMlv%G87QDN;f2E
zsaej{`vzm}#O7UQ8-d6LTd^A<|EQf``x(J2o7>ELW7(0~vdb&T#4|t~B`^!=NiwQ+
zYy_98+Si(uQA-ewH$=S$8}uGF^va#YIb9pC)Ec`aYl|_vbH$!XOSVr@9VmA$GcC{6
zR^S`6xR)nnNCT*ZgB{V+YRzP_kM9%#Ouul|-e$S$DDOK))PC0heTbe0aUI`{ii&Ld
zM|=L?7g=eOsLz#LJv{!9F{)hrWhm;a<nQBi^x7G#=O+d1JE&&D*W*2f2DA8^8`+cU
z-t$4RwP2Ec7$>mfJqQwbH-f4j(>XF__wW={Jt%sa_tE|H48!G?yn?*xWjvxmE;p3E
zy)@`MtGtKw_hH4~0Ms8cdX9H-(&bl2Mt{iqeD3)>{yegbRj;A>Lv{|gPFJY-%@NY+
zpGH82*qANV{^B?c-Mrez&Rc|x{~^=(y~gE8UC8EOkGAPVw4q;5tkEv!{jOvj3wguw
zw5?3)T*Z2T(~R)5_4|aK<vm$&RMP^WI(oc*0~)OwlK8SOE-3gP^~WRQOO8JJSoTJe
zKN=2%ANXVpZ|`3Cp~j4BN_#V8oF1*BdqmG-Ir<Ai>m)Al^hYv9iF&3>uuyqlBV$fB
zPkw<I0dSh;)cWHEzm_?37A`LFEQFZ;5JZ94N!)Szd49yw)edw0b*DbB^T(72%_+Eg
z5X_@$Tbp0Esjg*!e-sl0D9zF1GUV_3l9!YL_W?j9H8DKO%$j5Ed|VQ(c+Kpo#*vK`
z?)8D^<;CQa$T?Uu3$x;zW&bbfQ0gCr{nsguf3%DrMO4p>d_UfnihhNiT%~eVid^{c
z6}W%Znb-0|sRoN2lQti>3Y9a2qpP`-Ng=eOY-ikw%#U7n^rwyoAbhl?XM?2}ju|J2
z$R|656}^rb=HF)bm;S3Nm25Jj@b-fVQI#Dc5Q|9Ku9Pe7uawNh`DuWvwR40J#7HmY
z%Sf*q&NKeS=X6hP*_6tg4hICWxkm1(0hH)mYjT>+{iE8G;B(4FYheB9+kY4$3z^=a
zXV*0gjGI_hn8%$7WF>>v<LSI5tU2{<#P#Ba?jxz1wqk($`a|Q%<$rPVC73!FkfJ`D
z61V*6g(p^Zm$ARNyKIWcbHu0Ah*8!m&lIDX+9~(MxC|l+%qm@ygUl`zFC+GZ(K8pH
z{_%pyL_K#vy&{9f90RuV^3C2pjX$xeAbE$6o{E4CH2oj4RP_rLJ)j&QdYqb?8W5{t
z`Cj7Xy(WjW_w=JvnFF4Slj=4E0nDSNGOfm_cocw!>SA>r$c%M`#t}Q6{$c2!BZ2Ht
zd+e?+W!rXK9Y%67I?j$OiR7JZZv3SAzC->qto`RN20&`*gtbW%E)#|`)P)~gMK5?i
zHtNXIL^RfMC+`_4C&T=4ca)~iSM(C~{T~ZH0&`Pp{}|sd{?W^?8I)4huVRhssl9*_
z?HG4?r&R?|xWUn1?K?G1>=l}3=N}J8#^ye{kZ+mE^i$ir_1+$#Z)7-%j=BUmG@{6K
z*Z)C#^cN>yiDWP-b2f}=LSbL7aLW&W{Wl@#+R_kf6H3VH=(2YsMxBmHvCPCgmDR`i
zY|V^89Xb;@&@atd|2Vbzy?c2<A1PF2Zm90)$*r&dqt+WHRG=O=Vcw;OqWlCY2?)|_
zD+4>T+y#qXo^BcCkR(7v#rL`95>YJ5!_4P0ad*0S64%5Z3C=~V{$0K0U^4oo-)&8_
z#H}Pp|EqykX;FL%O-nVPZ(`xP-qB;(D;>IPOP}PM_k`5p9*x0~N&k%}|J#2S*M4zQ
zE}2`wmD9n;D@))uA1ah<zf`SrB0`UU45FcPBOP7~fn0|Hmp+A4ZanxNd6)#4{?}Io
zpsOAbuh#3pj3EGU08$D%s?}rmC6MfTGj3ckeBM5<u-WRS_eNlq+E)d)p?qLHf<|n+
zmbYY@lRR#bo*#s!%HYdZNNu5b&G6)E{w%K{wrv3>`?Xctax_>NJ2{6O(enB}UPW>e
z*ysC&nGMLgcQYn58r*t(aYJ%$zVH6G_sqXzK#A&$fb1a3d!XMr6)k#uqvq6cL(GAx
z$P@`eI;*&yhHd3(!I?4&(HaZEQD*PR?z0&1RpB}4XH3T#Y~)xi9YN@+Azxa-mgYS@
z*$n<S_yF=}=AJs7P<?$JM$IPB?V#9Y)tY%e;)RLWN0O{O5-Td73{p#tnS)o$&KSDz
zjJfI^j1ttV?k-{!;s$EmZr1%f&Hn%6GuKKQ21WSEUCPMk^7C7U54&rP?psV`gak$D
z0^diF@(s!W$D$CXWIy&t6UYe1+>*25fl`lP&GZ$Y$WqKq>N0BdU|WHxcQB{3(n61q
z4Ou3e{|`SUs8JU>z7Q$Ow<(629hElC1}>SIW2@>8BZZzl5Q{IVK{0yN98F}ium8X6
zl^bS$2+`)}kE-7cmeQ`U+5<Ok_tVzuNV`bTFoZ-5Hq6eGx&4iJRW165>?QG4#gCI0
z`)vc4o!a}g9p7JH5wT5uClxDd^*oI3E2XFf!rav}8#{(73p2G+#v~WYG}_nJMgC7B
z#J}C<Yh9su)vwaK@b9A)?TO==YEI|6)rX(~C6H6TMb^IYm&Jv-j+ZHXBcnA@mWD`;
zQNf^Yp*rUYInSaq?-*~bP60vCgJw{@L;ikr&E`rSCSi12RkZ;P+gZ(HpU0KW1QnVo
z>#b$hGte**n`8O}u5P+(zOP`{r<=psjq7?m1`)cn`~vsmtm`Ns&U+A}y(QhB5k73U
zw9R<FEN1G>qa>ZdgTs?E4J`?pCO;nCNcyGVx+hVP%Ua?S%1K&hs%$9K?AhYaSL?Q&
zEQ%a>Xz7>a@@kaxF!{7-KZzhylL=vYr}+^pZ?e9^Q?VAX%Ym7Faq^X`wJp%taaDf)
zy}3AMQ{7YCodufUfInrv=YWq*d*_`qk%+*;ib_Ve6Dj|4jg-1owrr#gRGcwI0JNz=
zj;_Fh7;PccDwOLg>@+f0ICUPS+g9nI1j%Qu4P8yN&O}*r$M)D&Yq{ZfRZppabpkN~
zH2388u2S`{S#4atzJqU3y|sKwQZTI_s6aJi1m_c$nb*I(sGRxzKA+!C1NjvI16(pq
z>JBZS#6-g5PGU(4GNZ`nEuEaU<J0sL%>$h_(~SwLab0vufJ{(gJv}8|S`vDDsRswL
z2*~VfP%N;HxYBJimy)6$s+!oj;qQ&NW4v1a%GVhMyG5drxbzBGy5|=cZgC7kBcw}h
zDy&^90(sYtvfumSV6)vyF3YU1rR^_?HC!XV*Cw#(l%-s+u1XaOX+V6lS5M$keenQs
znG~!rvR8d4;36*43RQvUy+583#NXmeRgwYFp{@c*hJ;mV>hZcG3>*$rfXwR0hKF`0
z-UGnVM5?9`l}vm6T7TM#wXdD=PfdCT9g_;xqV#7I&2zVepg#4&MLDrPd7>Z5J0HY-
zj9r#S*3*x#s2f~s8IbhA$HI+-1RM*H4RdM0-ba8^7yUq}GCC2pt7>PR?YtHVbz^?Z
zK`ISV(StaSewndy6wCbZR>Fk~NsVMj>(Bg$F13mW!?J~D!#3GkcB7^Ii;bcTqb25O
zt5vyVNAomB>RwcqahX--ka#LHQAo>Qz?T{dp{`SF)P^p;l8gTlxyfSTair{nt=f`k
z?#FSH7wA`1W<<<|Cs!OnidZPbY68g}`!0VeX}0}nlU<h;Kl{NvxVZxB#X;TohkcH8
z?f7B3y-<J1;WMppP5W6GVjAHiubGI9^{ganiFMx2b@22dzccul&?mvq|M+P=et2cB
zuiYko0=w8OYS44a-F!(N9J@m%^o7~NRkBk|*D%=0fz;SusD*0jBoTI?`!ddFzU&(%
zXYRbH`TNsq9VnJ}|8`9HZyvmjJ--x42));QnV>mO2|hkEVf{_(b<Ub}ry5xDu_cfW
zdhRp%H?Y(I+Zho}4sIhg<a5t<6drTpdcS5F<Z5-BHDx*6dSU>!GTv)jcG8A>t_B~8
z{-{ZGELgLuXNt9?UNy_Hbxct_Wol%*H8rP4&6GQP1nv+y*;u{+v6*kOjMdh@O3<1X
zcy?A@KChP<`$auCc5dN#@LF5iIja0OMc$>GatpIrXd{2eH^LvXiLU?T{52!tuHm!*
z`;+JIH*e*YVrlu8HOq-{F|VxoW6qZPkTl$%;5t78iXS?Lm&zvvpOb-~)0;HFb=n`e
zR4?Z7Us|wFBV|uy$~t0Rl^?#mK7P$8$f0iQ+iKZw(0=uJC-n$@CUmGGkhlC0#9+Yv
zQAl0_aiy%&*2#|)YKgeiT`uI*#i#Pc+GkZJ@ANfZ|MtKB|I-eZSp7D${G_<Q!{k4i
z76G>zBfz%|XJ=<Fp;eVt3CqnG;D-D20qS%lPcfTkw+Z@r*wx*y1)^MBVoz@UeZA~x
z05SP#$mqgg#k&0MEV$VEj)cjcZfZysnfW{ZYPW`*SFGdK-HJwMO$zuKpPybvp>(Gv
zo8NAgT?ZAn<V3lI$w8Vh6CIZ{out-BM=2N=oLcG^M`g*TUuKB0`3yn9#7J6Lcujih
z>wgU${zUrbcD<r@RtZLh&lQ(CAM@W~B-^c(8(V}q<!~0P(4@+bD}aF=atq~Of5>bV
zp1JfY|C()q;6+^;UzM|1)`-X&-q^0kHO#@57wovrC1*V}Z=~++Snkp7R;*i5Dw)yp
z2vxrrnY9qy>(e}xRS}e2GrtA|IQytp{eE6bNz*9mjJ6Lu6Cdf$8IO2l{jpp&C$+bq
z|B%_83T6{)FkZEKx;n4Kn=nh&Yr*BIIhRs^tfS@*fgI#PY-L@)AM@9BNwC~j)1Hf$
zi{xhQ-7XiJ#}3~1$LOkysq`belL6@9XZm9f8{Cg{PE>yW+yyI9DkvHF_{uP?3}cLc
z%_hzFidv`EathN93Mac;0t;imsC~GRHoteHoLKj(+dX5G+H@$R*mR^Um3*LdgFPE-
z-{qs)HSyw5pfDNqalr9N-w5A#d8?(T-oS)0{lsD{PD(N|=J^_PD8CPpe(K=5uB4r0
zA-`BKm%AvY?*s_ipxz<il0Dg&tiP)$m!}Vzx>91#>Y3}y;&7*F4WfnSJ;-`Vw9lHg
zLH0?dsyj_|=Yp4B26g-(71f!BNoe@WLr<a9++3qp?f8zB{nX_rbIP2^Jt<0|-IQe1
z%ap45xNE<DSojF**ZciFTvVI4X5hyABD2bLkvE@?hqW$hs6f-H|M~3AX%H(x)eToU
zRJXrufvzeg$q)D55v0;^k-(S=OI+Wjh#uu!@-h9qHWZbzZ!l&e%kn(ap{h@pUcBBN
zen{m*pKkR?9Pq)dRXgqVlU=EtslLXJH`~)Xbj^ts*rl#|X;*}LKDSV25=8WuFMDY_
zW1V~VAj+`6Eq;Ik(l}jGfA}3`tR=eN=$WbwvU<LyQNsx7rF$dPg%<kzfmmOg+NM?h
ziFgmHZ6(H)0^MrNxmf$PuD2Y=19KbNI5c7FQebPojR8*(sDY}E)5&hOV*@@Wmb>-m
zT@P*~Zf?hUuoDL*DOcp4YzQnQk?J)Ja&nRe!jODl?CUDY^^((WN!08&>@85W(|U-&
z*x0)H5fx=3b2(BI;y*<}RL$1juuYsUGl!J93D?S)9@_VE__J>y3$Cmme=)Te`nx{Y
zxLz-*e}7tOrH18MQpAamv=rVje|Z{cd7k_2fmh+4uxGY2Hi}R)<kc6LQF7=6jzKC_
z6rr>;>aLQ=FLyF5Ot%XXxy3xIHm8~c$@301%q0ZOUvbrCpP3t_*qTq1Gl7VYKSZ+a
z2Ly>P7uhy{k%9$dwLpAgx|x-U$w#=v$SXw$MDkx7u;eAfd7C7F@WZ#(*rBS{=zCVK
ztDS=o)n4XfI>&n6;6|o|6WO2ZyR=F=mx=T45d)78iCl70Sg^EjTQglsTN;7H#DRhu
zL%0tfUQZeH@iG#jG@kfKPVk~z_0jWg-tNI`$IH{m9+Pw$n#zz3`ox;SzIeU>A%TQ0
zT(y^2#rn?j0^}#9pQu6*4KM5TgGOtsu16c@?{<Tw_r64Ag8z`A!hlN7zF4{Mof<V!
z8HUb}NYgs2d#_@C2UpIW3*6f~bDS1!M(3fFA167grZlPM4E9BxCr(+hirY4SRCHly
zA?l=}iP4N@q9dd+gJ6k3%QMOKodc_otBdKb{VF!9bQj~eJ+mcqn&hHW4Ns%j9ve0O
zTXmJ_IN-dW8;6oI&KYYj%`<{N3#dfQ^Nye97Pu%yY8#FA>|dHR^m$JTy;a1H`u=of
z_Q^8DqikD>3)JnSdBQ65Av7?1>hk3M=S}3Dx5KcMX$bt!+iX3*TA3J*e_CdNKBtKs
zqUZdv#gs@slBxqy{~nK(`{$-Tjqi#BCI><oW$`n6$5Ml&PZ<bdp7xqNk(i?#i59)Q
z(0(>}aR(i+8+Q{Uo?O`Q_-5XxVYAY+_=0$sgyQ7{GnouCSW`h<N%9o`nG>55_lp+o
z#}mO^(0JH#!P-HOFz2J#D|RcRWF%~Wx3ID{kXk|;dT<2K;E}tx=s0jYA07?iZN~Aa
zHG1;?H_L>7^JnI)6JwM`MVcYW#IiaaMK>3jGG>ginK%Lh$wd|BxDSFNB>yHGJGFB7
z{k`baY~kmpm4LMKh^!T-&`!!PWa4^U)`X5k*DGZnF}?SF2`Rnljih?%2BYSE#7$Mz
z(evUB&X521rqG*2<du$hEZmU$eP~W@Xk^74QhWuEd`0o&eB7C={;9ESYgvXbIr_V_
zJ5?CTUh=e}IXUCd_;99aG)CSl0BR7gdUa)TG7~ioc}I(Yd~eU7;2ljtu|W*X{cAse
zRngM^UVj>+4A#EehtKD57^aF{q_IE*J2YUXyNZhkykE;Y8<}xT>K5k(T@36eNDHou
zzQWT5WJ~+ZsAT&sxT|V3{2}WRU`^aACXRgk?Qp8WdNor^m|3n>#+k1H_pL#PJngEx
zk&0|q7#C}HI0xIwPJ3^|@3(YSw-%M2d+>hWtJP+&sBn7jxxtvexV~S2hY!~ccmx2#
z`l26ZcxpGd7(?MfI8_{PO*-RAS)1vlD>LD^qc4<KQAxboOjN!@x=(yS>9eb6c78;|
zt<D+&kA&&MZmc^k!sqK-1zv(yHO4-*uy3re2`@H#+?_a4Q*q@><i6K8aznD`E#Zr_
zdV3X}3I(0w2^h{wVrEpDBp&FHU^19Q9FpG17Z2s~sjmbf;65B~aYta{#$(}kbd`-!
z6lflU2r;7~zML3G*Ed%etN5nO$m=sp(fy(?K{zf^P&ZC`_L_f#O3)aqecdE8GM743
zoAhYEskB>pw_eStAk@&(_xzj6@69i<hW~o$*ZM=qGd#oyJ{@om?PeL^u#ZE6PrOAc
zLrcACmw(^YDL_I`EhJDIp1UghL7&;FKxQJ<6&oC^W}_{vU4QG~t7h0mak7n>*WXEd
z_&3h8o{`ZYz1>B+RF4`a-z#WIq$`70Q3!L5AWYYdn>gd5Q#Wh77I@&T=%dvPbcGK!
zd}nd>Ca^x&p+R5W`eW4ad3?wxs4GFtr@DnDdQ-e2^6{sdN6iVsXh>O=z)hN{lx+}Y
zBbx}@TvF|`=rH#FspIYZ^{L;C2RxEl17qqRZc{O2IPUbnuX4jRk&an6`L7Nzzqbt(
zcF6D}5#qnsY??laD`|<(XOR}CmcqGtt@Ss9<KxO@j(CYYeR*l%UJg&do5hq;M+o?c
zYwmB>EbAqhWIc((L+`tNQle&7VXRTej9FaKYzM}0<qex*e6%xQ4~bD0cGtmf?5ayh
zSn;$$b1C0R*OSh-D2mM=jO`=C>5+1ho{#GfoPXMIg+lL}jOojO4=UB)m?fj828!T6
z9lql#IWyi3A>M*MIZdV39;!KsH^YEH)$a-uTf8kWys=s%vO$bfct~6hCUJuq*DV^S
z*Z6M9{mG6%vV+nCOL!a_Qs14*<{TVO4FtF#yE4-3=HmLU5txfBLd5DYo$7YUtGZgf
z!tYkcKZw9v{ZsI6JO+d-SzhhLTRJMdv;<8tw<O1$xCY}*#g;J(KZCtZ2(GN<J)=}j
zjs$8+Gd8jD`5@os6{Ju_(}!rTAGBRx2~#-F*ycWBXE@N`qOScc${Mn|TejK*uKQuM
zA8WR6g<(ll)&QLfyh113o4)y-o49(cuw%~&eh<DnBGW{x{h%h3R*clE?+5zD!oW1$
zAi-aPp<*sz*z=kbBm{nYOncH$>wUdWQW!JoO@Iyv%B71^<aPE}ndCi`<L~!$of|OG
z`JHf+x5!R5HG0B8H%f=Oqkl@8LLBNaM3A?bS9P9i75G?m9FfD~DI-!7Kc(y^-0H;L
zh_Mc}d!W()4$TT~%m@~G;?VdW$H}{!96yn}{xxUg)d1W=JyXxz3RH`)udPc%*geV8
zW%#n)oX~Ri<;_A2<tu~cZMEjzJ7QZBMwgS@u^Xw5aw)9|eWAKS`mMLQB)C+-hjLjr
z_dtcs9kf8{*%EM~w?N4FK`3+J8i!_C30<QQYI>$oOlv&(Q~0fB-7W`LQ4Cx0P>G^(
zA@3XQVL_U<mB)QedA6H(BI7}%xt`YNsIu}x`_v~Em5~TN^fXD+MIdImq4`<TXr^IE
z&hec8Iow<Npt6!)LAL5J_w(71^6C%0##81toT$uUzXrF24`H~PTWqc4<sB~9t#-a*
zf2X}WdgV3yC6BzJ{=aHs;XmLwPq18J;U{cN0ZIv$jN*M5kv$DZ%|(=V4YMVy{QUJj
zwUoI!RmU)UmVqZHM8b?8)E05(xS+$VIh2!FJebzOz3CH#oH6>!c92hLr2-Qwy23pe
zSQ!};`tWc$oDxOzHbKZc(r`w~(+N+hFcPZM^JLZk#i%P-GP)4kbXXrDVl66vLZMAS
z0yN*8-^MuT;hG;W_X0eHK*+AbSQOXPVBCE<1Bpm@`W_WWrzDkKo~e{tghUdorFY5B
zs>&SVxgs8}u=!r>76os@CJz&^)Div?!o=#qH@?7Z=#$I-v*Pn=_4N@kH|*z_%igkg
zGa{gf88KUxIQ-swem%EZ72$W$8Vwh(MoCP(k(Rsz58i0cYme=SM9?$f_Z77f3@b&3
z>QJPwR-aby^UA>?A=!Mf-Jm?rlaWYDcElW1C4G>f|1#+RpzXb*np)d^Q7oXSpcFx>
z0-+O-(2=sFC!vMjkroJ5sv?3er4vd5gx(1y^xk1fhfpN+E?v5GDdNqy_c&*)^^JSZ
z-TT~e@=xX*@0@R)^L?KAl;00buZM=4567~F4JV}x*Ke2cdeG*;_9w(H{-9Mw?#Kf>
zn>i=o)2=@2x&<c2E~YVxeJf)kq~|&?Ah|L73)L^#?${J~ToP)>6m@aO%mAu)F%^sX
ztdRSBHZS%#cML5XeQlk1Ae7xu+a??GO)Uj_N6W^%tFa%mn%@FTPAsJ<?zGCSb~zqf
z)^RdP<zN>Qdx2U00nrIWKQiUyte|iH<KfqTm-+v#^G8=JM#mxQSe4I((uE<*9`IV%
zZusG$0BP$d*+?}g9$XN5$&ldk7Yzw({oxZ#hevkn>_GD;_c|sRFFzQwnhn~qwVOi*
zLWRfYW&Q#c>yqO=o<3_FTx%c;mO#txW8?zSap^8uYHrp<9Y-`_mPGxJ+x*iB8ZDS#
zpIY31lbo)Mxf%W<nsJ`~q|DbUnOGXR2$=IdJoA~lAx0d@?UyYkqZ{7fc{-Jst?(`d
zA+zE`)ytQ<!87LDuQ&ar7~D7T^Rre${~Gj-?`hJHZMk=f^Mt7Z24hG2txMvw8o6~k
zV_0V}n2I4O!<zlc7y|{Fp#}{{R~i4FqkXNGQidfbHAk=WZxXNBP!R~e;*0}C;{31b
z9{o}uACxEDYrJ&6-kk!@YB`oRgnk7K#fzH;!h?+>8R}TQ;b7QK;C~?Tk#La+^A+dJ
zO?1_U&gD&E2o^5X*)JZiVwD@Hr@eXIYy37%z{=l291O3Bb>GDO7(J1$yZmPVVWH@y
zV|Y=A*4WnK{-0^TN#2!^utqwLFfcYHZ;n@;qO-=VV~aVB&Pi9d3h~epjl8~)C0<JU
zbi3ZhP!7%CBpTnJG^Q+VRelIzPEi!a(B5-1;HZa9%IWTLdYHdY_uhP1ggYt)n#JYQ
z4?4p3-~lL4l3T2e(#wWjp-Z_Fcpf4w^UEYa=;?2gnBq$L(hM!Rv-5;C=GP|KxZdkV
zf0@h0b;~&Szi#h~W8=|Q2|st}8C^9>%nH;CM?ne(+Yki}6(zWGV}yU`u-oe{a2_ku
zvrX3frZ!Qk{Ico=T@cLU$Wt)Fo=o&Jf_an~3tZ%ksxy6AC?Fb12uxt@RDSpIR=Nz^
z_jIhoysOH3v7wFUVKdHs&SLF32rLeA<tiAcFHFen;VBl6;&rJ8wzJPQz0<T??%6h)
z&}z9>JeN8k&fh;3GwsO$x&o1wS3^~is%FzaPqi%d$z=7I(L*#sc`sZ!S2ZJ8CKAyz
zkqCBZ?k1xz6^bf}<&fkaRKL1MzlL?D?XGS4%&tt3o5t(GK>Jk@_XnBH3Gt7q8X;D;
zHKKvBeZDygbd{S0y>V|Gh)X%rh7)Y^|H1Z~#M?m`$5+C1k-c5Fc=PF*L7+({ak2kz
z{ceRfvz*IU(TDBB-JYE+Z6_`Aoj_gdN}^Eb;`5r<;EoTfNv%Y0v`<0E2BTE2bPAv7
zE|}0PFAz&!<tQ>4=WbsT$gqb{NLPpi*w6Jgxsse0MiP<1!rwB7@skf6s#~C4tJ%r8
z<tnXL<n%O$M6p2ruX*a{JF%SNehiH2O_bqd;Yhqv%|-`?RKMgM3iYzDa<)N&a?z4$
zNE4EM1f|5rRTB~wnzBlnVQtAu(h2g#1dKatYoXz?L)UlAz>x1PSom#HQ9Ht$H~74f
z%+ux4IV~4tu$-ZYN>l)MoIR+#Be@~Wji#d3%%ZU0pC$Ijq~8hiOV-YjLmmG^lJ^&5
zG8-_Frw>^H-j&;Wp8d#``vX|m*q8kNVq7*s(OOZXGO!bQ=>mkhz;oQkfDn<n=&yoz
zP5w+T>*`7uFJ8FXV*2`a^6p0;Enkp*^4tPU3?D`WWgQV;n<a23AAPHR8tQ~6UH9-J
zaB8F$Y{g2Gr9G>RS8^Z(Xc|%u7y|}kJLlA`LEs%F3zWagXet>n%Tgj}B(1K{0jnQg
zG^1DZmsU0?UGzxTBIEgR)Vlx2r&6z(kMt1L^6kOJFN&p3@)R)}E;?p>c=os}MEo8O
z#v78uc%kLeiA;6p_iFOJ<OC!Mh3~#<t1Fm5?kmM7tayqh57C-F2sbD+efe@a%RvMV
zd`O?!`Xx6eo~DJ`Gx-#pzFxZP20pGx4ma~p6YYW-MnZVU%5^oK<V+B?;ddhMB`b->
zqyCpv=RJ${?X6!+sQO*kR~zJ_UY(wvI-kqTlK(sgQ`^>5(BEM(g=6z=QO-G_sfYy<
zPs&s?aPJ3O9UzevD0^6OD^Iq<*=#NjTjaJPLU$(|f0Y8Mp;xPE6u0!f^Qf#%;0>_E
z(Wc8{2fRkGhK>;4)7CDEQUR5xQh)81B^t3lGz~zOHc@0p_>A4wXYWg1hw(m-@yiyL
z2Cu8<UdH+z2)zzAfJUqsrwODwJ@0W6a^%X5<!bHc@W_!PBD_Hn?V<&tz0nQt)IL6y
zJxqHwX=dLZIT^}xUQ?xtGl?1LiWML4RLSYGi2wR3(Kxp7QwY;bhUOUxwe2JJRq%`*
z=4rUr!?#f-Tcn!i8e2Armbka;pyI7zk&Jsz{0c1Ph;7NiMrW_Nd9`^LbQpJ?`^@0V
zlTo8hYYLn3&@7%-1${*5v=YzU+v9gHUT+>8y(L0e*D?xls1Rl98YCs}LO)~~%`L9p
z{|+4`@VIrFbv3>)4lst6yJTG@*+@)A1W)}$SXe~98a{+Nr)X|Kygj_h`V|ef&Ge+V
z2fGiWV&OBrq0MO&Xq}p8y@Zk=rk5X3-_Lf4B#FxTQWh^E>sCadzGn?Dl5|?YpRTZ?
zmnfaf2-k-;0M~v~K=K(dzA~ak*q>%3mHk3RH)xZU{-rmtjuB;J_>x*6UwfH|Z0=e+
z4OGGsVSq9`*+p2lfKLf#7NrZ(41^t(?7C-bTg{T9sGJI`USxDXlLEQmo{H^zV*1cW
zsaxE5PYuC|7Bd2i+}MN4MXJbL9Z*o|@UtW*zVIn@HK-Qe>)tcFmDv#MuL(e`q6L+J
zPj4DUX>hB){$)2HmI3o*GHas!sP&~+kc#WJqUP3m2N=P(#d^s*Nw*R%ln1*|UP5lN
z##wn8NVc-%Nv1s(>k-m(D&n<Kju1gh@)wUf0Hhk<kFKgV_#9L0KBiz5(xPGQMr4lZ
z8&+f3Uw&qwTIq3axl1&7bCL_a4h{XtoRVs0z9o;8=>UtgDEAAitJ42P<*Q5`{VZP1
zW6|AIEp=tY%%!I<GBoTStU7I&h_mCFGU;p3w6s9w(PCKNnt-s%m?Ff-If3I#wLbQz
zLa9erha|qNWs(0T;c*KeGfm>RbCyHfGaBYGrp~E&b88B}uc*JLlbwB-E_3*5ypF>;
zU7<dS$nQ{l`Dn2G%j*<7LW5HXY`{Q#%L1zxerhJ&L!HiR<_eYz50gVMipw^Uh>vhc
z+~a>ft&4LBCJ-nl4Ymf*rc{MaB09r$-aG?DB%q||*Z(16)=Q|rUI{(#I%JNT+1BR3
zCloOBz#3Z=+KR)|8D!D-107;b?OOXd<n9VW=2aWx;QoAkWfbzdMITzp-u&9CCXp?X
z_WlvA|AC3YaQo3|_2y)|tp1g=on^5yj>D)mZ?o0S$w@hAi&E@^ozjNN1dDkUb;idH
zPVUdUwN<*1)u4PkD%MSyUoc6?rHR+}5-aXn98@;JUfuTN{Lh+YC~2O?HI~`eNY0qk
z$slhv{5@vu;&yJ#!SGfVnU)ZmI$L2NMT`h!GAR28U-929LjRk72|EM%9gUSDOrLz6
z@m^oa@?eD`bW3?+X6}yJs$hAQ;u3NPdLtS>@@qM1roHDFx9slLzvwhb=em-6<<-jM
zZL!XBjg?|2d+5iR3muz7>)tbYCX*YBY?0NFj_j>d4AY@CMThndglOOUhmLWlCSS9a
zYO8F;hVPd(>qMILPgPtS(fn`{TBGuKd|prtCi1LzcS&D=va(QwYcM95qX)s+nCjk@
z#7WM+2Zt;Dac^;EG~e>>iQz#-d_cv+zjsf|U8aX>7{Sn2?u^S|&qrgkQtl-ULXpcc
zIgl{HYAj$NPDQoi6a-@WNXol}XDIDBjCAhdvz_S;qtWK@nyapjx92rG2p7wc0?oN^
zyeQ~xDE>*rVe11^nOpyxy7CX@gu<&in)Of5Tg+*GmW&f4lucy!E49O@V<uT~$<D**
zKIYM0f@&T6B7vmPTbIR2YI0t_z2xvg>0|z>0DXsJM}hSIgfiZ+YK_X3qEDCno}W0?
z0B9t^T%UFJj7aeGw-o^WUD3?ESM<M}3ynj8Gty4(aecca>|?;}<Q1k=EHXkkE&a?b
zt2~5OW$SABNJqsVzsU6Mj@a9v5dllGuR!RayE9(j%G~Cs{~*>Y!%jVIC@_yrdPVcK
z*$JDOL12162Y9!F7$Ql?#vRz46gzP*{3c<$O8r^$$3N!d|MbLzs{r5D*Zkgw(foxx
z9(O?A=tgA1b5T(gtVu{$N<(krUWDq<Dc6lW$}O|q#fKGg>utKW%}qOq3kAb_lI`CZ
zbLGw~ODac5l?2_GQi(bYtIL~JdYz$*P!-*;>Nn0LrWKn-<ayJyt5D%u*$Tazv>9@G
zaX8_+0)J;SZ4v1grNF&xJ5O(wxWX4nR9?u-j8uz1;d2P__*Ka`X0qQT>IYC6>+%#H
zy(-=u*_*t?M95g!#Ag>~wSkCmsI_e9jiU&`UXg*E_MQk?@!`Y!PF%<9CzXY&r46-P
z!9fQmD|~*1i1Xqsvszwax&m2-Xy9b$X|l~DQG9!{&CHrp2){%Q;%ouGYau?$#kD0W
z-N2O>-^UAeR6x7zA)LQLU{AeA5%9#7au>&gUs7G$YzvtwHPMvWFMwJvIr6~%JTGZ-
z)sDnrPg7>UK1XFs3obCOr%B{W>bOf-@@m4kKCu}BmqDP{L^=bNmfq+DmNRw|kAN(V
z@qzp<U|aA2yxq}3_=;6v&Q)|!qoIdW%^U9EgBS`Q2s6+Wf3M@<5C1faZr~Z{j*xm|
zxorH?lz_x{ip9Q}^r`bo0$x+NuWNn><>sRrhczZ)J>iiP>VzepQ_cjo>B&a!9Agbq
zP5xK}7rh%;;gc(_*8hPq)vIS{Frb_;D=KQH7GX=}!_JoQ5OUn!(nywx=oVsll8nLK
zuF_&5)o}u1+|^LWw_psY<DdIPOAeQbjj@I*5_zlRu%=9pYuWv{-R~&uDo=rPRyvF;
zu7&fop1)*+ek6Ib(TrND5?shv@xf+A0)Le8uqZxxwN_%((ZeY@*xsmDFAR$8jOWbu
z9#1OiqSkVSh`>;E>()eG_J^Iewma*?%j1-`uw}km;@h7_23<kXL_WXu^el1vxv#1f
z8TB$1sj9T?MG3IrmB*sbnf?2q5ys>N#f2LWW1|?Ez<`OIJ;zy3d9_=?9#=hlgOv|y
zC(mok?55&_(c%Q*Ct}7AlIzD0Vp+6^Y=1i?6a(9hO0iPp-uHyW0LoK#dXA-qPIUP{
zJx2#k@j~TW&a+wzeI*>aMzTu3)(A|RBSVV!Ton-zS~n)3WaOEkk}VOZHpw6cA^yhu
zK49>;jjGpW8_XLdH-Q#y5Dw#8N4rdFV>#@6yDBg}iv)27%&kvbVh{)K`}ec@wyC!)
z)%lYr;N_aR$S~X4LLFZ!h8;gQq9r&dqc7oMCmj9LEuu&YX)e(a@kUOmwzAL827zE1
ze_gDPGe2wvx_(gL4(kA*1jfsc#v6`~>&)LvEE%a3+EofS$+YjWRC2B9Y@91Ja^-fQ
zk4g*OVqF=qcYUEcKJ3Ran_<!-IY+N2MAzYQOT)|VX4}F_yR`a-qU-DLW1ox=Aay#?
z9-5a)mn&e?{A<R_<^Wfeo|Uw{yK2PHZxUcPhrwSc>Qd`E%Mms#S1^kl#4r^NI}|xn
zztMRSiMznwQG0K1+&-3ylNm659!VFjXb5_=rpx8xI;T&4W_73GykOaUK<#9oWz1iw
zTU6gmj?>#mBN*!l<Z}_=vJ}m?sa~ZgS64;UK591>VTV=MVuE-=#~`w2JIRChMi|W;
zhacpP46u^Pd*zn)=ZVWw6?p2U)u#+#D;0xH>AJIJYuQ@rzmyk;HDWv-$e{-t8Gr*g
z_5M=}o`<!u5SCX&22~&cu9WL+t6K5o7P1BwsW<9kOK^MOC3USEZ{thq-<4B!*nDlr
z5;Pbl?WAO9X4tJ47z>6ab5sWs(W&C#aWP1z`>g=coIiw%b%B$r&}XNXB{=JzCw;k^
z88>mz)Y;F1raQF_f^3xB0Zopa`F|G2sh01{5cv?rC{qu1ev?q<^x+~qTUI3=?+iwd
zX%Wiuw)E1K8+QC$O(T_LACENd1PF3l=GNPqNQ#Ui=qi$*6oht;rEj8x!L8(8(G8MH
z5TW-sdJQg6xW<=7>D=!A@_mQUQ+IQ@kT~thd8591$y$iZ`-#aopc8=Me4aK*VjaZ^
zFb}Mvg8u9YDqv+hpwJxz?J6V%4K*Son7eJi8=Ue5*O|G94M`UiZVJR96oDKeP&X_m
z)rqs2BF$&C#z5xs3Dy2iqN~8HL6&G3+SqNA$$eQ%3@f*+2Up&spp(r`77JY+hS%qz
zt-Km&KYK0Im1adSIt6d;Eft>5a%PC^l?*prydm-64q<G2G;Lo#)Yvt)RWS8y1@5a*
zd?G!uGRfQLJmw-|$j-?A5*?zQ1BGGGs*j&F6F_qNk4##~zRsu@SM{4Se0gQt#Wro{
zP0tcMw%3%}7<H~-$}Lrte9n*A=si15O#7E`y!Q{o6)EnY`lN{#Z?0%fD}sv<E0)}H
zJFRK|G~~tib?hctrDL;m#I2zp6Eoesa9x;9yw3r>^^Ta#u!VYPq`){!T*?D%9x{xW
zor_Tuqox^oMQn-VV)OWD|3gv$tcTdpKNLih?DPYc@DUTkWnbAq<u)rV42<hu0$T^-
zvq!(Q!IB6*e{7`@+@U3kg@Sp)969R$AbD&U@wCLom}u*IxVq6r>^PmgNq43H{^8TU
z;g3Mr!|~=7veEbVdL+hdBDOeRR&m+^l9%0L3%|QdsHZ?Knk7ph5FI~aJ}d`Jo#Y1W
ztyfK#NAf_<i=L_hLyn|a^R%EGqj5Z)wi7GQPW7B?5tnT~Ic7#@%+Esy*&8OKOblP$
zFjIUymNIBE$?(>v=-pHxa+$_e>Excwf-6Qy1->Ei09nTYc0cxQy3FHh?MdD~2q!x0
zt*&Dn7RTw$TYo<KP10D+K2D-<@wYNRR;RLV{%cgog~B)qss4^{Y1@;s3zesvcPiE!
zudLfPpM1y#YXASgXEkH`_K)3Vc>~Al-z4Onz+$3dWjx=R-Ka*D5hqInWCN^}4npvG
zPC7H9b*<Z*3@FAJw0zbpl9gdfY2&hYTE*|)yeylU-3f7i=Z`s;Iv2W-;w|s-ty^=N
z-b_Uj{8e=zz*@nM@-BVUFRu&Vv$laYo|RAIBpz+yexx=HV%x8|tq9JZshk5Ividvu
zD_Xwo?kf&OOi)U$FXO;v&Th!8&fDrL7*)(BGC*d^ORPd5UBa${3bQI15MQsx4kIFX
z5y3f0{;}{8L=U+@Kux6W<>IP*YzAGy|EU7_|9ao8xSC<E-b48(Gd7u%NnNMHpUhWI
z4f!0C=VwCp1*YV9!aX?(smXzvgB>pURQQS_4|Yc8lWA15+9r3%-Q>cgihQ~eA!l2#
zlVG9&oeyfFRjjB+5i3CGV~oz#@;thWaa@+J1O4^dT%}NhGRw+KO4iyOx*6nYGB80t
zhC&IXc=KtobI3-!Tqce~M_iaQLb^u#W;BAqT~;q9BUN@E=uegtCDaJ>$moAcsB8Dj
zZx1t!S!=hBHE#AXgcVNnk4w5XyzJTpeX8=MULU2|T*n_#XX_Dd8hK+i1?N=GIk1?<
zN+MT^q)nGk^U8>)Y4;YNXk>)+6mVJwmW3CUv}eQgnVEW>3dsT>yPGT~M8deWrJ#KN
zFT!iqwqKO_Xo_BKhoNQr+PGE-OKkx%K68%3F84>6nee~t9M&I3UlcuA8a@$X4WsE_
z$-w5nbIC9Ge5O-^-zAYW)${~HGCQxBH})JC-nh9Hh~tWSTl9MYeGzi(DX;tLIC#w;
z^K$v8i!$|8c(|L=jed9_>0&O|_Oq}#1Z^@tt-Y#Jpd;X6yg~lShdNFRZ`Tu9NDj_m
zH{CSO;#^OnqN40@f*y|~dOaT4Tma3f!-i^Zkrza1j5XBGE$5=D?oAEUxrLpvHc@aA
znw-<-7ADmXzzwva{eip9;SNI@9?~%mx)6Ug2@0~$qRHx~qDh!p63McTTFc@ml(nx+
zY?civ$GAXOiMzI6`fVIjCiNNar9dW@C)E=uWh8S$9_0-JGYPrXN?;BU%QaiQ3iA7P
z(jdw%?*bNLjb5u&?+rkQ3(&JVM|&uY^;0`){yLSL4U75v)5ufI7u$(R{!-svP<^C9
zWG3tf@z;A|H$1<!uF#1eZqz0`(Rk74;-H@W5+z+ZPiE`iwME=~C!dCYC|`&V0PTE$
z-6@r)C6X$(@LkzWU1t+t623CC0N(44cltl^6~C0&G5V~$J7QX_W5d?J_}s>b-sFhh
z7L<Ti$=acl#ny?e;?di8DLv6NYZHrJKc>z9B}M)_)hAcLmNWE3t3adj#QKX*W}~M@
zeBCNJBNm`*uvGh^(Gv=ppC?l8t7)gRM5_+bKZa{@|CxVDv1zy>*mLse-O7R1DHEZr
z+dO)U&GF+++)2-J#*PHm64~^0?B#ZJm;96)q>aHxz90xka&J^cSm5H5Po`~Sq<QqD
zF+54}jmyj2ns;5H0}?Rv8@#&vDoCurIH?neD>M{?p1X_?`1^Ddk4g0l-?kQB@-8zb
z<#|bB7Yh=9ozPAh)*U*w9S|HBTqaWCw59L;HBI#5IpsH~ZS5;!{##w7{4gFKO?#1r
zO~NEy(r@b9m(B%rN8tTY0v3mse8?49i;M!qMo%F=8n9nGi@A*Y_-`HjzjI2;UG}5q
z9!o2yU8ob>G`y#|*Qo$7Iq@a$os{Tiy^}9D=+<6rekHUi_|sJL;F(C%?GJ+OTSxLR
ze?1g$nRdyFx$2=eR%X#Y;dU8X%j+q}T!hW>n`fjC!OQ#1CBxH%kiG9^3DmX&e*J59
z+Og=qicoxnDZ!B;ieM78RJlYmn34O{pEgU$E+Xb_+ys*>sG_l8lo9E;?gO-p|5|tz
zvr3O{zSL+8<6KoyZ~gST7NT=Vt9rZ%+I5l07CSgJHhgq6*Ul+Zx29nr!stYyAJ@kV
zlQ}=I%@snGIBf#go<j{mPAIB`?AeY}PNygirWZ|9B~{@$)Nm+e#e^e>ZszN`Cqhjk
zOyk*^X!66>@1PwL<SHv?u8m2ZKGvs3@`cV0wyT84KCSYNW6Y&Y-+B@CmlEwh3q)fb
zlc%*#;|nSe<Wc~e7-bbbwsj0=0DzkN2X2#XyFA*lRd!rT%sVIg>}r=v^3rr_!B{TL
zFp4q0kB(19Ru`jI{O8<n5+U#SdF9@5BiD8%WfM>wV6`aFwlT+@XWAttc!Ux8HFyAx
zIMey*-|1+h_E~UiZ-M`0Vw~LCW95hI<G@e;6S@RDGTR`!UY$tDvVHG_AHxmmF}KGj
zPx$CQMTRSy6-`%<>w2VC(0Y6HGpR{w3YsK(=4E?B?I^JK#R|#%urHF{9Yj;@uSpym
zi>X`kbb2uINvZV2fdcZI#GMnjKH#Q#(%Ib~)Y?K<YPZ3c;fWl7@bZ^8TherkI9CP-
zu=h`u^)B}!e`OOkKg2O@#dQ=nyRr5gm*j{TQ^$5GsTdh|hrD<}U0c*uY-eOTXLI0g
z$hkNtlj90vHHjCUhdTN8sclOxL(w|eEb`S6-__ftWZAfY^qDQDC#!u_OI%;RyR-8b
zU!8K4bk^45t@jQj`gr%J5qs8;|NQwW1*7-X2%p=Tn|H~sagWn_`jg<dXG0=oyyhNJ
zgvJ8#$omM592*z|ufl-MqvEijnooZP*gkn+$W@VGrKU^G-~vQb=gol+6?#LFajLD$
zWcwk#9$RM0QTcF>L*0r<hY5o=UDGPN@85Nu(b{vO9XdL*Q90o&MJUNq)Aa5?l<b{`
z@{H|KufDwnIH~PUuG(`o+BLHEh#5?aF%^42UPRGEsLD}LOEMJMl>n3~0((LSmGy^I
z{S1T0=WVKF9mOKL%=>fXhbC<gW0Wb(0jM%`t?4XH0K&JYx<rny;%yMsmd4A$EqE7v
z0+iR4@>sLA3t8QdnGM|W&O8U+MjHb#fk{7f^fxF7GV)d<B{1>YiFHyxo?I?Rj#7EF
zC@x=0g~%VlG@I<B&^srZtX2ys@_s@U5lua8^R7t)n?!2ykWM+ZH@;L!#S5R1W<Rqr
z2`^7vUsWm^+^S#EWT1{80iZl4?2{LZbw%sgheeOHHjN7PJU*~Ae&D#^C{QPUKR}SL
zC%=m2PUMB%c^0R9tY~NYD5dx7J&D6jhL`XpD9o}oI`%QYma;mnv8WojrA@Ik+eBL_
zZ(=J2c&4D84T+%IrxYF5onKKcIB$0-UY4hBEu>-31G331Nn^XUV)wSfY*jvEJjh8K
zwZY&*F47VzorK5nLw0$#h1Ds8)tmK!LT*yL##pXwj=4}GR=D6LeWqVX9*3Yx_`n-%
z6K{74gzhr;lWCrEMTt@%7GS?A$ICu<Nkwn9_kU+={jb$}V^C{LGfQ>&(9~mWk*73s
z%yCB<sh|A4MS>k4O#$7OB1o~=`YJ}Vsk!q40DTDKH|mT&g<Eu%FW7om2dWgZ2A;o9
zAG+Nz{`1%&IJ5;Ieqt(6xY_GwufcbCTugM1)EAX+;xt%CXVi!A+Oyr6&^~AQ3xN`B
zhgBbnygLQqT>hKnrh6?i8Jz6Px^CarY^-+y!`2@+Owckqg-<}Le9uKPLNr(sIHeDq
zkD?0ia+stF_UqJ%(dWOu!`D+erXlz_&Sb^6)Hr5pru|DHla>Ms#2$xfgapXzCmvj^
znxbX0{8?S;0Y#NcD|ZJ@a`+lMf8d+D_n(>wmD&vZudW;d*E)m8-k3OADIO*|H#@p(
z)N9Z<ioOBV=78hppQ?#s1Q;Hpyqc)s%#FF)y;1W3+v*c*4V6SFn8Qw$!vvWO`eqhg
z=gBw#at;5>C;IUhM8S4*m`??7)?ayt2FuKbVTHZ6=oy_6nqaiSEcGwW-lmWIE>G~@
z20j{rQ+6{bu2V{ch9~bu9g+U7lbgK{Api`jB(TiQzV2&xccm(71hf9hQ`>uJY7T3*
zJ<*zQQF+Xc&3NV``na?Hu$@Dm$Fy%7FqZSZYQ#G&l|ziCO$~VW?U0070ko`V+ZtC>
z#}Me;>2imm?m=XyM1^k{jQE&pcKXp>{~6@;@m0b!(8As~NPB9P);7DSHo*ix`V3vS
zsxFh8JhSMz7>2n_yp=pN!2%0XTd;nIG0<EiBJ5d*iy5|a9ZVLXa}cn>+KX@qEf_&S
z*VVa$S2`LW`ldM!Woq=^XK44Fcn18Ls1wgt@1D!VS}^R}APB&G=FmlQ%DkE9mE!|k
zsO=QtS?&|m8~v-BbsBTn`mMm_wUxzPX}+88Hr%mML$M4&&Vi^Mm5b0gEVXVUGW67l
zj3Q^Vht2p~qZL@yR01m;dR-Ha9V?_PoS3K|s+=tW3i7(ri*hPq8JigPNOGu8{rKfN
z6a0!Wpxh;2R$Ys*(d9{1xF^sMJ25FaON^%kU>r;mxuAbC>_GMm3`4G>#mi&8p#>7>
zO0xz=Au>4&%K6Kfg)Mpu+VGrBDT*%fY?p^9FjySk(ubIFNzg?EvFM(sTOvo|)B_Y7
z?&ZlEgiO+b#Sv!J&Orj{n|rY)P$ErQ-5HrH{m0Gg8~4=BW!oKu3ot^d(?Hz`5pwFM
zFe;`}&fMI%Dt30Ur~X7o<O(v~-P}*&M$v@njLPA_CAa@<yh5Rex=#2fci|kG3XKez
zfr%XVeElz2Y7KdH!`9HX)9Jmq6<?R_ii|WXnpgsRtpgm4b$lr|qdpwY42TUhO>(aM
z7c)Uq2|M9Enx(`?Iqyk)F?2d|%A`UZBeFXj7l_ur+%$QR=FSVzff{VFv$M;!>$1L}
zdg3Zzz+{Kf)~KCD0I=ljM9ymSe~~&rx;6D*+iCuO)h+#4NIZ3xD#eQ3#;Qtqf?DWc
zYt?s$F(v;xcH<<sS&x4)Fhi+}XNOa~uzt7`xUp-%b>|Y*2dW28YJL{ag7rpRcIT2v
zdG&%vJzspcEmPSToBdF&T7OgfrgXjiOLx6Qk$hkTIi<315@;^QJkfouII@cu3=*%M
zinif=>h=ff@7?)7KVHW-8>0{8kinFXcH{A7J)rW+jy&(IZl2Z89&`$XtE7U$qe%U|
z{DPHmuTypt_0218sCQ(b-JlM;zK_q|Q#Y~&XBPPs^(H0>+VoFrX7chm&gTr0&P^;W
z==kDD_dX@mf_3hjE{hORD5%dRo!zQWBlW%Y_%eFb>D%Gd(vyqT{<>|ib1yaI+&A;T
z9_^D$rq20ZRs1~JEY3|>1v}o>c##xh7+1|^ds(8G1l&mVR&t3gqj7M7$0h469fCrv
zE@(ncSKTX2OPF0~1<c_4CW@%Y!tirwfG%HX4LX@Cms5k0qK2M*K6~n1qS9^K{|#XV
z+IYdxa6A8LB~S2NXuJrgf=xd%To{bOL!xQt#kl{JS(fqu*3AWH&vxb|Ecd^YGxk~m
zeLcyzJFf2|?&>SbKn_>jRuuNl0v0-K@rTt$nI7r9|90oOowOTH99BMV`$!;ixr1j(
z`vtd_Gl(D3WPsU-6O^gehzbvrof4Df4>tgVnS#}HNY9p=82UlXTC}bJhg#a0Qt%F%
zc}_)Bi&5Q;(-(Mxvu>UxeAOz%zE()+Qi@O3QxoB(kpq;o_f}p+knM8((Pg9Gs}as;
zFEgKz!0nR~t`=P#LJJSfs-@kU6LXH+i)AafvG_q=zGSXgt?TVt<FMM1anp;pg7PMH
z-PZ!wZf>1wbe#^%W|0jzH=*?}O8eMM3Nx!a+-w9wZ}G6Y#}i>rt&aEU5cRdJ*X*1j
z%;;?yCRnThaK)Pk7|9Fc<xDg&PSD>jrg2PqycO|AdrykJO=GPOMTROU=A0@q&$fB~
zO=+H5XH~y$;Jkn^n^(<nEc+&mg|)h<zAgea#fg*WlA@4TeASsS%BhmrAHmX1ecaic
zn{1-zC8^2a{23PMoe?%K^Xcu*)c4w0(|0Fv%1Js?T>YX24hjiR$A-D3!oB)b;897u
zyx<}mEsvXty+4@W0NTYkSldSKZ~H^yZ~iQCrANxIo<2yy`%qKr5HySuDF1wphxn@}
ztJPN{gk82-SX|O$mJisiba$TjR&6ab^<)ub@-6aS`VO#o=xEjlW~fa*N!E$v(Rt^1
z0c)(CeD5Eo&sx_UyvIU7m?O7AX)Ybs-6lC%dt7}v;ct^r-cm_ubZ(9{Yzy~ZbBRzn
zC0SEUH2kM&Fpkqzd^2jX96ax#s(rT=8c4UjvcWT2QPA5P#+6Ab5$)ij;>pe#pqpD|
z@z;nhMk)$6KUSRd=i}+;Lo|HhTk?Gg>&+_<1kfwssW2W<cP%Z%3xffK&$F&0kJ(Ds
z@OxbB#ylC&&USL!CH^7K!$tqg{nnK?hW`4V?EOI7(x!OORWN+85HR~oRMLS1#BR{%
zP6m$!amcm+Y)$`b8wu_5dvKRt1bH}tc~`PaAAcMR`Yav$5ntRZxhLsDyOQWagRLmY
zN!s0Tsa@wj^yb&Jf~zCO27Z$;+3=%><A`peBJv9F66ATFa8W_-PCV;sS}pkUFY;&i
zt^-;S9FL_q;|2YcQEg$9ELf~JXD%YvD7pi!mxo^8TvajCFjG4dT(+XV#=LLP|FO$1
zD}Kx6g&>DpmR2`3OiW_zCF}Q9CAcXug{J)0tDB@j+v)KeVybq!$)li4VUB?`bFO|6
z+)(@+C9($O>|&A`=#9l!BJnoDV(|9cVII#^Q&^^jC4Cin-1*$o?HYY^qrLXb@gKk>
zp#Pj6gva8s?$dVfwpj)uX(b*Rt4$a<g$Ou}<(%COx8Z_+a0sm<dAl~b2Q%b+ZpvMp
zFOc(+2-x@ncc%55q(1|EeKt$wOqMYAVBxUA<1yQE5JS%b?96-I6<hRkVEAWX{{~}L
zFF3oY2yU+!kfB|-(mK0KghwQP#Yh(A;jtFN=H&HhT;bGI_!L1Z2B0Z}v+4eR;w^9g
z>-Zk|z`5%lfNL=3xOexUlj?k-puVJKKAicJ$gkajh@ZW#ABkgl-x<BUXyja>;aX!D
z4i*->|7e(d<l9eycm3R{4N-G}Y?ymbQGCu(W|&3sIgX<@B6G;$!Wx=xT7SNTPl#(9
z%W!>1GTrF-`3PS*%>{X|i5dhvF_)aTXb96xM$aN#-q-UCf~Sxx&H+8?Lfp_As3u6}
zV0zz?3?U?mtBp2ZB?xa0F>2&E*~hu2z|*%@7eO#g-*FiDr}3E5{N`x^x7l(F?yVSf
z>I(VEKxn+IwZ!v`=2={I!#D?^2sakzS_w<`0fBF@^n9?r&xAjorSBCsyKyOl7wos}
z;1^y8ChJKE7=a|rPn2R1l7Z9czSU|+7=!b|vVB21z_aCPPbOUsg44=wT27gXu@lF0
z7^E5Id^w1|h=Tes-AB%dzsA7Zk@Zx$3B93yM3NAS>GXoJl>`P%sj^Duk}vtH<ot<&
zBrH?GkT$0;fwYgZWL*z!D5ygYcbw;cN%iPti#vE{-$?lTor0M<BVQk3luDmF|2r9T
zZEL+&ylqqpiY4)4wYFtW4AmMdXA2f}=O97sG}~x(E;;UU`Q-^rtY2tEvuf_DGmR!#
zy4=w4bm`{8i8z6iL%(L5d59*tDrF0o<{ac>Ozaq~km_IU*;|T^DQ8#SgW%?-LZX4d
zP7Mw_Ub|16=&*^nqu{jGPmuY_sp+>;85o7}+2%9dq{_h)dsZ2%SqD*7>ejrQCf{bA
z74#D5ayS`ER2s+}MW!MFidLpe9Aj6bibMnxTgv?MmIpX3?h}4D{l)?DNy7xw?@V52
zD^)4`wQIHC?&`HolPIx}^os@U@^F_uADH^#=j<bUVelzV2i*fJbhMH|BTNjGYEf;O
z0}(^ihRr7SC0lUQoUSbOi;c|g6nxfUe+rM_Bu^YXNsHFg%$F%Yj17~UNyKVfpniYw
z_`|8(yBX7h8N^OQ&-5)ARQ_;LTjCIy;j#$WJl7hLAKV>Z&CtAh;S7{$7W4Mr=9&0S
zVv+K0wVkQMU;RYlH_5{;h+*7GV{5=w+u#o&)lm}u9^$aLwG#5AEt5J*3rg$9mE`K5
zADTttlKTlrKZsrvsd(#M@OgnKs0Ijm>Hd--$uaF0%JI_(bDKW@@fOQq9t}ph%YI_a
zb{ATfSV)Nlf`hEp(H1Uh#ykz8(}x`39aEmuw;>-#{~|h&Ah-h;L&@_Y6eRIPTxFMD
z2aso1Z_h51D5niEb$odsNc|`qMqRgCp0GtE^WKF{rt#E_ra>o6>>w#`Or5CBRoLrM
zSyfb>lFqS|8s0s$#7HR#qK(>BxzY!Sd^qFwc&~H(st9}{PvS&xAsfmeVeMratp}c2
z(dl#Wzm2!We_xvCG3^yuv`eC5EHGC(;4W@V6Y22D)LR*PC28PZ@PLa=0iP>n<z$il
zIwtIZ2|3H-;s6#HS$jx~oOJ%6F3q*kJ`Ptddy|a!j_19&ym~kxfK&RiT#Bl%;XOo+
z$*n^{OQXBBSJ7YjanZLKRw483oxmDGb^tKPu*6KKJXNZ@*tT{Nq)5&AE;BFMRw!Cj
zRVy>v?e&e;1MU?@!{EZIRa>uQ7aMbu<zBw-6Beq<ut#Pl+6@oJTF7}{6Z{ncv>7rf
zvT>>nbio#_PA$=q7qY?~j6EB#FjLeZ@uIwGUm;kH+z4}r?;~@sbcg{$-NU%La?kbI
zC4IX4HzG+KAhLK&l(CNt7h%H1-aUX_?h)^2+5|WUubtZ<beVFmgY)W$Lea#ZTMV`I
z?zGD2hd-iSjFr^CR49z&L1l#f9}NS>3XmB?@e7Y6J^O&*66D}m0{c97XWt(Nl7`_A
zoS?kTN?UT;d%iADsfRVuvfda4gTUi_^1F87r(<|!@_1lpg8JAJAauO<bjsN4ny1pF
zMFGe^zAF$GVe2DVG;Q~^Fo}GoBN<N)Yh&m!WR#xvXX5tQEFn{YLuG4a`4`B9i;#*&
zuXl1JlLX~(b^%Q`y&-pVWTCvww8J#I5#05_>ZBwzSP7mM;orTL(Y^;o6e#18TX626
z)TpM%el$5JBCOH{uao@wxL4~5tG%YfWnO#!O3y-(VWh3Ax0xspwtZu)8|4cOXPzaS
zSsIl2yNaIM_Qir|YrNmNat3w-8rvgF8hE?sIK=!PT7yfkCtZkar}CS23q^J#YkhyJ
z&q#I!qje86+z~E$rW_zahzZ(!c>bCVV)*vSiM(31A>>!wyY!Cp7gA3XF2d??sk8^j
z?*d@Q6c>wHeZ$xbi(jQJJ|n3%on#pEO-*xlGY5JgG0)i%(DwLrB7xDj3Itk>$8yiL
z46l$F8l8nRr_<%=ZCaEg%w%+qxVkg?9Ow;W(yEPltYk96k0K&j)LYRYS#-!FyS9v}
z&?OW5l@5-9YSga}_^THt){QRrpC-9*T}}${<~2CCYOpiE?FF=M=(dfRDJsQNCC~sd
zo@RdSr&<y)oVTu#m=3F<FLFg|MelA=PgI_Otz5cLzi&-nOnwH8if+@?fqijcPsKt~
z=ySh-TRX=Bb)|Z{@}O7p=4+Pv=)2j{Dn~eXK{hO~9TI~D^hT<v_O#`{B2|20;*IA$
zFkI@oD$OjH@e|5ztRmuEo3p>;s#_hDTui8hCEBLCq2HpII#saF*|V+vqo*2PryM~a
zEDe$BT_`k0a+^D;d_lL)G`p|>`#pna?lN~s!NBEwJ@Df8B+HNZ0&HO;F^MavM9*+=
zQ&tZbO6Bxp!)ZQ!bcZcRFC^jBg)d4xcAg?Py3KX1c2cWZbc<*ZYUrr8H+96l#zVvC
z4l?JK@7P%xc+f%p3kWD-Oufgd!?vo|D88ZW6&XuU_%DSI79J({MlC5vMR<*hys5Fq
zHF1g```HKfR!Z3wbh0<|Jt3xB$pgsf{Hb9qscPVN`r`dZ7k`#nG|5y=j=Q*gD71Jv
zdMFP7C&<-lU!<NLzt&thkND5v-~X+CO6$!Uye6hEIp`N#cs^=9d5o5JZRqmgmhPW5
z2?#dLY;;_;uh_Gz;ID7~;|AW>Foeab7e4xzfg<Tl<J+mNtcLCGWezH{k1O|LpKRr1
z3EN>u5svC1V_3ihS{xapV6T;xr1T(sYh7u^o4uP%>v5Ezq?1B%Ue@WG`PNRb3Vv$X
zX4X}C&f*!#hA9i}grqlJjZs&;h(nrIq>9N@?}olS?`vw8{)jc<oRkq0_dAWY)j6-D
zy^aYMuP622!iU*bVsXqhxGm_(K((*(S`d2ng`{&efu7IA;BsCp%Qm0RFB!}S!2A;R
znj+yz-B<DE`r;$<x+sy#45h57ySh6MEi{W$jJa5?SgSD69eK3O3eh=Ia^m0+1ekY!
zru`ae@fS-$YmdLP@s!@Yxk(;HH`E@G=IRA!Qsk?M!5`}7O$6<25vp`_($JqQ=`Qq0
zn2$`CdW4GJg(>^%3XPo>j04A+BAKGQYXZD-6v)YxVz-k`IShW)8Z4lY>(OhHuA*&*
z3=uX43lF)%J#g=F`k#rh#*0z)6)92h-z0CKo?_3wI&e%Y!4cnokSJ7pGPQP9ucy<Q
zZELQgXZ0gjBDurqZP-+fqd5Ca0jVZv^r{~xFL2nUbL@Y+^#56S15(ySp*Y%4Di`NS
zTG1PCXKAsc_LaPeJch*tP(IxNB>N08)eT(foP~g$<CCwxls6vGYY_3w5+)H7Zr)K-
z6DsvdZue;6EBZ=rd(m}6O2KhkXca85zRw>4U`U3drEVDKbOfm0%fpL1(kTnEhou><
zaw26hP<`1sQZ4uH89kafZS5M-T<Om`G*!JX8!of<`>M}OguH#lDMrkddzRt)XDv_C
zGljJT*%=jRFGK#?nNR(^n8E#MB<r@C*p8as=}M=l<NfUiuHp=_cD037DU`JLBw-9*
zBXlx`Y4aX=Y&FhFGGXw%v#GCYtKco`m2393Rb%~d$Lb@FjEMY)xV|FW&S1h?zmH8+
zZsPZoi@=<3aS2>s;}}_V<^b-nKS`>;s{c6dVeQ$f;7Y0v$u}ltM!Tw8oBGhuGdQ><
zm4zLduowadZ`;k`Jr=eJ#U{kim(Si|#>%=sW|LQEF6|m;($kZeO4?BshZ3`%$r^Ap
zCkM772^}UaY4VG@Tm5d8u+tUX?~Mmymv=X-D?k_O+R#@xJyki;4%R@^LKPMk1IXeS
zl2Kb7%~05S{N9wpsM#sENw~PGPG|T8-LC=eF|a~TEq;%Y7OF#?+xtD)mpZ=k?gUZ3
z=EW~mgpS%=Bm{PSald*&-5BKghOeEf2U9qw=ZMgd#sX4AnT85iYtf$3lv%FX^;dGr
zy#T~EL-4D1#$M2*tYLZ_r&1g$i=q<ENyQVI<WkzLWDB<za$DB<l29;JMz~6r&6yUd
zUn`XmE^sxW!J4hs>fn)hJ$ae1*e&37vJTc$l)=@)mpe3@=dtI3kt1SyhfT|HX7U#u
ze_?IJO{ix5TqJr0ei`;Hu7oR_lO`2v_vhP>nXCuG_Qr9FkwTPJmEGZW&oXQfl*~T*
zOk<a{-M>l18*s*1ykd+4=p2jhI$+j!G@d$ZVw0x{V%~$TIXN+7#t#bt8o=?&khhs<
zbAvqYOnb`nrf>GknTT-@cW~+sM>Y-AraZn0Xb1P2e6K?8O%AqQ60*1o#yCtb$~7<{
zi?<;W&3+adHXr(VBHr-Qd>F4Q=jHnGM#-Pi`jg<pZYKXeW{S#p>iM=nURfj|UJ}cw
zb_(pRU9@I;_Fs7S{=YBr{B;&dA=7hs-`V?01cW}34NRU)w+c1qZ0x`x;87mv94!Db
zv4vxHj?LKK%JTy^lY^AlFe&NvXw|g4ZgW9qdIa|$BpXlN#8cnj9Qu%0_M3!`b-_jQ
zCu!<$5@E}00mGjSHSaqk>?4~V?rQH!OWu*Z^sQd`M(26e_2Fm5e%3+S%Y-%9k75qx
z6Jqi%%93xiFZ&|A=WfNERUNa=>Qr?UKCX|7DmfaJK1-E|@~jXK!g8Cma6c(^x$ruR
zr%UOoEGy#ob3BM!1f-YU7dli`{D^Bw5~E*E!?p2Rx0NbR$HD@5XNnFn{1>Yx_z}l{
z7OUh795BVbnkpG`iu-9cZlqWL>wN^2LCzE@^)<Z1(S1^qeF<tv%cfeXHs<H7WIuJo
z`=1s3qy8KJR{Osf1|g#IB?>iGom+xdpOj$uU+ZD?>l_2~XZ^La2O#JI4*)2aH>o&@
zL8|pCw6~T<``=dZ@pabLY`wU?Vvru9)G%q@_VKAWvOC+OrZvz=Q^{q6#cb0Zb$qmf
z26JIqw`4;j+;OM2W0v7QxGQY<b;Ex^k-;^Tf95-u^&6FK$NXC%uX<IdE9|p%om8pr
zlBd3v&dSLlP{8)w_4$SZKSg1W*|&%wpqX|MYaa(}w}L9e%KPL0=-|En2<2bOsA&1^
zgpw@^>@H?OwWnkYy<qq-Urtu8cJ^$o=WE)f{<P|+&?F^D_~nCqru2COG0LDMi6?$O
zG|yy-R4$NG-@e=0M3cI#9-0nIP1EH7qYn2hOHTNZcQ{qmv!oo)B?yc$sh-PU<&1A(
zpo*I@n04j64^AHT)X~em{x;fJhwL9udz;x-9_hz%E8bGoO<}CFt1uCNXg$-6LE|0u
zu-N=3VhkzWnw<MDMTc{5$mEac;>LIxmR9F=QC$71=yQ6O=297PpN%nj?4g`R)Gv)D
z_qnB73p!`4wX3zT_f2YRwN5ADB;Evt{8@B8i1v_n+l~`ZKNK4!eoQndNd`G<a<^XN
zfUn0NtB3f;NQ9=54xa2u+<YKedCy7frIe#1G@AEFfl%EkT7pJ!#e!gxw(n^+apRVc
z=*OOXe#H4o*ll0<O%=EloR?S-*#$9Lx>22H^;B={;4sdWQ3ouP$)E5_r2NA@?H{!%
zN=!tazsja}C$AlZeYT2byl1d~m^!N{_H&^c8?lX;k~2tLj-1U4-XC9^qaTJvaERFD
zgoG-0`KsZzyW`Yb@p*+0ykp0$2sU}E0d|9wyfS$Apoq8Oy&gSV+u2{+Bjm2_WTq&9
ze6LYZKB9g$eqfopEKNNF!G~)oO&=VizHG`EJ4V(}T$?gc>Yns>?YM1?a*#KdL>pZB
zM<x;kZ0XTcv3&-st2Pdyvj&|@JaK<Og){Dnzg1ToUSdcWikj-^jpninFBTT&ok8O|
z0&8_i3wc3SDmh-G6I-n7DoaU6Od&ELY;R4kf$K8%`E04C|7(fELN`;@O5d^kqGM4e
zBrmy}w>tzs1j`>H{P!LJYikS6s<{m51UhkE;W=MVL8q^a#VL9Bs=uZ5x$4&LarbhH
zA0~n}tJOrA3Itu<wQs&!2-E1Aw?5?zDJ8Q~>N84F4?iWG<j|p;i!$+1U`6K=J_xL@
zLx)bq@5iD?{P$oA`}{_PR<vcmhyJko>U8T$?$#W7=@7<is{~zEQCq8URbv~TW-b01
z)-<Z{;{KI%)abKrUQi-6A1?2(WNQdHdSE6G<cLwx#0=%m?DlbeUX0le-Sig_Ptp`$
z>HNe^H$y?LQ@dvyD<X=@$V3jiz5n&>;j91gV*bznyRuP@4)Mx?$tf<@UzRij4_Epk
zIoLvHi=~4Y4Q9UP-zDT0Ac_1*92^`Znk0tTQyMofB@w?#mPCb8A3AgW>7Oe4o8(6_
z{Sb)&2@x^9mgbMvA;u3sHSTZRX=QEt?*F-qm8|j`!5k#QHQk$4Vj6HDRUa<w`8FKS
z7MDZX#T}^mhq5fjt86DB{CySEamTbh<ht)|K3=#@UHymSBJ32T6{LN!T|-??rDd1K
zf62L=A<+Fxc>9BqD4L%V5|1;A5UBj@-oxC4jLQO;CL;&^sHon|$uQQ9`6JKT3*`6B
z3lct-MRbaPEC>TnvdDwvR_$UH&c9sl27VQ)3eLEuA)WxzjY#-^ll*-=cm6gM5;_tJ
zI{x%m=c6FN5<?oD!k98AH8ZXPZejw%tU|@(vcXAIJF5E>-~Z>l{_S8?@YFeE!8IM#
zSEA~!UUKA5@oWyflbau~QE@6V72yHeFTXm`<@iZNRXZQNe%19tui2K!eQ?6Wy7-&K
z=tsEuwEfr^x)KkqAK$i{pD7S$G<8x1hLJ5j<<N;}BknHOH~7rQx9{o_+zlJm1fiP{
z7bieMZl7%3uidk=;2BX2mx`mAJ$bjKf|-lmcO5-bRKmfNtKbW)k~L&T7vfu8?@~ng
z3|h4LwN%_8k2tpcQt<(&cFW^$V;~-$qcVQMj@%lX+WH?x=dG-(d^k}ZY~9);W}3>0
zTyOj)kuKRO^DXi#yy_0A{nd>m(W<<<&zLpZ{6-+J*Ssd)iSy?n1~WG2#Yi2yJYx`8
z4_AS!<V+DUvR&p*T}(q^scZu4TK6?mC5pbUB5-syf^0gu!l42!2JUgyW(#GRA0u?X
zY)oq1-~SL30K+~us*z#!_xjO#c7A7#5oSe9n9C{(csfbLJ@<5r=|j&2;LT5x(i%VQ
z$}1|PB^#qyk>wfl$-gQp6F7OSnI7U1#19NeW2sDO>E;o*5>N;+Sj9}{YQ_k_^xsxo
zOo+u2o4M9}PXwL#M?&7saT3PjEGcN5PI7xX&U&pK!rtaE8r5YxsHCzyO9U6njoZt1
z#w?1kYl*t`ew_JOCi+Gge+c(*M2>Go+<#!b*&kb%_i0B`9H1e-R9I{7dm^4Uwwz@0
zu1L^@=L|T=TmIK$gUDYn@7eu7A6^zI0AWM*!BvRbw5^GH0$hr_@W^tA4dPP95wo|;
zDx`uo*v(SjRETgn+r4ae+Rgb|79u|qE9dYZI!E=p$3HeMK=y-oYNHienSYb0sGnyN
zu|;vGBi{`>GQRg^h_cSe-uvnKi$P@lBVED0qFL1}OUe&9T<^d7IUMsZ4IkSMC;x|j
z;KCa8Ug{c|boBCn)y6%Eu7^88+74#Ume0Q|8Nw5H!iN6nea1_jrDdNCR+nd)WZF!k
zzXMj&`G2Z=>wqTPzJGYOh=?E|E!`laTM-5u<wi*e(m9ZpHjswF$dS^cdjlj?x?yw+
z5|T<c*SqVv@B0$3=ed5r_j&(&&wsWt7Uyvu=dthk{gld(MR+)CO(bQU;@;rOGXBT`
zPctJ3HJ$8(UuENr$c|6mE>i0nzi8AbrLFqNAvnh;6~tF#9FfOcc!H?_f>Sq&%=Zu0
zCE5X(l-3%`ml7Q!9~ZUhXBbzFipvO!MA7@K$C>D0^T?Mbdv+HSZ;pVheL#xiv80;j
zt`Fx^%(2?IBp41N#Epvv?lSxUaQ(QTwfGz5{$97_!&cWTltn!cC*9zjeD_t*f>g{b
zG6TnC20;ji*$3Ab{(YKUqUmz^ai<p}m7msydB<1T*hy!5nh1W7rkId*CLj`Vj+`uu
zTaw~&8NF5I>y6s2*WsBdiQ+VbZF@#M))gl^;lO;&U>?J;`mRM0hyl2|iJKaT0#s1W
zoIp67O}PST{}Bi?1!|Z{%=e%JwiC=}?(D{#SSg2s-K7!6_)t)0<?;zn3CTnPb4uFK
z1@g9AGj{Azcl%?w1QCj^cci}H{7J_h@d_bQd_F}`r!U}Um!9ogfdn1PvaKMySxrs8
zfZMG&E3v&*JqkNU7H<NnB+xXL7sJdn<?o8PGx4Z7$%8cv{IWQ`z1SmuW9cKkkmYd4
zetJ1JyqW_anO7bb7YqUm&YDcL1Fo91slF`GS?)+v9(L+kJtB==<Yz}OkVi+UmxtQl
zVv-Iu(b=rIJFM8hArKy`)SdVt(hYb=?%1|6u3ANsku8441RowdFoU;0wXd8o&=KCu
zTqs(84e;0fj!80r<uW8a`Vl3n6y{NG(W=2GdWlf3e}WGF#qkzoQZ5(=6$2rGPf=~g
zBnkF{HZS{e1O%eA8X-WYF)S<$5B+s_cb^Tc|Gv8Y`RLlEx!Eh9%c*aV;4kxTXH>*_
zA`%|Vq71rvsUfS{EgjZ^WA$@>A?{5#HZ+7z7M5h=vH@)8&Bwa?LM^_a@Z(GC>e*?g
zDmn{RvmaO?K!~bhL$m%qZ@11^M42FR@w|d}_k^c~wKHh%y)w2)UXC|Ys4~5cN|}_-
zHB%j$P2+U8%U_KWRl|e4K0B7CEYq4$(mv4m$O#)BCP~ufuzPm@oEQ2G1>N*qyhErG
zU8hD#Z$N5*St!Ep+WQ`EJZ?9X^0Sk`dl(UU`~m1DiLy@{B=h|Pfa_NG^?hBvr?@A-
z#Q)&kd2`#2`O&-f2Y&#{1Ww0&jen5Vv8$9P?@NcRKHCtzRTM}c#Rmx|vX7xYwgxf_
zi&--R!IYUb>(5))i}X_J=paQbwYt69yWGz;HPJ1d>gji9|MRW^?_}Q-+^1KnlsD&6
zm2YBZ!@D$Ol(A70Jw8h}Yy7k$oA>PP+R?o|Sw~<?<E`8ceovW1lL`GCpg>|Usn9Z|
z>*xzby^-3WRagu>tGm00uDfrVM%KiK=Y-m%kr{M$sj^wt#-;Hy-r+{6v$hu9#$`Th
zG~QBN^tv+(63>Z)=jeg$vx^|>@(P;?%TJ0_iOXc-e8L5cqB4D7d)-^H92+D*q89y5
zY@cFtjXrn2&_e9nz413}@(CL`-sXy4ZVdWd!XhQ&G<g+3RuXPgKKbaW5EVeiE-)t_
zVX!w*j4WsMX5+Q;gl3e-C4%;Da~!6#$m^p}sVysl0-m>Ylw_)Tz0S$;XchBAz7GI^
zw#2()@aLT%Pj_hcc$}pxU&`8`jsM!yI1`Z-p-gBGiow+hY8T@qICaSqkX=}(F&|Z$
z9}dd><Xq5g^Cev;x^Pard{AiMZZThS1#(Am*zRcC80k8{f<9z`Z;BzmsRwm`cVPX#
zO`}{S7CjaBo?->+lZmg@yk7q~qyMvA%4*THM*GM|Lu12(*A1ip0PuK~1}#!l(`$uy
zwx_vd>dINe+g2q9<LP#(eu7!KcDH|Q?#muQOsl*=c_fU(qB4R=L*(FlmLxsTV$K#;
zFw`pqh5^cgX5NG?C>kbf+pkPr>t<#pdoS_17N6R?*yi&GAna0A%XGOPZIS0x>|#OK
zMCyV*1G73_8h_AMaQiHG!0&p?c^-3%8<jFz!_Ks@d_L5ydVMJIuI~|2seAt*ba20(
z*;vCcBoXbZT5T&Yw$TO&!w8d-f|+HH0KiZG*}4C^WdP!#-+USPfsXsp<E`B(@AkfP
z^bvOY`RwA2RzE!Td|9hL_J^>8t87C5{%rL?uyCC9YQPoscgIPCdu(RGzFH%*1;dj)
zb8S774r<Gd6C{x|a@ibOj!s6bi=ID6YbumsC9sf~7eBK~&$u~HaF>UEdh88;!3(n(
zVyqRkdDU;EJ(uFIKYx4#iZPv>-4%9dI-WQ8dUy83;qVWDe8e9B5`6RH_TG;At%6~G
zrfo^jHN#xRg02p8kv{<H_Tb5Qbp-E;1Kv#Zi^>(Hln2+e{*R7Q3RR-iofWs4`e?rF
zkj;-&Oe(AD<x?;0lC;-EF=A1=)KMK$k*j2>2FMI3Nw5-tu~sYC{27QZ%z+KlV!=dR
z{YXs`A+1gqAi^)8Dh-6iRp_u%ALaT!@A4UPet_4A{O0KSyWQub3nPC?%ho+0<>3~I
zB$aldLK!x%B}-Z$L@puc1q4|RW*xr?>UufJf<_v2NQ#vMp&ada2dW&c2-7h{&@`}3
z0K#=_lOW-N5~jF)vmHU_@0lIpfreiNgMBg>o&P4W1fcg!<@uf0Y$a(eMc;tJ;=;-b
zmX>jJQ8j?Ld2EpHQ=F;vuBFso_5#&ULAxI54alMD$ft0Fu88U>?Wdg-t2Mm%b|MM-
zY;a?$_0jJ_yWnDtV7eJXYyWSf8UpIPMQ*h$$1dr)Yhcwx(r;I-h0>MqhUAp+P&mSr
z<9DGZz~5Zt2=CCz94B;mHj_2;ZLe%XMZ~~7@8Hgd$cZUI=_FQkjaUZLk23tEZud_#
z-+)UiYag+URd%DtpM)f_V)_S*^EjSZRLHPp>LK5bouJ!8ecJ6n??gV8_9$iURi&Cf
z<>xw0NBJrXWH^?4WFseaDXXoaBvG$LzcMAh(Ls;EpT;FG3cExo)CDswERQB<pfsap
zIl@Ld`J(fg!P&hh)NYdvi$@8o3Tz}kA*PsEEX-_)f~)ZD!mVW%U{@YojXFyVIo~Ok
zb-L(-&m1wkT6*CS|8<vIf4z^iWu0@?NeWquv46tf&SeUYjGInj4n|K}vw0QCBL#u{
zkFK~}=8RwEwV;YLsa{x>94|4i7&W3gKD-s#tdLz{B(BVSOlXN|u3+Zv4#j5qnqzCu
z4An!l=M!i|T`*2F4i74xXF85P{h~jOz1tb;aUFXbX%@{4CP{6pvAbG&Rh>P5NVg`A
z)ih)=cDQ=ct9%rxqx5Ffv!|*O91)N*6(0R0z*dl7tr-N;k38!TI!lfgZFYXYI4<qE
zbxlnJpK{06;~LdrmMECmc8~qo3<S3~|Evs+cLq3QO=~<U4_4)yxJN-j+smTJAZHl$
zSt@?;aT@Cum@2i|i^Sek!9DL#O)~A9o)O#*<77~SLfSCKD8|63gWejvr^y3jB3QEa
zVL^RSb|69CQYnc8?(|-Jfic*REf-}4;de!wyQDGx8Hy@rz4)X~KU-|AsW%`*-~8yB
zZ+ib+=}$uNp2z?zI6ks+&78CfPV&-v2(=?DW?AmW?>9r$=Xtod#_++#x6((|xs$-`
zAm0$}dao&9R_?pYkwarj7yP5W=m$WEYSWKJ8!fk&y*{?|spR_o^sW7qzKWcCB}j8#
zedkq{&|r!aJ4X8ZvNovPS$2WbDh>m=l%JnZG<#%@#1V^mj|ktia1!NN*-gP<7D~)`
zl5qWOY1VG=*^_qX@f!Pr?@KyQ+!how<&uEU1*^0okF(4v1Qe2Y-&b$8>mzuxc0}4S
za89{vQvkpMKfNr5P7F_zsQUx3baQRsf2vY^9Cj=2&8DrKq-pwTs#)JldC+0)=pE4N
zx<1Qhe5^Y_qoZoOuok}Vzm62-?VH~9n!kO;b?Rw&4<|H@Jtt6n!l1Xe&bEmvw6pJl
z5cRtgxemWKOMbAj#*39lz`rUFazt<5iSi7h;8oy;TKu7e=pLBH5&f&Kn2d0d7<*?m
zBkCGr+&66WWm{Ldt~w#^ILL6l@x@6Gw+mH#@<hLvLybdO6#WAG1Hj*Y_0u6MH%70O
z_<gDV1F&FmZ3u9S8M^B(+f(;M%Y0Y27#s)%HL+^8V>2XLWEKM1M9L$qlIR~tQQsWy
zA;SAQ8+v%&IRM<Cv)v3|)h~TWd}qQVM;x9Sq*)#;u0IsF=bJaZ8)u{76)0oo>N2hC
z%HZrQyggRWCUJZj79`LgWD^uq`Ick#+eg+ikHF?0$@t7Nl-s(6^w+;m<A1z!)9&p4
zz#16Gv%Z^2gaVot_{s-<&!}GOE$o5l4P{Z~fh;U6ENf7kY;tLD=C7Zhb+MNb-|>u%
z)E+g^#gWH=U+wUKeXowDw`*8O+YB~GRJtpL<LkH}g#Em^L=~shV6<WG+vgCWH6G0n
z-zuc~`n?PW<2st3o(w&KrD{Pw0Y4o+fA<p+^6gFx-BKGiiYYT1WnAw-H<E2=9VwHa
zIk6p@w<X8hvK$hXBq2-=Y5HEI+&v-sI{U8w-0sErq213Fcif)kqsn<d#d!73vS)lh
zXW@t2R{OIivktEUh%4AKPBp4@7I}2`+7XNN+*F^^0%@MQ@I$w-JWp5W*fe6Cq|$k*
z2?o`X<C1L%t@@*_SrojXbHbeAc4HzoXHn!BIOABd4lZ7IGZ%H$0+ArZjd?89{Co*+
zR7rkmYsrN)Ua@U3lg^u=kl~fkI;45N!$h>PG>CYu+qZ*nkIgqt_T(}ED$<tl-KLh1
zACw{`2~qGZtnpbsUDonM%AD!^V7Aels#zfY&<{ceiWII>qT-{d>-#&_ZJ&6SkMM%~
zEz|2KZToSim2TqKBCe%+@R}*+$PR`ZX1h#2F4CurbC*W?_yeam@^02Z{7Zw<>-8zJ
zVwjyNW<p_Y`?;^49agUBe1*Gr2mbbj{|Oq=4SVgSQOD!0=Th)f<m1wl-@f8M9VI~G
zoSxU~l_HHwJ?D4d4vTEJ$$7_IV#7Wgw++Q)tL0%}F^N4Kc$-fBb^rKkxkrq8rg=T9
z8Xh$K#va$X%|oWhS!3Z7>74tsnj96RS^2cVxASgU*s_7t92%K`y#9X!n6GL`;Cy(7
zi5d4*230h8_XG<T42uUrX$$pg(rWR#x8}NXA`IMS-@5PK4LP~6GoaC@d+;e5uJkYY
zLZjT>o8`AoXh5b=alKhN9(E>?orq$W>h@!gtqzuG7{V2A3|c+x1+?LBk<{{qwfGsl
zzIccte;eA1^XwmA5@i#j@xMfwp1=6lxf?QT-5?ufS&hk5i?*4@JP7W2B=2=YFKnI_
zqgtz8s`z=P1DqRWAIlO}e`Q<q46K&(sW1?s!GF~$WqC=D&!*E^TYeRcA#I~)n@AGW
zkv`*u(oeF)as#fqhClgYZi%a`l~E?>E+#T{2XZ0nXYEO+L~vX2nWnmM%t_peAQLM~
zue?AArbgrzR^?`kxLu4Zge`hlB<O&JdVe<JY=0Ks>={a2pV#LUR%`F$V5#dp#qu~k
zGN6-H4rSNBngHap9FgIIa`c&#Y>e$cV14v}q{>Fx7FBo#&M>?ZVInvl()%Rx83=K?
zBmQo9mdIR*#lVoR=rYPa6YL|9s1b{nsP|uKAaX#v-(wjY8t9PcY*D4tCayeAdqT$-
zEfr&{9KjNkf>rlv#{g6x$?rX3?J1)b(b`P49<1rXtFGZ8pu+JIP5+`HMjQ4SzQk9o
z5Z&!bC%l2MtSgWrTfL=MCVCpCWprd<P8J%}<tq+ihdJcFmD~|IB2^#gf`1|{Ljb+0
zmvDRjBTSg8?5w<Hf~(o+eTWj?|1Qmh-$hZF`>G#QU}Wp?s6jb|^IoQk_~9-tloxFM
zMYp0Gm$6b98WPlgz%1~iPiUE7Xtg>9wIoeHf1B{;e%ZJWgvnn2*1}+RbNd1K;Gll{
z;f)g6p>7&w+k&Es^*8xJZX;3F#<kzncv6xWL2-IJq{UKUh+b)=;Y#GsA=-(q8wk*u
zJ%L{gUHyVYVANBbftD9P6>_No0<~+BPZ=_=IqKM00Y=w05hKC-PnWMD2YXn&H4WeK
zo?SOQ+y^j?uZOD`iyf3@`%3ew8ObXg6!6!L3#)k7Ww%cQWnjwPUt>K2YkqvE7oXvc
zTPXRaHedUmpVB^KTRJ>b<+F_({OgC<p|=%=omO!c(W?&*_}KL6&$3?tLY5}1Lv5}F
z2v|o|3Bv?v+<tK15o&wJ(kZ7IljQ%7FGECk^Syoafs#FMS)JR_@sUm?`3S#e2WKR0
zJUKR&^f5`4!~i;MQkMkfr1q!9(xZz%X16Fx<b<*JP~I~)=IO6VEA*UYA2Uk7ui~{6
zm(VZQe&-5P8f{TaRNCijeja2bC(N(mGI2HcL+=o+jZi`_XFSStLV{uf=f%kdE^z|l
zH<<-9KyiK8srsqaY9r+Ue#yR3mKOX6Ae>HT{#f=fU3UK&uCLsXd%N3Vkk~w&2airP
z;?aqRSFeoqHRE$6A_}&A7*3xCTg;HV>Sz>lE!(gMJ;Ru*O$E%e$<PjIDC+0-)taH`
z&&WKL&Te2PLo3Ejm1R0!mHY{Rc#alIC{6aXI%}4Z)opQ1+HHO#jRQ|cMLAFHPhrot
z$tkc}g#}!@tigQh+bf(7c4+u5QcMkkjJ3zMnBG&GV|mu<-5&sp8|QKk%kL)e?Vj;B
zczJfx(Omyb7q#xba*MBS__6VNS$3LPoz7yyY!1Jn@CHKWg!!ky4^ojE<MDP4>{2K4
z3?~m|x9|B$J>+jR-J0x_q}H+(>>>A8%t-8gb8BAh!C=F*$t?VgCjLj{iqWq@5I8K-
z!)39I)T^|R#>6dIy50{8Ju;zkP#aBjbWLT*kiE=dZ}>SEZ(^S{Qn$FpU9AV5x40_u
zpmpmpEWlJKP9JQ2a-$q3FaLaz>X>j^U~^w^OLi+W|I|qF4*(fYv*j3fkaGA^B^M+p
zoQs=HB@8jmFXnrE^7)(!dYaL<Ldf8ew`XLGbU!xIb&HL^c-%hx{#LU_uVkCJWMg#C
z9_$^wI|XJQDWu`<>}aeue4btLF`o#sk62+-&y7b#XrOEeb>Y3{T(l^lUaqPW&GxGm
zV*)nZn!m5fy;*GL^px(mTji^XD}-*T6J)z0X0iE#-(8TtTX<^?s*LN&hExJzusHxg
zPyhPYyuFmsRng4=@NY@=KtWAkE-I;Lv*?di5Mhn9PxT=wFtD;T!Ytl-%t7(KZ^yZ~
zZo<lboQbGvUgt@@N_J74KN@FazQ|b>MMXu$K+fPK{fX{*j161b+#PMfL`$)~EAi?v
zdP5VsWn@bYUUri#^P%X91w$KmnZ+N$H;<&}*+Ml+sBzuzbOi9pE7Z>D`{y?xdEYg_
zm23+f3lk#GgIi~n9SOLA($)NMtbn)qk&Gf*uYMkO=YvgyegFuUf1RN@W6%tkml*aX
zgY#RtBafJWm*A^l$p$qMg(b(`*OFP9>W39S&uH$(rpl&mWxPkyGo3cVa?%J*j8wk;
zWb;T%nMAP%{sBlb?vn7y_pB(9cr!P8ZFs7Hr7gca`LY|-Y?>}Pk^KkHBoy=3@q4KC
z!iPkld)E7GHd|((baa(!$byGV%Ar=hHgW6>PRfu;N3Z7L1`CsNS*$RJzI}ZkdWDm<
zyawH4R6EsuY4`22fy)Tu>HLajLGZRy9+W24b#JI5H`%I7H=P*lB*|^hZIJ<FmCB-Y
zP-_xZcC}Tn%Dt~`o`bqkAoM-q%ronVd22tiv${HhnDN%S4|dY@;C36=$|mL0<@G3A
zI#h3?^6+7-LMThvTTdF%<OPUwc=r60B8qccKGMX;2N0M~i$KsCrn;o5A?lH~wWW07
ztR^1^KYU}``X!x#8Q8gFE+zl&ZV*^62&rsU6UBl|!9Kh(bcrOk;Qm!<_4k=DheHby
zVYTO=qHUeNJc6p9`t>kHC5(<eTe(i%lTcxc*%pR?ja$P<JA2e+n1yFU;)<jqI+ARa
z&1*SQajX(G6=ODXkzTy2ykB&#u-f?=qc+Q=n4)i-Me$sxquW<>8tUOdu{t{dIY)_Q
zG9%wE#k1DUeTdzix@!$uvH(B0;e28C{$C8YKFE@H1#dA4PL@%v>xX2LbyXQ&FQ|IG
zdUXmm9;)X-$;<Y(Sj8L{?Z9}`Lb*iAB`ey0KKM?w<AJ6-ac<fAYu-2Q=aRC6iTG|J
zfh#^I=+fwws8ct;T%nV5(zWx$D=gCtIacTVgUlqejVmVze9H~rysF$CRS`%IiKw0A
zj<p$>H6*8Vl+(9ThOAJ>>?Grgtrx7-FVExA;RCGj@UWzW&vq`q;_>RxauN;oN4XM@
z%&=pZ=Z5?7Gg@ys3uIw+U7NmbCxUMs-L)54yN`8}*=gT!juk*Dnlt5k*v#=hoi~Ql
z@(j%nWKGXdpNWQIHY_=|9ewAy?^hdXgw^D=%^*rXUD1kR=X46uc&0I|Vo(j1*V&+U
z94kbjgQoQ1dE73@lp$5*<dPY9Dg201GQ|x3NGDM;Vev>%To`1tr|>>Nz-19UIr!L(
zmWGrq>E+NQU)^z%h7BG~mez_FiSdeCiWz%xs{nM4uH88J6nHkAPJza|dFfV8s_oet
zSPRld4;_^(-^~^ojcRvyLr`Xdx!w9RcGDFwC;Mk}!oT1^d<2u0;a`y8yI*kNe+4c6
z1ql-W1qVJN)V!wk3lfz63v&Edc;YYE@t;29J<WXJz8qSJPlP|=#m|BY6*u|i2WEH4
zf}iGA^swYGM68AHc89*p%Hm|R5d^g&r`id-esf+E7nQGrt?-BV_<VK>=P4~U${V|*
zHyuOuGoD_a(x9|WGj7CZQWwS2=QdJ);<Zm_1^{FTHyJoQq?6BXve~vNbgS>In$*{O
zHm4oGu4uQwnA0)JDAV;)Up6BT#=@N=k9ZZ<xod6hm?aiZI>*V90h5W9_g(S%a83W8
zJGArAh*fDB#~kS1lgq(;tJlI!)(e<-wWyA~9`q17x&-xSQ#tMgVFUrnkL0-=oYymz
zNg@?mnCHh+-b>6Ppl(|Y&4gF%j`4Uh4fUjQvImV9Rjv<4;N(`cWG$}p8!7-yjeVE~
zLTghTZfnM6s_V18?+8+hTWqoHxmtce`0{GCjgv|RZL<LCC{@{!`Wl^qRxe6?*oPot
zMV(d5LjNf1`2k}NuTWkU&I!k}?Y&-Gu0@Haz!j!KA(iGTF3%m8X@t-zj*_S%wpb@Y
zC>4VvHoMue&Wq&Dn_<oJkXsW=`OJ^p-IG*ACfFi7YXdU%Jk7u|APe)AsY``*LwdCv
zLVlWP&P}HYT)t|v*sD`@&J<HR)#xCOZuH=g9x_8H*GX{K(~Fq)-GiR&mFRJiG?OCP
z=#D~rb64SS`2q6B{jv8*$>c32qHB<$_O1E^%C`l`1HOUGiGVl)B8)#0<{xE;uMUEO
z>OZdo<eh56<vLwP6x77yh5aoAzY4`|jJ+NXbuC8(7lDm`2rNWxgGmuJ3|WY3OKR6E
zFqb96<Ryv`ww4%e)1D+9#+PQiN-zprxRs!WK}`5QvrMabP4G*A(CDEXcicC%xp6ny
z%g0DL8hd#{rIom5e@s>$%PK9iauO$jPO{=uP%9s0?kkxTVOxgltq*Wj9u|8&Axxs5
zfo!#4b_S^5lFrp1*gBRMfFSErAGr_B^xDN{t~xrmI}(l;HFb&tAnZ_x`PG?KSX^9n
z)W(W*J9al!Epe5IKL5p@?zuQcudpkgb{J4NCw>PeR-^<K>+*+yw;!Kn@`i=Aw>Kx)
zI_cfiH59iE9j{@ncE6kUhS4!qgP#?vFT~M5@0E_1HbLE)$J#HIU%D9&Mp_pmQYEv6
zJKHz6R?V8T5MCAy8iv;8cpWyYKDq~_pQngfZp59SYVLgKUzxd@yrT_Y-FoJDKg6x1
zk^5EYtgDhkpk{jwd;Z*m7&HA%#WDcvBdMpAV*b?=!C)v!w!BwVe8ZFWUp3CZ?_C0Y
zcf0Y<OxPnl{G3X`l4oZ<<@#yolHp67He%f%E?PQthB7k0PSSU@g1F&2^a%R%YQY5{
zR62OKV%<9AG@NGlY90s-MjwoY`nJ@BoREyLIT<vLm^Y-ECuG$?=}Yt6_Rv8Jl@@U^
zV&6DTo)lMFuL@<_%Tc`y{<0&=ihMMA@(T;Rl_WG6E`^x4RJMKgX=`L&xz(wq-I(0&
zo&)Edp`XH8(dFS%v6%DlJ_n`mkJXlqo{J#Qx)$--wwnJ8{+dvw*^E%--b1fC5g_=w
zn9}I4-V~r-ZOC+85FCNI^W9Fj{}3}Dh6H8d(_oOXm!d8p$Bc>NIePFbpo;N+mlNu6
zHW2}<9ezO#IzqU0$XqrU%8b7>!OH{|%y{RnkNTl~)6Y8f{Y+idtIoZpZD%y5PiJz=
z&)Rx&9()$j7Yz7}%EHQ^yqd-r?)r@-oZB@%+(DchH=bSl1MsOYFR!AZ9HFjyZ6(^g
zFv&b1x<eUy;%{_*KIcwHHW4x(7fc&VF5Ur>z}R}fB|f>w&=bEz{2#2>KTZ4#eE0n5
z*~a7mT}&Lck0N=KV>$c1_Ry#(d-F!0k6x`NGK|?U@(=kM9_pIy1+W4Km}Y)XRx0(X
z<?};IIHq)}2oo8@Qan%&(<kwZ7Ls`3tp28PB0%?41hL+8CyrXwqam(j?JNp+lD12S
ztvD$KaML{=W%+v4EG>z~t#=95#;fhERLQxbl?7;A5~1j&ZbWSj2L>WewaEig(*49K
z=|Dba?n|kFE`8`dKYJfrV&7|C2x(0+Nt8%cW?!^qT<XhzkM2+bs?`~gdEds&tjgi$
zN#}`O(82P)NB6J%HHWFa5qFxMugW>_w&e2-=<<HN2`+m(gbe4bn8r@BlX}yF4<Li(
zp;H3Yfa^(YO%G285*&AE)e`sek5qdDLdA$6Djpot{X1s!Dv|R(Epe@ebK!=jS$gVd
z%-y1@8!7@ksjfD!zC=4KQ`?q>#`7lWqoE&aNEi~$XRGH5cch0;*?iH5$^}Eoc~gV@
zy{m+`Dx+-~-AJs8$m|<(Sc<$O6TXM?a?bl&v-exD9BrglF5?4{f82?2ce>R`_2mVm
z^;`Rpmx11s(3j!Fx~^XpJx<88@wU?*&xI1*$A+@+Unz5jKOB6n97ypITDe2jnekZn
zBY9|T!cLO0g+6s;d?cRZ<~D^{XNLV`AWC@dDw5RqWfvE6tv(Nu^T88yRog;WL_Whw
z$`s7nhLcz9AP|mgHjPhe9Pvm@`d~=!TA&-V;aWsjFp7QUGuf}qgbkQavSEA%RA~Uk
z#BDNGz#)2CT|NQ)!DqZp#%l}#iT8Em%xw*pTrCZD^Qh{RwoR$hl{E<BU=lsm?@=G(
z`wGEKj|=hb9k*RONs_=QwQeSp;oC-f4}&U-Xx`$?JIxuCCQC>?iU<x!GlBdsZ6|Qk
zKtxc-;J(oEeFw_KTuQbma+1y~t9nO;qqI?j&^$v&RBp|I%(uvrA_r$5_t7wJ@nPCN
z(i~i&99bAh*a9k15vm-LOIzX7nhMF~NRfk{Kef-oNO$$!eZLxkT^$ksx*<A|89z_q
z8B)OJkZZaPC0N<W;oP0ofWTGo&aABQY%xgT?Z+(pz$>*|JyZjyGI4`6%`Uljvd|yF
z8NJtB$({}5vIpl{PQb6mD+i<OGGtoK1b%K1^q)yq?=|Hfi@q}zsb>{f96DSVgGty-
zA|=8_VjZPJon}qP0$FX(#pekjKk{{gA>Dg{st={W%?3K`+LYmjAXJB1N_fl&+2D2!
z@Eu#fetcrZw}w_9t+`pQ+@4+08(n343(7@R+2L$l<(!Ia?UOJs+Q1kKdpvW0o(hG=
zXfIb^ZoaUd|DSU_E2r4WCW5n(6xKjUxMUPVwU^tJmV|8fM0ze}1h{=B-vYFN$1Mco
zGEJO5Omrh{j;Uv27DK;_^uL3V;;7_c24LEo)hT!oa>0)-ZKX%%4}gE|AAknH+TSk+
zkoJf{AwSBMsGuLSyJu~4XO7~L&pdq!ZW~wGX^nrADagU}7u#xTs7CmwqCOnOUvsuQ
z9h^5%caoKCwE1~`xdxN%u3kT)YlvwS4sfo@UbLmIKYrI-4?b5F{sWMrk|aqWG6LJM
z;E0ePor+w+gL+lyp`ymHZn-_nRUMR1-x$8b@)FXlq3xUP5kGZ?*tfhYY7H96FsL@F
zMj5knPr%|uIZg)l4*!>c{T(-n?tm3-)>z~FwE}gCljI3OHglK6Hl_BY-0qLnwFvEY
zK=gVHVr-XEj@)?<pxv(lf1?d*{mqv5!qLNrBh65|G&Rewo9mTPw9e#GMDDh7-ZDYG
z;evFq#G5hF@HW{YGr-O)>+|g$@|l0SQlhlQ5hgOsE?u-O90e=+FIJY!hMqlSW|DFv
zS&b84O2>zqu<i%F@BYmQnF4*eap?4hL@z1|tcZ?WDLV;LE~<IY7L+5EL)jizjfn2?
zknF@?M&&{oueDrPjY8I_y^ssnzb@?}(owyo_k`#{qmzFWsq(q}lNr1%=Y>?Q`Qg7g
zCSTaUf8TAw<mNTJ-q2b!P`<Aob+0VDG(OR=t;R_%sQo00!HMcMAjC$5IciDkA@^4J
z<BDdM=Bi-nmB{<nqb!Qn-pL^uFT+^^?=7L$z1-G)-&yKrV;H%q5IS!R&Cm@WP4e}t
zf*<GP1(FsmHTEd%l2Nh5VwTIXOwt(fX8>tpwfFS$ubkK(4sQ3zHz~hFVpUl?<loAy
zj(tp>c}u@V(_XoJv&+fA_F;!WbsVLbci<wl>s8q3wb8^Ci7bi}szgbA!Pdd1Pel?Z
zT`8^N47IX^O+y9!EAHs=eQgh!Uv5Vh%cx;!nOV{~f%xnpFC+qULUoDf61jHv{P&?<
z>szt1ql7Vmkt`vExilRF=fOW7L=UneyX%^u-1xS8S@`>X%V3jzvORHn@N<1p#`pzt
zfvk3A``)PLA$Vo9^hgGn3Pt(hQtc(Ou04v1+j!dsCs(@l(em-NlCU%HYf2_exbc$Y
z?5MMA6_X~qQE^KOlQ>!6<kWcydkeRt3Q<z6zPN*Eq;By*^%~lkcXKkxT%qoregGH<
zj>-z^Go8X*es-BXdf0|%s*AL&#i?_3%Bg<bwu3wC(OSL5!~{#*6SYlHM|zVmDn8eQ
znVGlUcOlWY7o0RN{$+bTwf@!HT+Z6slY17m?F;3qR6VA_(Puzr5iIa)HoB|kwJ6R$
zFMV7-Cmv1^N_fYTlo^rEuk<xldF4YN@-Qo`4<9Ol70CEFbjnWGw^hWC!RU+g8FY6D
z-|b}X<Rp@R%qJH<lr^bH3iINWw`|8SQ^k$Z(|N`AFN@w^=X5`;Wz9(ZJaXVk0Fi?-
z=lCuQv2w~*O->6m02Fx3Pv15^{r+<7)wQhSBr9ZASG%5Td9WTPs9I=M<tQ;(a9onI
z@yG4udT-A?M8AZro=i^7;?;7mVSY`C;JGG;bQcM4>}>BMW*SS=4^Uv6JJ_z``Y2{8
zQ-x5!RrH<M)C%6n9CJL{0gNlE*&|hG4)wv17~PqzT5tOCCdOCuPuB9!jT6IenL{eY
zMa54PNg*&ASpRr+R#tZ;NZv-)tI-&|a_=}Wa=F53T4$iUIAEh40eUZDR%>Zfz__Q>
z$zu{KzC!3Y9f_sa2QR^+>?UCTgw=pcztz8lK^irqqpDerPDg%wu>iuL$5nnbwVjTg
z?PO2u@(z#acn+XLa$=>;`_Emy(CvtjyP5deKS@Hhe=ibnmFD?*`f=a^C;TjYK>RM;
z{Wx%yVJaKDFDidxwA?gOzbHN1v+Bh^<vsUT!Maau;I~8c2jDyZ`HfN+J#o6|6w?lC
zr^1>72tk7uxp3*X#A92<0UB#gUAuB-d=O;A<jseGXa_lwp-d2MMp+)z-C448r#dgJ
z1GCkoQQ}Q>J_PKWtMs$5aw%O(yW&@MyW$YeFrG-qBNVXX!)r8wuRW8}cFF&&JvqV^
z2k<;Hku4&v_yLO=HaZ2G+p!u<GeK))CrH|aiB9nhJCEz~R?Ip=B6zxr$KrR%scS(U
zcU}%Z7wY10!BD)h+C+Aizn%sQK91wsmepavu&$$Uo|~fVNRPbDI#btS?nKp_Q(?!{
z7^$Gx0Tbo8JoWM~G_HFTt*ad@po$X<ue7Z&yom7EW@Yj98E>lA4axE^-?JC0!nvfl
z<JVA5F50mT->rAlp#(CJDP#uU@{cf;Z;;(4jijL7td&;&IMGWGO-)0Mwp0~r%mO^R
zUFV)OehST+kVXRJ*Il9^_7CwwPI%6<<SFeNS3(uo;smSK9cK}05|Qv+cpn_8Z*NL|
zc6Y#9%*97CDoaeGI!m!v)04Gd{!g3<-O~&bIkwhKjPVYZ+{DvuRbOAa-BGCn+JP0Q
znP*j-Z$ZnUtDU_8JSB`)nDN`LG%`r<p?$TMrT98g?N=5Iu9i?&#}EzmvbK?28`h4u
zgC#me;Q4GtJ*16Qk91hSd5s%hODPAOCK`}B-V&E_&m03Oz6|Kc+3}VgMzdt!6cl1f
zajoJBX0tHU-=EsQr^!gf$cyn$zsUrFA;*G3`{jSXIK8B&|H|y=q@S(G{z21H;@J4J
zb|-@l<dzt*Eb9RoUhYS$0m8FN1)*p>b-24**W`aofB1U4(Z>BfB2+Eb_xh-J`)`o?
zMJ1y21swhn@x_<)7XBQc!$lXYioCIIzJsODMb4M%En`lc<_8=w%b4?{Fop{fzNozD
zc~OqwSRG@S6rW8#Uy<$7#FOyF>d$*7%}#<&v`ad)_gb7yDO$}kSM8BNNLYsdqME_e
zSH5ov@NQg9!m&dg@fQdg^q(ad_h~*i-bg<k8#uXufd^hhy3^+3>7AYzMtv1&u$K<~
z>LC5r<;l9}J73}_M`de%k1E+T?kL1&RMT|Vy@R(-wK3-5^IBV@nH@}-`6h{EFTPVV
zX0)!6AwS8-Z_ZECc6H6P255(6H|1lKa>Lm=HYvNU>IHK$OGn0X?5~5o?(U0<3Nu_z
zPIX!8W3{#hi`P1WZIk)m#BZqmEZefim3*4`f%<JQ-+BE~o0eYWR+xO7UmD&a<)s%{
zm#unInS7Bh@EiBx!OTTDt90Fj<4;2IbCbr?XRrQLwg&!7*vhxgt^SfuINvQ)?WMOL
zd6BPN_QyNOvIHKnV`NO;ut+B?S~vS97~SO)i9EQebd^S_%AH}v&eu&LVt8V-!stc*
zD5@Nu^>J8PRkFq<*^w=kpSh9uU>4gbPLa6FrLXCdUkG+q*q)h>o#q%#=i}(o9905b
z;z<6rLI4y5i`a2JPq!J&dW2elvl<c&3acAJ*5}GY--cFPBa3NktrpwhhwOfjZ<bu7
zguq>iwWmLq=hJd!syfy6u01FztF#cl7hAN;^u>wUL2Pwd8Se{`MvDXsa~wd|<@LjL
zE*`>5|2X&;2@$`|pWj|WfzAf@k<@^QLZ+5#)Zh1Ly;nwzwpW%#FIG3D+$<EyrE`AE
zM^@!1N9M&6u83&{`Me*wu-5><@WPRrz<X$C#|}?KaX=yg%i^=~ZhIaJ0tgGs$XbAs
z+A0jDkFvL6+D#a;iO4psRq%gibNyLQQk84IF@H`iVKfU>?KZU{g+8-h-NC(vmNBgC
zV*u+5CeL)_(pUry*MYjw%GA<UyeX!IS=(+jUru}VZq>aXl&Vss**r(^8Ww!;a4m`u
zmOH=90N%{wDmg6%hA1?fH_~R_%M%^^Om~#z9nAh6@5e;&tPLeY<ZSKv#g;^04~Ns-
z@qddC=3r&9@agU=qpW34jD0JMPFQ<&>OITaLKZxbLwo9Jtq_-0Ima`;WDUi)JzMYv
zU{u1va-y^Us;dD+WrrT&c>S8c*R%D8Y!Xr4_fFQhc{8u9lam>IQ)4u3)q*IyYItOP
z|Ix6#q^(kA>9k`mg<4N!rZTTbuns*)ahbk@-=+PP%^kS;yS7YU?A6B3J2TsKPukJe
zd|dVMVI9s!tQ;8Q47F9@)ak<hE($zt9N58f!r}b3!+HrR!^QtN?))9Uw-<oKZXcBM
zFxhJi*GM0i%b&q?G8{i+mzX1C2wSj6F>tajy@b>?O&#nK)Bd37a?;t#b$l)tABu%~
z(6<icyr^On`S}#G`dGKbTu50Eym!|N@9=`~#gzgSoZ&((RhOIYSGG?i=G_nf=pGwL
zZCXfb&d3E({qi^-O2cy?hXtQx2^U*o$oVz0ZR%^;_4R6-K^e_uQHyuEpCh{b6QV{y
zaSSpRpjGb;>{yqpgB0MZ23|WTKAx(p8$H+0?O1)Mna61$iOFM4WTbanqL9@0ExZ<_
zP>YD-_(;)enHFJr-_1y56-(JsP{jJJB39R~S(m`E9f*{-7|dgyzBPuAAZUEm`tyZW
zKNKb@eXRHFE^TE@-e5chkXkd3EhcNm1*g!{1i2dx!%8EOvt<CmPx8Av6rE+aCZE;V
zr}uT!CpiFHd4gY#8$ZpcZBT7`EM(|Ovk8V_p+B%DEK$BIaSpOiU!px+F+_fWN=R(~
zqQVY`jVT>_+$<|A7m%eOP$>1RKPz?m6KN?vG@H3AVyGH{*&$}+6I!!jWwZiL#YFS9
zf)s10{-qDiwXHOTZPmE52CThZwJq8#<O`uLtHu%KvbO85sY7k2<w0{%o{}VPly?VZ
z6O9LqoOlY$QSm5U%r|>=Yc9DUe2fG&UNE{*8uBpa9MW++_r7=T+Z+F5^b0$E!%x2e
zO+Su`<Hu`1ZXB8`ZimWdG5rvhDOi8-W4EDvL!M!5=ee(d(QcS5j$}>x`tXUte4upn
zJcAdlCz1cnoL4c=gKI(}a=q^P>;E8c1T%cZd>Q|R<KM^a$*OWsS)+~v!&)7;p_PF}
znIEQD@ELd1CmS^m*i0vS2@Pdl!C}4RRI+&MyZch3MqxT<%*4Czetj-!$uC~R!wZ~`
zyZ3IE{ED#=9;Zjx@@XOTOexM8Rog{=roG|37TgTZ%YAC4Yp=}Ai_6>uf!KY2kuOYJ
zHVkpEHh2&AWJ%#UI9t6@_xs40D0>SF+jV(4j*c%7=u(g+xma~qG>2Q=Ev@t{M!ifc
zi<l^?SZ5ZHpA}a2Trs{<FLyoNcH2uf+3@>tJ*OjqW4&tgP%H_xrVcdhPgaDK?+K5!
z<v2h2k%j0==#OwoqrQfgp8{#5`^4k%wudpYzurMQV_4Rl&?tKRbpG(deqQ9ZY~|V>
z9gH<xX|8{_Z?)eToH-y(6`V;<;YtD|plC7Y$u4ReQ-(l-D>LCpFDg4Eh1dvm;c!H4
zu;iNuk$Ht3N95|)F8FRbNI#0B-?Hrda6DAFrA5ECxXKbGW7p0Lw;G3r*Vnx=rXyVi
z09aywTh#yI!6mI%*2gLOXFmlSe`tNEQqY>qU3GOh_=L|ctJi)_sCi?)#!-d^<mJXJ
z9qOoQf^sYPH0)MN_b4euU%*zKX)&u`t0Sxy247&j-Ql`&CkFX48On&WmsyRN+ufIr
zyjJ?Rrbhn7SaQ75^46M86(Rp;@(2pyU_lw)>}hX88cruz0AuMXCP65>vLOFk=jvvf
zP!ba)CfgnifAK7XsCX7i^yDN4bKd$rIY{KAkGwGPCT#MACL^bBS?DbL2d6`Wt;V?T
zlJ_Ej%2j}|uN-mb%W&dYZy|CWtMAxcp5_XQ!)0?H1l7wNWVy&EhQO2K@7HuFxSY)<
z2>?{~Y>9F{JIy{tTPVlgo&zD{%^lgv>_s3l9ZXDoXQ2%<M?o**9SkXjGWF6pc{1R`
zbsOA}_eQZ4gL|1}-nY-xfN8Z@Klaabv|`I&v@t6YmjV~b#O+lV?wvUmoySKhxp!;c
z)MsFug!wcbxKkp22~5*7MZ<HeY`?7u=7_2am+?C9k^Jd@@fR?BL5rGh4kR7Er1eW`
z`u}rWIZqh3qvx2gW#`h!^To7AdY_wnm9Nahc_vAE7)61jDt(b;6;(UBOP=_2Vt5%|
zT9pwK(~3vut(^!6*fnuP9I*YJa85x%rdgjLyalWGCvv}k4h0JaNuqa;T2YF_w{Nk0
zL{kWdmKP!hG%Xj`-w6q~;F#rN%-edI#=0-@Ub>rC-&Y>a1A0tJh6uHABU|&8R+wx(
zxywUTu~NlqXJ-rNb6kM0TcqrdwXJihY_35DBP^&`Y^J{WP^UlB|JDK13zxP_q-0~1
z(2pli6C&-J;{8NAm)8O8cD~zcx#YpyY(?hks3T#vthLK#nk^MGhjLhoa-RxFsKn)r
zg1eL|I(EueJ2VYH=tIgq6u1Ur1qNK(lI?IAuFabkm5EyIJ@RHgtuyf?ty}A$>o>Gu
zogCb@iRkc4(=+gglU9jOxK^pdtyiL0j;Qd>pqJOncc3<Fl>=w654km6wvas0yK~>R
zq;g!tTt(C#WCHPWN*hQJE~UM1U}as(>GnI3_#Ab!37?S4b9J_|n%JSyh<vJhYjT<@
zN>Gv{8?_|}??);-R-T8=+BD%*F+yEREFu%6U<j!~83{f(lj-0_=_QiRRbfKTL0xjO
zC4&)?gud4}eXPk9ZQhYi8MZ3cz`8c1Vh0OEX2BOJGnZ^VL<Am0c{(TPKTD#3>LNto
z2Jb&dRxv8+7FB)dQFVfej)k32#VxUZ0t_0>jTP&MO9q;cof58sZ^CQt@-pY>lafv#
zv!c;2!};v8SPdN+oS13cCgE|oLDlS=a$rt~vI&^-Ls0z}`bYnLQEV$K34|N56@Px~
z$<wMsx8L0z49Da6Ec-vE!rILMm&pm$CkKnzoYjRy8gl8MYw2$Zw3=s+8aI36GOg9(
zmgf7_xRD>Xb$m}!zm58&#UZVU%dc0ADX?4Fd~1w?<OJ+Qoo1S`E-G5~x=gNe97b(s
zs7;<^I01-NIPLlZn!isS1ND@9cHpW7AwgkSNKj1l*x7Y?^UcVEo2j|hQ@pU|lx7~F
zK}2w1mLk$w;Dq&YLM(9A03UdxG!&B<slTzpH;I^*`#3}`HWXpKL~@VK_xY5PTR6B<
zh3rf5%OUk{);z8yxJUj<Ryc$gA928Hx@4FyAF^LyN$4i1tW&JaxB8e0k6lp6gd17B
zisO|@EMe*3&R<**lvdBTFz~O;D}P0O%u5<pQ-~LVgWY8nv)a8^hl%rZ91p<eXX&Og
zQ^YC5`+hFAcUJYDJ>AE5`4y!88w47lY^71mTA8RcVulpvK<cM8SdN8mB*f2H#~g=w
zs;S>#$@uvGvL2(VMBB7;M~a@i+7bpAn{`H;1dDF074d=Wu0RBmn!Lk#Av0C4@bm>u
zSHHDf0`qUk;Yh##;H>Z8%|+9;6?;QW_poHbgBkoZwR#EB<aaRH_<D>g8*hna`1Nfl
z_ZP(IlZdnK89&{d&)l)!ZDs@anAxX5r!TOlMj46@7sRhOE@)z7gDiXso>?TTW#+ZB
zNqyuA@qTspL9=cCarkItpDuW+8u-5UEnYeerbZ&CArP&sJ5j(3F#I*=d1aN#i5r1%
zSjr9cGBnQASO)>z5f<pjj3n?=YsQj_GBYy)ftw(?E%LPB*`g9{-?@^;QYk=hOC?st
z%v6|^#k!Yot@u+!Rgo*nP&UUT7==odM5Wb=`2G@BU$GAI9a`~whu0)V#k#seBQ(1j
z8Y@Dd>tBjhisjSDnpCMoPAluh;#)h@53dLQ&y;~1ShcFl2#qamL0ZS)^-s}EDsF=^
zz|8uO#6_{xNjk~lL*ppYpxCl_Ch(>L8$jb&KtoG0pSnW_tV`N(k8;?DV?b9~mD$tE
zMur1SLa}6FS>~zIZE!gs*Oz~e_lzahJXi3?M;ZR_G{0HAr)E}gVNlAmj=F@J#7r(J
zy<7gEiBCgo{u4txTu>Jnt{XRpKH+4G+14@r`02l&?0@^q|D_gE{o<&KPSj<zh-N_#
z){vY8V&N?sna9ubslDK){QN^I8f+Hh_DU&SBJ5m5g&k}q=|wAQ&l=vrZwfQ=GRCm7
zHb;F54W9fnBkea9)^8OOKrcfuZ7#LJdQQtTd%;lA;koV5_dO2}o}Ag0jSfktm951_
z_`%rz82B~dF5K-n$mZXj--X9jldFtqKc~3av1xJ<KXJT7+(X%CM=!POKyaA@a3@5Z
zCEA9<dTslRGBxLj_@?8$OYQXYV^yTMLyJrbcG7;<4)o}1tx&pLVz#O=FNsLKXS}ev
zUO{)(_H6E^B;}{Rp>Lcg1@7D4Bm+&!ztuTHhTDI3d?hSh8s1gy-vwMU13viA*xoA8
zuM=7c3<@6WtTN#ZuUF$`TnU4bwFq9;j75U^+T}#_1IAh9kxSrlytijC=+CT6V4*X8
zvo~lZPCNvvGpPA_u)70Y`T`{XYMptQ$GPW;6%#~xgM=_aOP66(JvYtkv}-#;jQ8nP
zn-iivuZOL`z;!T>siMHb6A_Pl<@8pG%5_H`IsTlu$>1MDyK&PCAO6>P*njrsw_753
zj%dVmcX3(mhvGE8l$}_EQ*=@7f@nBfv|}ts*rd@t;6e(*6v!Y}^<*_K*$1cVXZ7sZ
zP{zt8a9Bg^B&+<qzbwuEM(^s0Y^|`LD}Uy9FTV=rtt9JlBD_aJldT>MtIC!kr-GNY
z;*qbE(CP}|%U$H}Lj;3XHZ66S!ULg9PdF`X*mXOAOroix%-LDlbRE3NeWoAF^pjd&
zMa*r@4YRl)>K&uh#tqn9MMa&G>q*$U*eb4=O!CD-OM~p8(h)?iVzOzR2I0k3P&=dZ
zXXal2w?=2tv=KRq=Nre=*ts((khl5#!+))b{ZZG4Pwm$@+qfS<<+gLbEx1%nD?)G1
zH{Ba|9p21jlHrg;m~8%+y_?!2tDUUZg{Nv~aL<YZBA>H2a~8MyntzOp7RbIFogJeG
zs5%l)r11ZAaacu)e1ra4uCRc|l@2tuEf1GhOXfUyvS$oblzrs+-Hb|omSSsfclE-p
zq_I=jv37(L9u&Ezm7S(_W_?bdcN%&V590i7X@OEBGHcG^^V9|W89h!a5+Nc~TsaE$
zBqN*$EaC!HG0yU1+qZ-ILB&>ZHV|^BtKYpEApS%1R665&qV*DCR&`%2J9#IXMQUw3
zNnFv00$RIM+^`+2R^VDa7q^tnX#?c#cKK`4|29p2f55Hma66U8VZ&l(=FodkpN5<2
zjxq{p$bM;gp<zB;P2yXK)lgL)$VQk7h2LHS1bi+i;utTs1_6MJV!8isPcio(ZZ%NO
zH%y@$Xm)GxjR>*H@VUJX7@1-XM;K}<4AJoQ%Fu*|VmMoAPv1YZwE^g#T?lc$5sg>>
zk`@wQP!?WjWS0W|^>0$wM|;}Jyb0(<gXgPV$VW`>_h`|IJ{KLO$J}#F)X0&?H+IXJ
zgsE6r8z$G0{4e3JuLeUjykQ>Z^2Yc1?iW4)TqeCR(3d$=tYzYC0wYh@taCWyj#0zg
z{5sZ|>f467^tsA~?_VwsD)LD>e3m@Esc>;99{~PQclfub@Ur{=m)o7&Lq1~pC^Sm*
zTV=U5%zmk^GF{b?m2(fZ@zKJ|?>?aGK+frXiy|e0{ovvAliXXF${3uljyf9Qbzc)t
z(kSwa>FWtBd#`@7I(npXItJ+bcxF_1qFd-qI*T7OV$)k`2g4=XsH3tEF?+2D8$R*a
zIh`G<a;IxCG6fqCS6Hi(&AH8zyzIRRYocBgiu}~dlTOlts1`~1a5Qe#*He4xg{5Cj
zHDYNfU~8C=cKy@6h<FK4RW5GCI~OsY_5+43R`9k)|9-;XRzvE^8k=e^m7%C3Dyg40
zkNt7Xgzfx;k*-2sLNG_-6c!SO<-@!9Ou-J*Mz8065Nfqj#RL_x-sBn!0D~W3m;|l*
zu#<uUNgJ7?MtFCn$a5<IV1Hn(1n;2K?)RsI(!W3IHGmT9Ptz5aSGUI7UPvi|n;+rE
irE0)25Dmw>oo*1)m?%;z#hR6=vw?wM2jj{g<NpupF&}UM

literal 0
HcmV?d00001

diff --git a/wp/references.bib b/wp/references.bib
index 80afb849..f47887fd 100644
--- a/wp/references.bib
+++ b/wp/references.bib
@@ -1,7 +1,5 @@
 %================ Related Works Section ==================
-
 % for white paper
-
 @article{10.1145/69605.2085,
  author = {Basili, Victor R. and Perricone, Barry T.},
  title = {Software Errors and Complexity:  An Empirical Investigation},
@@ -320,7 +318,8 @@ @misc{Sayyad-Shirabad+Menzies:2005
 author = "Sayyad Shirabad, J. and Menzies, T.J.",
 year = "2005",
 isbn_hidden={null},
-url_hidden={https://ieeexplore.ieee.org/document/5989519},}
+url_hidden={https://ieeexplore.ieee.org/document/5989519}
+}
 
 @inproceedings{Better_cross,
 author = {Peters, Fayola and Menzies, Tim and Marcus, Andrian},
@@ -386,7 +385,8 @@ @INPROCEEDINGS{7510216
 keywords={genetic algorithms;neural nets;software fault tolerance;completeness parameter;recall parameter;error rate;PROMISE data repository;project fault datasets;fault prediction model;software module prediction;software fault prediction;genetic programming;neural network;Software;Programming;Artificial neural networks;Computers;Data mining;Number of Software Faults;Software Fault Prediction;Genrtic Programming;Neural Network},
 doi_hidden={10.1109/RAECE.2015.7510216},
 issn_hidden={null},
-month={Feb},}
+month={Feb},
+}
 
 @ARTICLE{1717471,
 author={ {Yuming Zhou} and {Hareton Leung}},
@@ -399,7 +399,8 @@ @ARTICLE{1717471
 keywords={object-oriented programming;regression analysis;software fault tolerance;software metrics;object-oriented design metrics;fault-proneness prediction;object-oriented software system;fault severity;logistic regression method;machine learning method;public domain NASA data set;fault-prone classes;Object oriented modeling;Predictive models;Logistics;Learning systems;NASA;Computer Society;Software systems;Fault detection;Decision making;Programming;Object-oriented;faults;fault-proneness;metrics;prediction;cross validation.},
 doi_hidden={10.1109/TSE.2006.102},
 issn_hidden={2326-3881},
-month={Oct},}
+month={Oct},
+}
 
 @article{SHATNAWI20081868,
 title = "The effectiveness of software metrics in identifying error-prone classes in post-release software evolution process",
@@ -427,7 +428,8 @@ @INPROCEEDINGS{6349519
 keywords={Bayes methods;learning (artificial intelligence);pattern classification;program debugging;software engineering;support vector machines;bug severity;software bug reporting;software development process;bug tracking system;MLT;self-improving software;supervised machine learning technique;software repositories;k-nearest neighbor;naïve Bayes multinomial;support vector machine;J48;RIPPER in;PROMISE repository;NASA;naïve Bayes classifier;Machine learning;Accuracy;Support vector machines;Software;Classification algorithms;Niobium;Text mining;Machine Learning;Supervised Classification;Feature Selection;Bug Severity},
 doi_hidden={10.1109/CONSEG.2012.6349519},
 issn_hidden={null},
-month={Sep.},}
+month={Sep.},
+}
 
 @ARTICLE{1702388,
 author={T. J. {McCabe}},
@@ -440,7 +442,8 @@ @ARTICLE{1702388
 keywords={Basis;complexity measure;control flow;decomposition;graph theory;independence;linear;modularization;programming;reduction;software;testing;Software testing;System testing;Graph theory;Fluid flow measurement;Software measurement;Linear programming;Software engineering;Software systems;Software maintenance;National security;Basis;complexity measure;control flow;decomposition;graph theory;independence;linear;modularization;programming;reduction;software;testing},
 doi_hidden={10.1109/TSE.1976.233837},
 issn_hidden={2326-3881},
-month={Dec},}
+month={Dec},
+}
 
 @ARTICLE{979986,
 author={J. {Bansiya} and C. G. {Davis}},
@@ -453,7 +456,8 @@ @ARTICLE{979986
 keywords={object-oriented programming;software metrics;software quality;data encapsulation;hierarchical model;object-oriented design quality assessment;high-level design quality attributes;behavioral design properties;object-oriented design metrics;design properties;encapsulation;modularity;coupling;cohesion;high-level quality attributes;reusability;quality attributes;expert opinion;commercial object-oriented systems;quality assessment tool;product metrics;Object oriented modeling;Quality assessment},
 doi_hidden={10.1109/32.979986},
 issn_hidden={2326-3881},
-month={Jan},}
+month={Jan},
+}
 
 @article{Incorporating_transitive,
 author = {Al Dallal, Jehad},
@@ -520,7 +524,8 @@ @INPROCEEDINGS{4301106
 keywords={data mining;project management;software fault tolerance;software engineering database;fault module detection;decision making;project manager;data mining algorithm;Software engineering;Fault detection;Data mining;Project management;Engineering management;Databases;Computer science;Decision making;Programming;Software tools},
 doi_hidden={10.1109/EUROMICRO.2007.20},
 issn_hidden={2376-9505},
-month={Aug},}
+month={Aug},
+}
 
 @inproceedings{101049,
 author = {Gray, D. and Bowes, David and Davey, N. and Sun, Yi and Christianson, Bruce},
@@ -553,7 +558,8 @@ @INPROCEEDINGS{8952192
 keywords={Class overlap;Software defect prediction;K Means clustering;Machine learning},
 doi_hidden={10.1109/ASE.2019.00071},
 issn_hidden={1938-4300},
-month={Nov},}
+month={Nov},
+}
 
 @INPROCEEDINGS{8777507,
 author={S. {Agarwal} and S. {Gupta} and R. {Aggarwal} and S. {Maheshwari} and L. {Goel} and S. {Gupta}},
@@ -563,10 +569,10 @@ @INPROCEEDINGS{8777507
 volume={},
 number={},
 pages={1-6},
-keywords={Bayes methods;decision trees;program testing;public domain software;software engineering;statistical analysis;software defect prediction;statistical learning;focused testing;chi-square score function;naïve Bayes;decision tree;k nearest neighbors;gradient boosting;random forest;feature selection;select k best algorithm;software development;open source software projects;statistical model;predictive performance;testing cost;Software;Feature extraction;Decision trees;Testing;Computer bugs;Classification algorithms;Prediction algorithms;machine learning;defect prediction;classification;Select K-Best},
 doi_hidden={10.1109/IoT-SIU.2019.8777507},
 issn_hidden={null},
-month={April},}
+month={April}
+}
 
 @INPROCEEDINGS{1544801,
 author={V. U. B. {Challagulla} and F. B. {Bastani} and {I-Ling Yen} and R. A. {Paul}},
@@ -579,7 +585,8 @@ @INPROCEEDINGS{1544801
 keywords={software quality;safety-critical software;learning (artificial intelligence);regression analysis;belief networks;decision trees;software performance evaluation;machine learning;software defect prediction;real-time software systems;telecontrol systems;robotic systems;mission planning systems;dynamic code synthesis;runtime mission-specific requirements;dynamic dependability assessment;statistical models;stepwise multilinear regression models;multivariate models;artificial neural networks;instance-based reasoning;Bayesian-belief networks;decision trees;rule inductions;software quality;predictor model;consistency-based subset evaluation;telepresence systems;Machine learning;Predictive models;Real time systems;Software systems;Robots;Runtime;Network synthesis;Regression tree analysis;Artificial neural networks;Bayesian methods},
 doi_hidden={10.1109/WORDS.2005.32},
 issn_hidden={2378-573X},
-month={Feb},}
+month={Feb},
+}
 
 @inproceedings{10.1145/1137983.1138012,
 author = {Knab, Patrick and Pinzger, Martin and Bernstein, Abraham},
@@ -609,7 +616,8 @@ @ARTICLE{4027145
 keywords={data mining;learning (artificial intelligence);program diagnostics;program testing;software quality;data mining;static code attributes;defect predictor learning;McCabes versus Halstead;lines of code counts;resource-bound exploration;Data mining;Bayesian methods;Artificial intelligence;Software testing;System testing;Learning systems;Art;Software quality;Software systems;Financial management;Data mining detect prediction;McCabe;Halstead;artifical intelligence;empirical;naive Bayes.},
 doi_hidden={10.1109/TSE.2007.256941},
 issn_hidden={2326-3881},
-month={Jan},}
+month={Jan},
+}
 
 @INPROCEEDINGS{7476673,
 author={C. {Macho} and S. {McIntosh} and M. {Pinzger}},
@@ -622,7 +630,8 @@ @INPROCEEDINGS{7476673
 keywords={Java;random processes;sampling methods;software maintenance;source code (software);source code;software maintenance;Java open source project;random forest classifier;resampling technique;SMOTE;Predictive models;Java;Training;Data mining;Data models;Computational modeling;Maintenance engineering;Build Co-Change Prediction;Software Evolution;Fine-Grained Source Code Changes},
 doi_hidden={10.1109/SANER.2016.22},
 issn_hidden={null},
-month={March},}
+month={March},
+}
 
 @inproceedings{10.1145/2786805.2786813,
 author = {Jing, Xiaoyuan and Wu, Fei and Dong, Xiwei and Qi, Fumin and Xu, Baowen},
@@ -699,7 +708,8 @@ @ARTICLE{1033229
 keywords={neural nets;risk management;principal component analysis;pattern recognition;statistics;software engineering;enhanced neural network technique;software risk analysis;risk categorization;pattern recognition;multivariate statistics;principal component analysis;input data normalization;multicollinearity;cross-normalization;high-risk software modules;Neural networks;Risk analysis;Government;Costs;Principal component analysis;Mathematical model;Predictive models;Programming;Contracts;Pattern recognition},
 doi_hidden={10.1109/TSE.2002.1033229},
 issn_hidden={2326-3881},
-month={Sep.},}
+month={Sep.},
+}
 
 @article{semi_supervised_learning,
 author = {Eng, Softw and Li, Ming and Zhang, Hongyu and wu, Rongxin and Zhou, Zhi-Hua and Zhou, Z.-H and Eng, Autom},
@@ -723,7 +733,8 @@ @INPROCEEDINGS{7965301
 keywords={learning (artificial intelligence);program diagnostics;semisupervised software defect prediction problems;CPDP methods;within-project semisupervised defect prediction;WSDP methods;cross-project semisupervised defect prediction;CSDP methods;machine learning;semisupervised structured dictionary learning;SSDL;Dictionaries;Software engineering;Software;Predictive models;Data models;Training;Data mining;cross-project semi-supervised defect prediction;within-project semi-supervised defect prediction;semi-supervised structured dictionary learning},
 doi_hidden={10.1109/ICSE-C.2017.72},
 issn_hidden={null},
-month={May},}
+month={May},
+}
 
 @Article{Zhang2017,
 author="Zhang, Zhi-Wu
@@ -806,7 +817,8 @@ @INPROCEEDINGS{1383136
 keywords={software fault tolerance;decision trees;software quality;software metrics;fault prone module;software development process;large scale system;random forests;decision tree learning;NASA data sets;logistic regression;discriminant analysis;machine learning;software packages;Robustness;Decision trees;Programming;Fault detection;Fault diagnosis;Predictive models;Large-scale systems;Training data;Voting;NASA},
 doi_hidden={10.1109/ISSRE.2004.35},
 issn_hidden={1071-9458},
-month={Nov},}
+month={Nov},
+}
 
 @INBOOK{5989519,
 author={D. {Huizinga} and A. {Kolawa}},
@@ -821,7 +833,8 @@ @INBOOK{5989519
 issn_hidden={null},
 publisher={IEEE},
 isbn_hidden={null},
-url_hidden={https://ieeexplore.ieee.org/document/5989519},}
+url_hidden={https://ieeexplore.ieee.org/document/5989519},
+}
 
 @misc{vasic2019neural,
     title={Neural Program Repair by Jointly Learning to Localize and Repair},
@@ -886,7 +899,8 @@ @INPROCEEDINGS{5609530
 keywords={program debugging;software metrics;software packages;software quality;common bug prediction finding;effort aware model;software quality assurance;bug prediction literature;process metrics;package level prediction;Measurement;Predictive models;Computer bugs;Radio frequency;Mathematical model;Regression tree analysis;Computational modeling},
 doi_hidden={10.1109/ICSM.2010.5609530},
 issn_hidden={1063-6773},
-month={Sep.},}
+month={Sep.},
+}
 
 @article{10.1007/s10462-017-9563-5,
 author = {Rathore, Santosh S. and Kumar, Sandeep},
@@ -918,7 +932,8 @@ @ARTICLE{859533
 keywords={software maintenance;software metrics;software fault tolerance;management of change;fault incidence;software change history;change management data;change history;fault potential;code decay;metrics;statistical analysis;History;Predictive models;Software systems;Aging;Time measurement;Software measurement;Length measurement;Software development management;Statistical analysis;Degradation},
 doi_hidden={10.1109/32.859533},
 issn_hidden={2326-3881},
-month={July},}
+month={July},
+}
 
 @Article{Turhan2009,
 author="Turhan, Burak
@@ -958,7 +973,8 @@ @INPROCEEDINGS{8616596
 keywords={computer science education;learning (artificial intelligence);program compilers;program debugging;program diagnostics;recurrent neural nets;syntax errors;logical errors;bug detection methods;feedback system;online judge system;programming problems;accumulates numerous lines;solution source code;short-term memory networks;LSTM networks;bugs;solution codes;IDEs support;Computer bugs;Training;Programming profession;Probability;Debugging},
 doi_hidden={10.1109/SMC.2018.00599},
 issn_hidden={1062-922X},
-month={Oct},}
+month={Oct},
+}
 
 @inproceedings{Hinton2001UnsupervisedLF,
   title={Unsupervised Learning: Foundations of Neural Computation--A Review},
@@ -1236,3 +1252,364 @@ @article{10.1023/A:1010933404324
 numpages = {28},
 keywords = {ensemble, classification, regression}
 }
+
+@inproceedings{10.1145/3194164.3194186,
+author = {Campbell, G. Ann},
+title = {Cognitive Complexity: An Overview and Evaluation},
+year = {2018},
+isbn = {9781450357135},
+publisher = {Association for Computing Machinery},
+address = {New York, NY, USA},
+url = {https://doi.org/10.1145/3194164.3194186},
+doi = {10.1145/3194164.3194186},
+booktitle = {Proceedings of the 2018 International Conference on Technical Debt},
+pages = {57–58},
+numpages = {2},
+location = {Gothenburg, Sweden},
+series = {TechDebt ’18}
+}
+
+@article{10.1145/3133925,
+author = {Chae, Kwonsoo and Oh, Hakjoo and Heo, Kihong and Yang, Hongseok},
+title = {Automatically Generating Features for Learning Program Analysis Heuristics for C-like Languages},
+year = {2017},
+issue_date = {October 2017},
+publisher = {Association for Computing Machinery},
+address = {New York, NY, USA},
+volume = {1},
+number = {OOPSLA},
+url = {https://doi.org/10.1145/3133925},
+doi = {10.1145/3133925},
+journal = {Proc. ACM Program. Lang.},
+month = oct,
+articleno = {101},
+numpages = {25},
+keywords = {Automatic feature generation, Data-driven program analysis}
+}
+
+@inproceedings{Learning_a_Strategy_for_Adapting,
+author = {Oh, Hakjoo and Yang, Hongseok and Yi, Kwangkeun},
+title = {Learning a Strategy for Adapting a Program Analysis via Bayesian Optimisation},
+year = {2015},
+isbn = {9781450336895},
+publisher = {Association for Computing Machinery},
+address = {New York, NY, USA},
+url = {https://doi.org/10.1145/2814270.2814309},
+doi = {10.1145/2814270.2814309},
+booktitle = {Proceedings of the 2015 ACM SIGPLAN International Conference on Object-Oriented Programming, Systems, Languages, and Applications},
+pages = {572–588},
+numpages = {17},
+keywords = {Bayesian Optimization, rogram Analysis},
+location = {Pittsburgh, PA, USA},
+series = {OOPSLA 2015}
+}
+
+@ARTICLE{Software_quality_the_elusive_target,
+  author={B. {Kitchenham} and S. L. {Pfleeger}},
+  journal={IEEE Software}, 
+  title={Software quality: the elusive target [special issues section]}, 
+  year={1996},
+  volume={13},
+  number={1},
+  pages={12-21},
+}
+
+@BOOK{The_Economics_of_Software_Quality,
+  TITLE = {The Economics of Software Quality},
+  AUTHOR = {Jones, Capers and Oliver, Bonsignour},
+  YEAR = {2012},
+  PUBLISHER = {Addison-Wisley},
+}
+
+@inproceedings{static_code_analysis_for_security,
+author = {Kulenovic, Melina and Donko, Dzenana},
+year = {2014},
+month = {05},
+pages = {},
+title = {A survey of static code analysis methods for security vulnerabilities detection},
+doi = {10.1109/MIPRO.2014.6859783}
+}
+
+@inproceedings{Ilyas2016StaticCA,
+  title={Static Code Analysis: A Systematic Literature Review and an Industrial Survey},
+  author={Bilal Ilyas and Islam Elkhalifa},
+  year={2016}
+}
+
+@inproceedings{cognitive_complexity,
+author = {Campbell, G. Ann},
+title = {Cognitive Complexity: An Overview and Evaluation},
+year = {2018},
+isbn = {9781450357135},
+publisher = {Association for Computing Machinery},
+address = {New York, NY, USA},
+url = {https://doi.org/10.1145/3194164.3194186},
+doi = {10.1145/3194164.3194186},
+booktitle = {Proceedings of the 2018 International Conference on Technical Debt},
+pages = {57–58},
+numpages = {2},
+location = {Gothenburg, Sweden},
+series = {TechDebt ’18}
+}
+
+@article{Software_metric_for_Java,
+author = {Balmas, Françoise and Bergel, Alexandre and Denier, Simon and Ducasse, Stéphane and Laval, Jannik and Mordal-Manet, Karine and Abdeen, Hani and Bellingard, Fabrice},
+year = {2010},
+month = {03},
+pages = {},
+title = {Software metric for Java and C++ practices (Workpackage 1.1)}
+}
+
+@article{Code_Smells_and_Software_Quality,
+author = {Kaur, Amandeep},
+year = {2019},
+month = {07},
+pages = {},
+title = {A Systematic Literature Review on Empirical Analysis of the Relationship Between Code Smells and Software Quality Attributes},
+journal = {Archives of Computational Methods in Engineering},
+doi = {10.1007/s11831-019-09348-6}
+}
+
+@inproceedings{Overview_Static_Code_Analysis_in_Software_Development,
+  title={An overview on the Static Code Analysis approach in Software Development},
+  author={Ivo Vieira Gomes and Pedro Morgado and Tiago Gomes and Rodrigo Bossini Tavares Moreira},
+  year={2009},
+  booktitle={ }
+}
+
+@inproceedings{Youness2013ComparativeSO,
+  title={Comparative Study of Software Quality Models},
+  author={Boukouchi Youness and M. Abdel-Aziz and BENLAHMER Habib and Moutachaouik Hicham},
+  year={2013}
+}
+
+@article{Chidamber1994AMS,
+  title={A Metrics Suite for Object Oriented Design},
+  author={Shyam R. Chidamber and Chris F. Kemerer},
+  journal={IEEE Trans. Software Eng.},
+  year={1994},
+  volume={20},
+  pages={476-493}
+}
+
+@article{Iacob_Constantinescu,
+ author = {Iacob,  I.  M. and  Constantinescu,  R.},
+ title = {Testing:  First  Step Towards  Software  Quality},
+ year = {2008},
+ publisher = {Journal  of  Applied  Quantitative Methods},
+ address = {},
+ booktitle = {},
+ pages = {},
+ numpages = {},
+ location = {},
+ series = {}
+}
+
+@article{Liu_Kane,
+ author = {Liu, X. F. and Kane, G. and  Bambroo, M.},
+ title = {An intelligent early warning  system  for  software  quality  improvement  and  project management},
+ year = {2006},
+ publisher = {Journal of Systems and Software},
+ address = {},
+ booktitle = {},
+ pages = {},
+ numpages = {},
+ location = {},
+ series = {}
+}
+
+@article{Liu2006AnIE,
+  title={An intelligent early warning system for software quality improvement and project management},
+  author={Xiaoqing Frank Liu and Gautam Kane and Monu Bambroo},
+  journal={J. Syst. Softw.},
+  year={2006},
+  volume={79},
+  pages={1552-1564}
+}
+
+@article{Farhan,
+ author = {Farhan Alebebisat and Zaid Alhalhouli and Tamara E Alshabatat},
+ title = {Review of Literature on Software Quality},
+ year = {2018},
+ address = {},
+ booktitle = {},
+ pages = {},
+ numpages = {},
+ location = {},
+ series = {}
+}
+
+@inproceedings{Mohammadi2013AnAO,
+  title={An Analysis of Software Quality Attributes and Their Contribution to Trustworthiness},
+  author={Nazila Gol Mohammadi and Sachar Paulus and Mohamed Bishr and Andreas Metzger and Holger K{\"o}nnecke and Sandro Hartenstein and Klaus Pohl},
+  booktitle={CLOSER},
+  year={2013}
+}
+
+@article{McCabe1976ACM,
+  title={A Complexity Measure},
+  author={Thomas J. McCabe},
+  journal={IEEE Transactions on Software Engineering},
+  year={1976},
+  volume={SE-2},
+  pages={308-320}
+}
+
+@article{Cognitive,
+ author = { G. Ann Campbell},
+ title = {COGNITIVE COMPLEXITY. A new way of measuring understandability},
+ year = {2018},
+ address = {},
+ booktitle = {},
+ pages = {},
+ url_hidden = {https://www.sonarsource.com/docs/CognitiveComplexity.pdf},
+ numpages = {},
+ location = {},
+ series = {}
+}
+
+@inproceedings{Fil2015ACO,
+  title={A Catalogue of Thresholds for Object-Oriented Software Metrics},
+  author={Tarc{\'i}sio G. S. Fil{\'o} and Mariza Andrade da Silva Bigonha},
+  year={2015}
+}
+
+@article{Shatnawi2010FindingSM,
+  title={Finding software metrics threshold values using ROC curves},
+  author={Raed Shatnawi and Wei Li and James Swain and Tim Newman},
+  journal={J. Softw. Maintenance Res. Pract.},
+  year={2010},
+  volume={22},
+  pages={1-16}
+}
+
+@article{Ouni2011MaintainabilityDD,
+  title={Maintainability defects detection and correction: a multi-objective approach},
+  author={Ali Ouni and Marouane Kessentini and Houari A. Sahraoui and Mounir Boukadoum},
+  journal={Automated Software Engineering},
+  year={2011},
+  volume={20},
+  pages={47-79}
+}
+
+@inproceedings{Fowler1999RefactoringI,
+  title={Refactoring - Improving the Design of Existing Code},
+  author={Martin Fowler},
+  booktitle={Addison Wesley object technology series},
+  year={1999}
+}
+
+@article{Din2012AntipatternsDA,
+  title={Antipatterns detection approaches in Object-Oriented Design: A literature review},
+  author={Jamilah Din and Anas Bassam Al-Badareen and Yah Yusmadi Jusoh},
+  journal={2012 7th International Conference on Computing and Convergence Technology (ICCCT)},
+  year={2012},
+  pages={926-931}
+}
+
+@article{Kreimer2005AdaptiveDO,
+  title={Adaptive Detection of Design Flaws},
+  author={Jochen Kreimer},
+  journal={Electron. Notes Theor. Comput. Sci.},
+  year={2005},
+  volume={141},
+  pages={117-136}
+}
+
+@article{Palomba2015MiningVH,
+  title={Mining Version Histories for Detecting Code Smells},
+  author={Fabio Palomba and Gabriele Bavota and Massimiliano Di Penta and Rocco Oliveto and Denys Poshyvanyk and Andrea De Lucia},
+  journal={IEEE Transactions on Software Engineering},
+  year={2015},
+  volume={41},
+  pages={462-489}
+}
+
+@article{Liu2019DeepLB,
+  title={Deep Learning Based Code Smell Detection},
+  author={Hui Liu and Jiahao Jin and Zhifeng Xu and Yifan Bu and Yanzhen Zou and Lu Zhang},
+  journal={IEEE Transactions on Software Engineering},
+  year={2019},
+  pages={1-1}
+}
+
+@article{Emden2002JavaQA,
+  title={Java quality assurance by detecting code smells},
+  author={Eva Van Emden and Leon Moonen},
+  journal={Ninth Working Conference on Reverse Engineering, 2002. Proceedings.},
+  year={2002},
+  pages={97-106}
+}
+
+@article{Reeshti2019MeasuringCS,
+  title={Measuring Code Smells and Anti-Patterns},
+  author={Reeshti and Rajni Sehgal and Renuka Nagpal and Deepti Mehrotra},
+  journal={2019 4th International Conference on Information Systems and Computer Networks (ISCON)},
+  year={2019},
+  pages={311-314}
+}
+
+@inproceedings{Friedman2001GreedyFA,
+  title={Greedy function approximation: A gradient boosting machine.},
+  author={Jerome H. Friedman},
+  year={2001}
+}
+
+@article{Dorogush2018CatBoostGB,
+  title={CatBoost: gradient boosting with categorical features support},
+  author={Anna Veronika Dorogush and Vasily Ershov and Andrey Gulin},
+  journal={ArXiv},
+  year={2018},
+  volume={abs/1810.11363}
+}
+
+@inproceedings{Kessentini2019UnderstandingTC,
+  title={Understanding the Correlation between Code Smells And Software Bugs},
+  author={Marouane Kessentini},
+  year={2019}
+}
+
+@inproceedings{MurphyHill2010AnIA,
+  title={An interactive ambient visualization for code smells},
+  author={Emerson R. Murphy-Hill and Andrew P. Black},
+  booktitle={SOFTVIS '10},
+  year={2010}
+}
+
+@article{Vaucher2009TrackingDS,
+  title={Tracking Design Smells: Lessons from a Study of God Classes},
+  author={St{\'e}phane Vaucher and Foutse Khomh and Naouel Moha and Yann-Ga{\"e}l Gu{\'e}h{\'e}neuc},
+  journal={2009 16th Working Conference on Reverse Engineering},
+  year={2009},
+  pages={145-154}
+}
+
+@inproceedings{Broy2006DemystifyingM,
+  title={Demystifying maintainability},
+  author={Manfred Broy and Florian Dei{\ss}enb{\"o}ck and Markus Pizka},
+  booktitle={WoSQ '06},
+  year={2006}
+}
+
+@book{gamma1995design,
+  title={Design patterns: elements of reusable object-oriented software},
+  author={Gamma, Erich},
+  year={1995},
+  publisher={Pearson Education India}
+}
+
+@book{mcconnell2004code,
+  title={Code complete},
+  author={McConnell, Steve},
+  year={2004},
+  publisher={Pearson Education}
+}
+
+@inproceedings{LenatFeigenbaum1987,
+author = {Lenat, Douglas and Feigenbaum, Edward},
+year = {1987},
+month = {01},
+pages = {1173-1182},
+title = {On the Thresholds of Knowledge.},
+doi = {10.1109/AIIA.1988.13308}
+}
\ No newline at end of file
diff --git a/wp/sections/appendix.tex b/wp/sections/appendix.tex
new file mode 100644
index 00000000..2b821497
--- /dev/null
+++ b/wp/sections/appendix.tex
@@ -0,0 +1,425 @@
+\subsection*{Patterns Dictionary}
+
+\begin{itemize}
+	\item \pattern{Assert in code}{P1}
+	{If there is an assert statement in code block, and name of class doesn't end with Test, it is considered a pattern.}
+	{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	void foo(String x) {
+	assert x != null; // here
+}
+\end{lstlisting}
+
+	\item \pattern{Setter}{P2}{The method's name starts with set, then goes the name of the attribute. There are attributes assigning in the method. Also, asserts are ignored.}
+	{\it Examples:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private String title;
+	void setTitle(String) {
+		this.title = t;
+	}
+}
+\end{lstlisting}
+
+\begin{lstlisting}[language=Java]
+class Book {
+	private String title;
+	public void setIsDiscrete() {
+		assert !isDiscrete;
+		assert !x; //ignore it
+		this.isDiscrete = isDiscrete;
+	}
+}
+\end{lstlisting}
+
+\begin{lstlisting}[language=Java]
+class Book {
+private String isDiscrete;
+
+	public void setIsDiscrete(String isDiscretem, boolean x) {
+		assert !isDiscrete;
+		assert !x; //ignore it
+		this.isDiscrete = isDiscrete;
+	}
+}
+\end{lstlisting}
+
+\begin{lstlisting}[language=Java]
+class Book {
+	private String title;
+	
+	@Override
+	synchronized public void setConf(Configuration conf) {
+		this.conf = conf;
+		this.randomDevPath = conf.get(
+		HADOOP_SECURITY_SECURE_RANDOM_DEVICE_FILE_PATH_KEY,
+		HADOOP_SECURITY_SECURE_RANDOM_DEVICE_FILE_PATH_DEFAULT);
+		close(); \\ some minor changes also do not affect, it is still Setter pattern
+	}
+}
+\end{lstlisting}
+	
+	\item \pattern{Empty Rethrow}{P3}{We throw the same exception as it was caught.}
+	{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	void foo() {
+		try {
+			File.readAllBytes();
+		} catch (IOException e) {
+			// maybe something else here
+			throw e; // here!
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{ErClass}{P4}{ If a class name is one of the following (or ends with this word), it's the pattern:
+		
+		Manager, Controller, Router, Dispatcher, Printer, Writer, Reader, Parser, Generator, Renderer, Listener, Producer, Holder, Interceptor.}
+	
+	\item \pattern{Force type casting}{P5}{The force type casting considered as a pattern.}
+	{\it Example:}
+\begin{lstlisting}[language=Java]
+// casting to int is 
+public int square (int n) {
+	return (int) java.lang.Math.pow(n,2);
+}
+\end{lstlisting}
+
+	\item \pattern{If return if detection}{P6}{If there is a return in if condition, it's a pattern.}
+	{\it Example:}
+	\begin{lstlisting}[language=Java]
+class T1 {
+	public void main(int x) {
+		if (x < 0) {
+			return;
+		} else {
+			System.out.println("X is positive or zero");
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Implements Multi}{P7}{If a class implements more than 1 interface it's a pattern.}
+	{\it Examples:}
+	\begin{lstlisting}[language=Java]
+public class AnimatableSplitDimensionPathValue implements AnimatableValue<PointF, PointF> {
+	private final AnimatableFloatValue animatableXDimension;
+	private final AnimatableFloatValue animatableYDimension;
+	
+	public AnimatableSplitDimensionPathValue(
+	AnimatableFloatValue animatableXDimension,
+	AnimatableFloatValue animatableYDimension) {
+		this.animatableXDimension = animatableXDimension;
+		this.animatableYDimension = animatableYDimension;
+	}
+}
+\end{lstlisting}
+\begin{lstlisting}[language=Java]
+public class a implements A, B {
+}
+\end{lstlisting}
+
+	\item \pattern{Using \texttt{instanceof} operator}{P8}{ Using of \texttt{instanceof} operator considered as pattern.}
+{\it Examples:}
+	\begin{lstlisting}[language=Java]
+public static void main(String[] args) {
+	Child obj = new Child();
+	if (obj instanceof String)
+		System.out.println("obj is instance of Child");
+}
+\end{lstlisting}
+\begin{lstlisting}[language=Java]
+class Test
+{
+	public static void main(String[] args)
+	{
+		Child cobj = new Child();
+		System.out.println(b.getClass().isInstance(c));
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Many primary ctors}{P9}{If there is more than one primary constructors in a class, it is considered a pattern.}
+	{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+
+	private final int a;
+	Book(int x) { // first primary ctor
+		this.a = x;
+	}
+	Book() { // second
+		this.a = 0;
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Usage of method chaining more than one time}{P10}{If we use more than one method chaining invocation.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+// here we use method chaining 4 times
+public void start() {
+	MyObject.Start()
+	.SpecifySomeParameter()
+	.SpecifySomeOtherParameter()
+	.Execute();
+}
+\end{lstlisting}
+
+	\item \pattern{Multiple Try}{P11}{Once we see more than one try in a single method, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void bar() {
+		try {
+			// some code
+		} catch (IOException ex) {
+			// do something
+		}
+			// some other code
+		try {  // here!
+			// some code
+		} catch (IOException ex) {
+			// do something
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Non final attributes}{P12}{Once we see a mutable attribute (without final modifier), it's considered a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private int id;
+	// something else
+}
+\end{lstlisting}
+
+	\item \pattern{Null checks}{P13}{If we check that something equals (or not equals) null (except in constructor) it is considered a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	private String z;
+	void x() {
+		if (this.z == null) { // here!
+			throw new RuntimeException("oops");
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Partial synchronized}{P14}{Here, the synchronized block doesn't include all statements of the method. Something stays out of the block.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private int a;
+	void foo() {
+		synchronized (this.a) {
+			this.a = 2;
+		}
+		this.a = 1; // here!
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Redundant catch}{P15}{Here, the method \texttt{foo()} throws IOException, but we catch it inside the method.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	void foo() throws IOException {
+		try {
+			Files.readAllBytes();
+		} catch (IOException e) { // here
+			// do something
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Return null}{P16}{When we return null, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	String foo() {
+		return null;
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{String concatenation using \texttt{+} operator}{P17}{Any usage string concatenation using \texttt{+} operator is considered as pattern match.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+public void start() {
+	// this line is match the pattern
+	System.out.println("test" + str1 + "34234" + str2);
+	list = new ArrayList<>();
+	for (int i = 0; i < 10; i++)
+		list.add(Boolean.FALSE);
+}
+\end{lstlisting}
+
+	\item \pattern{Override method calls parent method}{P18}{If we call parent method from override class method it is considered as the pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+@Override
+public void method1() {
+	System.out.println("subclass method1");
+	super.method1();
+}
+\end{lstlisting}
+
+	\item \pattern{Class constructor except \texttt{this} contains other code}{P19}{The first constructor has this() and some other statements. This is the ``hybrid constructor'' pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private int id;
+	Book() {
+		this(1);
+		int a = 1; // here
+	}
+	Book(int i) {
+		this.id = I;
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Line distance between variable declaration and first usage greater then threshold}{P20\_5, P20\_7, P20\_11}{If line distance between variable declaration and first usage exceeds some threshold we consider it as the pattern. We calculate only non-empty lines. P20\_5 means that distance is 5.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+// variable a declared and used with 2 lines distance
+static void myMethod() { 
+	string path1 = '/tmp/test1';
+	int a = 4;
+	
+	string path2 = '/tmp/test2';
+	string path3 = '/tmp/test3';
+	a = a + 4;
+}
+\end{lstlisting}
+
+	\item \pattern{Variable is declared in the middle of the method body}{P21}{All variable we need have to be declared at the beginning of its scope. If variable declared inside the scope following after logical blocks we consider that this is the pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+// The declaration of variable list is match pattern.
+static void myMethod2() { 
+	int b = 4;
+	b = b + 6;
+	List<Integer> list = new List<Integer>();
+}
+\end{lstlisting}
+
+	\item \pattern{Array as argument}{P22}{If we pass an array as an argument, it's a pattern. It's better to use objects, instead of arrays.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void bar(int[] x) {
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Joined Validation}{P23}{Once you see a validation (if with a single throw inside) and its condition contains more than one condition joined with OR -- it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	void print(int x, int y) {
+		if (x == 1 || y == 1) { // here!
+			throw new Exception("Oops");
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Class declaration must always be \texttt{final}}{P24}{Once you see a non \texttt{final} method, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private static void foo() {
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Private static method}{P25}{Once you see a \texttt{private static} method, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private static void foo() {
+		//something
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Public static method}{P26}{Once you see a \texttt{public static} method, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	private static void foo() {
+		//something
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Var siblings}{P27}{Here fileSize and fileDate are ``siblings'' because they both have file as first part of their compound names. It's better to rename them to size and date.\\
+		file and fileSize are NOT siblings.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void bar() {
+		int fileSize = 10;
+		Date fileDate = new Date();
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Assign null}{P28}{Once we see \texttt{= null}, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void bar() {
+		String a = null; // here
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Multiple \texttt{while} pattern}{P29}{Once you see two or more \texttt{while} statements in a method body, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Book {
+	void foo() {
+		while (true) {
+		}
+		// something
+		while (true) {
+		}
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Protected method}{P30}{Once we find a protected method in a class, it's a pattern.}
+	
+	\item \pattern{Send \texttt{null}}{P31}{Once we see that \texttt{null} is being given as an argument to some method, it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void bar() {
+		FileUtils.doIt(null); // here
+	}
+}
+\end{lstlisting}
+
+	\item \pattern{Nested loop}{P32}{Once we find a loop (\texttt{for} / \texttt{while}) inside another loop it's a pattern.}
+{\it Example:}
+\begin{lstlisting}[language=Java]
+class Foo {
+	void foo() {
+		white (true) {
+			for (;;) { // here
+			}  
+		}
+	}
+}
+\end{lstlisting}
+
+\end{itemize}
\ No newline at end of file
diff --git a/wp/sections/available_data.tex b/wp/sections/available_data.tex
deleted file mode 100644
index 7f1eb922..00000000
--- a/wp/sections/available_data.tex
+++ /dev/null
@@ -1,36 +0,0 @@
-We considered available data in some open sources 
-related to the defect detection problem. Some datasets 
-have only code samples and labels, while others have 
-calculated some code metrics.
-
-ETH JavaScript Dataset\footnote{\url{https://www.sri.inf.ethz.ch/js150}}
-includes 150,000 JavaScript code samples represented as AST trees. 
-The dataset has been used in a binary classification task. 
-Target positive examples were extracted from the code 
-and negative examples were generated using 
-Negative Sampling techique. The dataset 
-was used by~\citet{DeepBugs} in an attempt to find the following defects: 
-swapped arguments, wrong binary operator, and wrong binary operand.
-
-NASA Dataset\footnote{\url{http://promise.site.uottawa.ca/SERepository/datasets-page.html}}
-includes 20 small-sized datasets (20 to 2000 samples in each one) of NASA projects. 
-It does not contain code itself but only code metrics. The data 
-was previously used for defect prediction, cost estimation, successful reuse, and requirements tracing.
- 
-Defects4j Dataset\footnote{\url{https://github.com/rjust/defects4j/tree/additional-projects-4}}
-includes code samples from different projects, 
-where each one has 52 defects of different 
-type. The datasets was built by scrapping bug reports and commits 
-that fixed them. This dataset does not have labels.
-
-BugLocalization Dataset:\footnote{\url{https://github.com/yanxiao6/BugLocalization-dataset/tree/master/dataset}}
-has 18,500 Java defect report samples from five GitHub-hosted projects. 
-Each sample consists of defect description, a commit to fix the defect 
-and a link to GitHub source code file.
-
-CodeSearchNet dataset:\footnote{\url{https://github.com/github/CodeSearchNet}}
-is a large dataset of over two million rows of code 
-collected from open source libraries. The dataset contains 
-Python, Javascript, Ruby, Go, Java, and PHP languages. 
-Each sample consists of a function code and a comment related 
-to the function. The dataset is built for a semantic code search task.
diff --git a/wp/sections/conclusion.tex b/wp/sections/conclusion.tex
new file mode 100644
index 00000000..e9b59f5a
--- /dev/null
+++ b/wp/sections/conclusion.tex
@@ -0,0 +1,28 @@
+Aibolit is a recommender system that helps improve the quality of Java classes. 
+The recommendations are learned from OSS Java projects using ML methods.
+Aibolit provides ranked recommendations for each specific Java class, 
+which differs Aibolit from others style checkers and makes it unique.
+
+Aibolit is an extendable system, allowing anyone to add new patterns and to 
+increase the training dataset and thus improve the precision and usefulness 
+of recommendations. Aibolit can also be used as a framework for analysis of 
+patterns and to decide whether any pattern, however subjective it is, is an anti- or a pro-pattern
+with respect to a particular quality metric. As a complementary result, 
+we contribute a 100K+ dataset of patterns and metrics calculated for Java classes.
+
+The first version of Aibolit is relatively simple and there is room
+for improvement. If the anti-pattern has found, we recommend to fix all instances 
+of the pattern in the code. Instead, we may consider each specific occurrence of the pattern. 
+We may exploit its relative position in the structure of the code, rather than just count 
+the frequency. Moreover, Aibolit inspects each Java class independently. But 
+we might consider the relations between classes in the future. Furthermore, 
+Aibolit's prediction model relies on patterns only. In order to improve the model, 
+we have to think about additional features, for example, information about 
+project domain or used frameworks.
+
+Aibolit is a firm step toward the next generation of tools to control
+and improve software quality. It is a complementary tool for 
+product owners who already use tools to manage software quality.
+
+
+
diff --git a/wp/sections/considered.tex b/wp/sections/considered.tex
deleted file mode 100644
index 087aad36..00000000
--- a/wp/sections/considered.tex
+++ /dev/null
@@ -1,126 +0,0 @@
-Even though detecting defects is an important task in any software project,
-existing studies differ in their definition of the parameters and objectives
-of this task,
-partially due to the ambiguity of the definition of the software defect.
-Among others, there are many methods of defect detection, which utilize ML. 
-Below we classify them from a few different perspectives, 
-in order to justify the selection made later in this document.
-
-\subsection{By the Origin of Data}
-
-As suggested by~\citet{10.1007/s10462-017-9563-5}
-the model can be built using the data 
-\begin{enumerate*}[label=\arabic*)]
-\item from the same release of a software module (\emph{intra-release}), 
-\item from a few different releases of the same module (\emph{inter-release}), or 
-\item from different software modules (\emph{cross-project prediction}). 
-\end{enumerate*}
-It seems that this classification was not possible earlier,
-when analytical data was kept proprietary and was not available
-for the community of researchers, only until NASA released a large 
-evidential dataset in 2004~\citep{Sayyad-Shirabad+Menzies:2005}.
-
-It it interesting to mention that
-in early 2000s most authors assumed that ``distributions'' in both 
-training and testing datasets were similar~\citep{Turhan2009} (\emph{intra-release}), which
-mislead them because the assumption was wrong.
-
-Later,~\citet{5609530} introduced ``ensemble techniques,'' which made the analysis
-possible in situations when software modules do not have similar 
-distributions in their training and testing datasets (\emph{inter-release}).
-
-Recently,~\citet{Better_cross} demonstrated that defect detection may work across 
-different software modules, using their historical data (\emph{cross-project release}).
-
-\subsection{By the Output}
-
-ML-based defect detection methods can be classified by the type of the output they produce:
-
-First, the output could be \emph{binary}, which means that 
-software modules (like classes, methods, packages, etc.) are classified
-either as faulty or non-faulty ones, as done by~\citet{Gokhale97regressiontree, Menzies04assessingpredictors}.
-
-Second, \emph{defect density} or the \emph{number of defects} 
-may be reported per module, demonstrating the degree of their defectiveness,
-as suggested by~\citet{Predicting_Fault, JANES20063711, 7510216}.
-
-Third, \emph{defect severity prediction} may be reported per module, demonstrating
-the impact of defects found in the module to the end-user experience,
-as suggested by~\citet{7510216, SHATNAWI20081868, 1717471}.
-
-Fourth, the output could contain information about a specific location
-of the defect inside the source code (the line of code with a defect), 
-as suggested by~\citet{vasic2019neural}.
-
-\subsection{By the Input}
-
-\textbf{Software metrics} can be used as an input. 
-They can be obtained from the source code,
-and can belong to one of the following categories:
-
-\begin{itemize}
-\item \emph{Product metrics} are generally used to check 
-whether software follows~\citep{InternationalStandardOrganization}, 
-as it was mentioned  by~\citet{10.1007/s10462-017-9563-5}, 
-and can be classified as suggested by~\citet{10.5555/540137}: 
-\begin{enumerate*}[label=\arabic*)]
-\item \emph{Traditional} metrics may include size, 
-system complexity~\citep{1702388}, and others;
-\item \emph{Object-oriented} metrics may include coupling, 
-cohesion and others specifically related to object-oriented 
-methodology, as suggested by~\citet{979986, Incorporating_transitive};
-\item \emph{Dynamic} metrics are gathered from a running  
-program and demonstrate the behavior of a software component
-during its execution~\citep{MITCHELL20064}.
-\end{enumerate*}
-
-\item \emph{Process metrics} such as number of modules changed for a bug-fix,
- work products delivered, and so on may also be used
- as an input~\citep{The_IT_Measurement_Compendium, 10.1109/ISSRE.2010.25}.
-\end{itemize}
-
-\textbf{AST} can also be used
-as an input, which is a represention of the syntax of a 
-software code snippet as a tree-like structure. 
-E.g.,~\citet{6676914} computes the set of AST changes between two source
-code files using algorithm, published by~\citet{4339230}.
-
-\subsection{By the ML Model}
-
-There are plenty of studies for bug detection problem 
-in which ML methods are used. ML methods 
-for solving defect detection problem can be classified into the following groups: 
-
-\begin{enumerate}[label=\arabic*)]
-\item \emph{Unsupervised learning} is a learning model which uses unlabeled data.
-The goal of unsupervised methods is to uncover unknown patterns in the data. 
-Unsupervised learning for solving defect detection problem includes such algorithms 
-as K-Means clustering cleaning approach~\citep{8952192}, KNN~\citep{8777507}.
-
-\item \emph{Supervised learning} uses labeled data to produces a function, 
-which maps unseen data to labels. Supervised methods for defect detection includes
-\citep{10.1145/1137983.1138012}, Naive Bayes~\citep{4027145}, Random forest 
-\citep{1383136, 7476673}, nearest neighbor 
-\citep{10.1145/2786805.2786813}, SVM~\citep{ELISH2008649, 10_1007}, 
-neural network~\citep{THWIN2005147THWIN2005147},~\citep{1033229, 10.1145/3360588} and ensemble techniques~\citep{Ensemble_Techniques}.
-Some authors tried to combine different machine learning techniques 
-(Linear Regression, Neural Network for continuous goal field, Naive Bayes, etc.) 
-with statistical techniques, such as PCA~\citep{1544801}.
-
-\item \emph{Semi-supervised learning} uses both labeled and unlabeled data and
-is a combination of supervised and unsupervised methods. 
-It can help unsupervised methods to obtain better-defined clusters,
-as explained by~\citet{semi_supervised_learning, 7965301, Zhang2017}.
-
-\item \emph{Deep learning} is a class of ML algorithms, which extracts
-features from lower layers in order to use them in higher ones.
-Textual information (for example, English bug reports
-or source files comments and names of variables) can be used in deep learning. 
-Using this information, deep learning 
-can achieve better performance than many state-of-the-art approaches.
-For example,~\citet{8616596} proposed a different method based 
-on Long Short-Term Memory (LSTM) networks 
-and~\citet{XIAO201917} suggested to use 
-convolutional Neural Network together 
-with word-embedding and feature-detecting techniques. 
-\end{enumerate}
diff --git a/wp/sections/empirical_results.tex b/wp/sections/empirical_results.tex
deleted file mode 100644
index 0536dd4b..00000000
--- a/wp/sections/empirical_results.tex
+++ /dev/null
@@ -1,13 +0,0 @@
-In our first experiment we collected dataset of Java classes and calculated 
-metrics and patterns for them. We calculated the readability using open-source
-tool. Then we estimate the impact features to readability and conclude that
-\emph{ncss\_method\_avg}, \emph{halstead\_volume}, and \emph{right\_spaces\_var} are
-most important features. The more details you can find in \emph{Report 1}.
-
-In second experiment we figured out non-zero correlation between the 
-non-structural patterns and the code complexity metrics like \emph{Cyclomatic Complexity}
-and \emph{NPath}. We observe that \emph{var\_in\_the\_middle} shows  
-highest impact on complexity. We explain it by the fact that the more complex 
-programs will  tend to have more variable and some of them declared in the middle
-of the method body. The more details you can find in \emph{Report 2}.
-
diff --git a/wp/sections/how_aibolit_works.tex b/wp/sections/how_aibolit_works.tex
new file mode 100644
index 00000000..940f3607
--- /dev/null
+++ b/wp/sections/how_aibolit_works.tex
@@ -0,0 +1,235 @@
+\subsection{The Idea}
+
+The main purpose of Aibolit is to help developers identify  patterns in their
+code that may cause maintainability issues. From the user perspective, it works
+by outputting a list of patterns recommended to remove, given a Java class. The
+Aibolit engine is comprised of two parts: an ML regression model and a
+recommendation algorithm. The regression model  predicts maintainability of any
+Java class. The recommendation algorithm uses the regression model to decide
+which pattern is better to avoid by considering different various modifications
+of the input Java class. 
+
+\begin{figure}[t]
+\includegraphics[width=13cm]{how_it_works_diagram_5.jpg}
+\centering
+\vspace{1 cm}
+\caption{How Aibolit works: (1) Inspect the source code for patterns. 
+(2) Count the pattern occurrences and put them in a vector representation. 
+Compute the maintainability metric value for the vector.
+(3) Consider changes to the vector representation by subtracting pattern counts. 
+Predict the corresponding maintainbility metric score.
+(4) Rank all the alternative vectors wrt to how much they improve the original 
+maintainability score. Recommend changes that gave the most improvement.}
+\label{fig:aibolit_graphic}
+\end{figure}
+
+Figure~\ref{fig:aibolit_graphic} represents the Aibolit recommendation proceduce on high level.
+In the remaining subsections we provide a more detailed
+description of each of Aibolit's components.
+
+\subsection{Patterns \& Quality metrics} 
+\label{sec:aibolit_patterns_metrics}
+
+\subsubsection*{Patterns}
+
+As discussed above (Section~\ref{sec:related}), 
+software engineering researchers and practitioners often associate
+good and bad code design with specific patterns. We follow this tradition and
+build a predictive system to reason about observed patterns in code in terms of their effect on quality. 
+In the current release of Aibolit, the model is built on top of 34
+commonly used manually designed patterns as input features. See Appendix
+for the complete list and detailed descriptions. Note that users of Aibolit
+can arbitrarily extend the model by implementing and integrating their patterns
+of choice.
+
+\subsubsection*{Metrics}
+
+The ultimate goal behind Aibolit's approach is to
+learn to identify maintainability-affecting patterns in code and recommend them
+to the user. However, as discussed above, maintainability quantification is
+still an open problem, and most of the metrics proposed so far typically
+describe only a narrow aspect of software maintainability. We recognize it as
+the major challenge of our approach and plan to research this problem in the
+future.
+
+In the current release of Aibolit, we use Cognitive Complexity \citep{10.1145/3194164.3194186} as 
+the maintainability metric. We refer to it as the maintainability metric or just
+metric in the remainder of the text. 
+
+\subsection{Maintainability prediction model}  
+
+\label{sec:maint_pred_model}
+
+\subsubsection*{Dataset}  
+
+\label{sec:dataset}
+
+To train our prediction model, we mined training data
+from Github open source repositories. We chose repositories written in Java
+as the main language. We filtered out all non-Java files and all software
+testing files. To make sure our data is
+representative of good software engineering standards, we only extracted
+repositories with at least 100 starts and at least six collaborators.
+
+Aibolit is currently designed to do predictions and recommendations at class
+level. For simplicity, we only consider files that contain exactly one 
+non-abstract Java class.
+%We chose this level of granularity because it is intuitive and because we
+%wanted to narrow down our scope for the first stage of development. Thus one
+%datapoint in our dataset is one Java class. 
+We filtered out classes with fewer than 50 and more than 300 lines of code.
+The resulting filtered dataset consists of 124 repositories and
+29,065 classes. Before filtering, we split the dataset into a test and train sets randomy by files with 
+approximately 0.7:0.3 ratio. The filtered train set contains 20,049 classes, 
+the filtered test contains 9,016.
+%More detailed statistics can be found in Table~\todo. -- keep this for research paper 
+The complete list of mined repositories and the train/test split is provided in Aibolit
+project folder. 
+%(\todo link to file with repository URLs).
+
+\subsubsection*{Feature and target preprocessing} 
+
+Each Java class gets
+associated with a vector of numerical features. We use scaled pattern counts as
+features. For each pattern from the fixed set
+(see Appendix), we count its occurrences in the
+class and divide it by the number of non-commented lines in the class (NCSS).
+The target value for each datapoint is the maintainability metric value for the
+class (Section~\ref{sec:aibolit_patterns_metrics}), also divided by the NCSS.
+%(\ref{eq:feat_vector}) and (\ref{eq:targ_val}) summarize feature and target
+%value computation (where $f_p^C$ is the feature value for pattern $p$ and code
+%$C$, $t^C$ is the target value for code $C$	). 
+Table~\ref{tab:features_example}
+gives an illustration of how a dataset looks like after the this procedure.
+
+
+%\vspace{0.2in}
+%
+%\begin{minipage}{.4\linewidth} \begin{equation} \label{eq:feat_vector} f_p^C =
+%\frac{\textit{count}(p, C)}{\textit{NCSS}(C)}\end{equation} \end{minipage}%
+%\begin{minipage}{.55\linewidth} \begin{equation} \label{eq:targ_val} t^C =
+%\frac{\textit{Metric}(C)}{\textit{NCSS}(C)} \end{equation} \end{minipage}
+%
+%\vspace{0.2in}
+
+
+We apply NCSS scaling because we observed that our maintainability metric
+(Cognitive Complexity) is highly correlated with code size. The scaling
+stimulates the model to find more implicit  dependencies between patterns and
+complexity.
+
+\begin{table}[H] \begin{center} \begin{tabular}{|r|rrrr|r|} \hline \textbf{class
+id} & \textbf{P16} & \textbf{P11} &  \textbf{P13} & \dots & \textbf{CogC}
+(target metric) \\ \hline \hline class 1 & 0.008695 & 0.         & 0.026086 &
+\dots&  0.417391 \\ class 2 & 0.              & 0.05      & 0.116667 & \dots& 
+0.466667 \\  class 3 & 0.009909 & 0.         & 0.009909 & \dots & 0.732673 \\
+\hline \end{tabular} \end{center} \caption{Example of a training dataset with
+preprocessed feature values. \textbf{P16}: {\em Return null}, \textbf{P11}: {\em
+Multiple Try},  \textbf{P13}: {\em Null checks}).} \label{tab:features_example}
+\end{table}
+
+
+
+\subsubsection*{Training} 
+We train a gradient boosting regression model \citep{Friedman2001GreedyFA}. 
+We use the implementation of CatBoost \citep{Dorogush2018CatBoostGB} with the RMSE loss function.
+For hyperparameter selection, we do a 3-fold crossvalidation.
+
+
+\subsection{Recommendation algorithm} 
+\label{sec:recommendation_algorithm}
+Our recommendation algorithm ranks
+patterns observed in the user's source class according to their individual
+impact on the code's maintainability metric value. It then outputs a pattern
+with the \textit{most negative impact} as a recommendation to the user to remove
+it from their code. 
+
+For each pattern $p$ we compute the \textbf{impact factor} $I_{\textit{neg}}(p, C)$
+on code $C$, which is intended to capture the \textit{negative
+influence} of $p$ on $C$. It is the difference between the quality metric value
+of the original code $C$ and the version of $C$ where the count of $p$ has been
+decreased (Eq.~\ref{eq:impact_factor}): 
+
+\begin{equation} \label{eq:impact_factor} I_{\textit{neg}}(p_i, C) = M(F(C)) -
+M(F_{p_{i} - 1}(C)), \end{equation}
+
+where $M$ is a quality metric, $F(C)$ is the feature vector $\langle f^C_{p_1},
+..., f^C_{p_n} \rangle$, $F_{p_{i} - 1}(C)$ is the feature vector with the count
+of $p_i$ decreased by $1$.%: $\langle f^C_{p_1}, ..., f^C_{p_i - 1}, ...,
+%f^C_{p_n} \rangle$. With $f^C_{p_i - 1}$ computed as:\footnote{We chose to
+%subtract 1, but in principle it is a hyperparameter of our model and we plan to
+%experiment with other values.}
+
+%\begin{equation} \label{eq:feature_count_min_1} f_{p_i - 1}^C =
+%\frac{\textit{count}(p_i, C) - 1}{\textit{NCSS}(C)} \end{equation}
+
+Under the ``lower metric is better'' convention (i.e., lower value of quality metric
+means better quality), lower values of $I_{\textit{neg}}$ correspond to patterns
+that contribute more to the deterioration of the code's quality. We rank pattern
+according to their $I_{\textit{neg}}$ and output patterns with lowest values as
+recommendations.
+
+Note that at the moment of recommendation we do not observe code with a
+decreased pattern count, so we cannot compute the maintainability metric
+directly. This is why we resort to a predictive maintainability model
+(Section~\ref{sec:maint_pred_model}), which helps estimate maintainability of a
+hypothetical code.
+
+
+
+\textbf{Algorithm}~\ref{fig:recsys_alg} summarizes how we do recommendations.
+For each pattern from the set of patterns used at training, we
+precompute feature values and compute the metric value of the source code
+(lines~\ref{line:init_F}-\ref{line:compute_m_source}). Then we compute the
+impact factor $I_p$ of each pattern $p$ on the source code maintainability
+(lines~\ref{line:init_I}-\ref{line:impact}). Under ``lower is better''
+convention of maintainability metric, low values of $I_p$ indicate that removal
+of $p$ lead to improvement of the metric score. We collect the $K$ most
+negatively impacting patterns and output them as recommendations.  Thus, we
+ pick patterns $p$ for which $I_p$ are the highest
+(lines~\ref{line:topK}-\ref{line:return}).
+
+\begin{algorithm}[t]
+\caption{Aibolit recommendation algorithm} 
+\hspace{\algorithmicindent} 
+\textbf{Input:}  $\mathsf{M}$: pretrained
+maintainability model); $C$: class source code; \\ 
+\hspace{\algorithmicindent} $P$: array of patterns used for training $\mathsf{M}$\\ 
+
+\begin{algorithmic}[1]
+\State $F = [ ]$ \label{line:init_F} 
+\For{\texttt{i = 1, $|P|$}}
+\State $F[i] =\frac{\textit{count}(P[i], C)}{\textit{NCSS}(C)}$% (Eq.~\ref{eq:feat_vector})
+\EndFor \State $M_{\textit{observed}} = \mathsf{M}(F)$ \label{line:compute_m_source}
+\State $I = [ ]$ \label{line:init_I}
+\For{\texttt{i = 1, $|P|$}}
+
+\State $F^{\prime} = F$
+\State $F^{\prime}[i] = F[i] - \frac{1}{\textit{NCSS}(C)}$ \label{line:f_prime} %(Eq.~\ref{eq:feature_count_min_1}) 
+\State $I[i] = M_{\textit{observed}} - \mathsf{M}(F^{\prime})$ \label{line:impact} % (Eq.~\ref{eq:impact_factor})
+\EndFor
+\State $I_{\textit{worst}} = \texttt{topK}_{i \in [1,...,|P|]} (-I[i])$ \label{line:topK}
+\State \textbf{return} $\{P[i]~|~i \in I_{\textit{worst}}\}$ \label{line:return}
+
+\end{algorithmic} 
+\centering
+\label{fig:recsys_alg} 
+\end{algorithm}
+
+ 
+% \subsection{How to customize Aibolit} label{sec:customizing_aibolit}
+
+% By design, Aibolit is easily adjustable and extendable. It gives an end user
+% the opportunity to adapt the tool to their own requirements and preferences.
+% Aibolit's core mechanism is ML-driven, therefore, as the user adds new
+% patterns to the system, there is no need to manually specify how the pattern
+% should be used by the tool. The interactions between patterns are discovered
+% automatically by the learning algorithm.
+
+% In order to modify the set of patterns, the user should provide an
+% implementation of a pattern extractor for source code file and modify the
+% configuration file (\verb|aibolit/config.py|) accordingly. See full
+% instructions in \todo (shouldn't we add them in README?)
+
+% In order to change the quality metric for training the prediction model, ???
+% \todo.
diff --git a/wp/sections/implementation.tex b/wp/sections/implementation.tex
deleted file mode 100644
index e69de29b..00000000
diff --git a/wp/sections/introduction.tex b/wp/sections/introduction.tex
index c9505f08..d3643efa 100644
--- a/wp/sections/introduction.tex
+++ b/wp/sections/introduction.tex
@@ -1,39 +1,74 @@
-% What is quality of code and why it's important?
 
-% What is static analysis?
+Insufficient software quality may result in increased development costs and 
+negatively affect customer satisfaction ~\citep{The_Economics_of_Software_Quality}. 
+\textit{Static code analysis} develops techniques to help detect software quality 
+issues prior to program execution. It has its practical application in various developer's tools. There are both open source\footnote{PMD: \url{http://pmd.sourceforge.net/}, Rubocop: \url{https://github.com/rubocop-hq/rubocop}, 
+PHPCS: \url{https://github.com/squizlabs/PHP_CodeSniffer} 
+FindSecBugs: \url{https://find-sec-bugs.github.io/}, ESLint: \url{https://eslint.org/}, Checkstyle: \url{https://checkstyle.sourceforge.io/}.} 
+(PMD, Rubocop, PHPCS, FindSecBugs, ESLint, Checkstyle, to name a few) and commercial\footnote{IBM Security AppScan: \url{https://www.hcltechsw.com/wps/portal/products/appscan}, 
+PVS-Studio: \url{https://www.viva64.com/en/pvs-studio/}, 
+SonarQube: \url{https://www.sonarqube.org/}, 
+Parasoft: \url{https://www.parasoft.com/}} 
+(IBM Security AppScan, PVS-Studio, SonarQube, Parasoft) static analyzers 
+on the market.
 
-% What is ML?
+Static code analysis can be applied to improve an \textit{internal} and an
+\textit{external} quality of software \citep{Ilyas2016StaticCA}. External
+quality is related to defects encountered by the end user of the software
+product. Within internal quality, two important subcategories are
+\textit{functional quality} and \textit{maintainability}. Functional quality is
+about code correctness and compliance with the functional software
+specifications \citep{Farhan}. Code maintainability is about how easy it is to
+analyze, modify and adapt given software \citep{Mohammadi2013AnAO}.
 
-% What is the objective of using ML for defect detection?
+Functional quality aspects are typically quite susceptible to formal definition
+and quantification.
+% (\todo: examples!!). 
+Functional quality is also an essential
+requirement in any domain of software development. On the other hand,
+maintainability is a lot less straightforward to formally specify or quantify.
+%\todo: refs. 
+Also, in certain applications it appears less important than
+functional correctness, although in business domain it is recognized as an
+essential property.
+% (\todo: ref). 
+As a result, there are currently a lot more
+research and practical tools addressing functional quality aspects of code than
+maintainability \citep{Overview_Static_Code_Analysis_in_Software_Development}.
+Another aspect of static analysis tools that may have hindered their application
+to maintainability, is that they are predominantly rule-based. Since there has
+not yet been a consensus on how to formalize maintainability, it is challenging
+to devise a set of formal rules to detect it.
 
-% What is the structure of this document?
+We designed our new tool Aibolit to help developers identify patterns in their
+code  that may cause maintainability issues. It is a next generation static
+analysis tool that uses a machine learning (ML) model as an underlying quality
+prediction mechanism. From the perspective of ML, our product is a recommender
+system. For a given class file, it gives suggestions to the developer to alter
+their code. The recommendations come in the form of \textit{code patterns} that
+are detected in the code and advised to be removed.
+  
+Our choice to design Aibolit as a ML-based system alleviates some important
+shortcomings of rule-based static analyzers. By design, ML algorithms capture
+statistical relations in the external world (data). Therefore, they can be a
+good way to model imprecisely and subjectively defined properties of code, such
+as its maintainability. Moreover, rule-based system are known to not scale well
+to the diversity of empirically observed cases, and they tend to get very hard
+to extend and maintain \citep{LenatFeigenbaum1987}. The ML 
+approach does not require
+manual system adaptation as new observations or new features (patterns) come
+along. In fact, Aibolit provides an easy way for developers to integrate a code
+pattern of their liking into the recommender system and to analyze the pattern`s
+impact on code quality. 
 
-Last years we see that the number of papers, which are trying to apply ML to source 
-code analysis, is growing. The researchers, encouraged by the success 
-of ML, deep learning, and Natural Language Processing (NLP), 
-are trying to adopt similar approaches to the code analysis.
-In our new tool we make an attempt to automatically detect defects
-in a software source code using ML and existing source code metrics,
-such as Cyclomatic Complexity, cohesion, Number of Incoming Invocations, 
-and others.
 
-In Section~\ref{sec:related} we analyze existing tools and methods
-of static analysis and identify the most important issues currently
-present in modern instruments.
 
-In Section~\ref{sec:categories} we classify existing ML-based
-static analysis ideas by the type of input they expect, by the output
-they produce, and some other qualities.
 
-In Section~\ref{sec:data} we make an attempt to summarize the situation
-with data availability on the market and give a few examples of the
-most notable datasets freely available for researchers.
 
-In Section~\ref{sec:criteria} we identify the most important criteria
-for the selection of the method and the tool for ML-based static analysis.
 
-In Section~\ref{sec:method} we introduce our own method of detecting
-defects in the source code using ML and software code metrics.
+% This document explains how Aibolit works  and what makes it novel and
+% different from other static analyzers (Section~\ref{sec:method}). 
+% We further show how Aibolit can be extended
+% with custom patterns and metrics. We discuss the current shortcomings and the way 
+% they can be addressed in future work (Section~\ref{sec:conclusion}).
 
-In Section~\ref{sec:risks} we identify a number of risks we forecast
-for the project and expect to mitigate.
\ No newline at end of file
diff --git a/wp/sections/motivation.tex b/wp/sections/motivation.tex
new file mode 100644
index 00000000..8b137891
--- /dev/null
+++ b/wp/sections/motivation.tex
@@ -0,0 +1 @@
+
diff --git a/wp/sections/notes.tex b/wp/sections/notes.tex
deleted file mode 100644
index 2734b441..00000000
--- a/wp/sections/notes.tex
+++ /dev/null
@@ -1,39 +0,0 @@
-Code readability is closely related to defect detection problem. There are plenty of 
-studies where the problem of code readability is considered 
-\citep{8651396, xxx66666444, 10.1145/1985441.1985454}. At the moment, it is 
-still the issue since the problem is too subjective. 
-
-It is interesting that \citet{10.1109/ICPC.2019.00014} tried to evaluate 
-the models for code readbility problem. The authors extracted 548 commits 
-from 63 engineered Java software projects. The authors identified a commit as a readability commit
-when the message of the commit directly indicated it. E.g., 
-the authors searched such words of the commit as ``readable'',
-``readability'', ``easier to read'', ``understand'', etc.
-Several state of the art readability models were reviewed by \citet{10.1109/ICPC.2019.00014} 
-and it was discovered, that the models failed to capture readability improvements. 
-The authors also suggested several metrics which were not considered in the reviewed state-of-art papers.
-The authors believe that those metrics for code readability can be used to detect readabilty 
-changes more efficiently.
-
-It is important to notice that code readabilty also can have a correlation 
-with complexity of the code. \citet{10.1109/TSE.2009.70} compared code readability 
-and software complexity. They used tools, published by \citet{xxx66666444, Readabil74:online}, to
-compute metrics for code readability like \emph{The Automated Readability Index}, 
-\emph{The Simple Measure of Gobbledygook}, \emph{Flesch-Kincaid Readability Index}
-\emph{The Gunning's Fog Index}, \emph{Coleman-Liau Index and Buse Readability Score}. 
-Also, they computed complexity software metrics, such as \emph{Halstead Complexity Volume},
-\emph{McCabe's Cyclomatic Complexity}. The authors found out that there was a 
-negative correlation between the readability and complexity metrics.
-It means that low readability increases program complexity and vice versa. 
-The authors also mentioned that the languages constructions as comments, spacing, 
-while loop, meaningful names and do-while loop affects the code readiblity the most.
-The authors also published a dataset with Java code which can help to detect defects.
-
-\citet{10.1007/978-3-319-95171-3_32} proposed a model which quantitatively measures the readability 
-of source code. The first idea is the using metrics (\emph{LOC, ProgramVolume, Entropy})
-as the key indicators which affect the source code readability. The second
-idea is that the authors introduced the equation for quantitative measure 
-the source code readability in real time. This measure can demonstrate 
-how small changes in the code affect the readability of the code.
-The authors also tried to optimize the model and they reached about 
-74.59\% of explanatory power. 
diff --git a/wp/sections/pattern_emp_analysis.tex b/wp/sections/pattern_emp_analysis.tex
new file mode 100644
index 00000000..5438f084
--- /dev/null
+++ b/wp/sections/pattern_emp_analysis.tex
@@ -0,0 +1,124 @@
+\subsection{Empirical analysis of patterns} 
+
+As a by-product of Aibolit's ML and recommendation
+engine, we get a tool for empirically analysing different patterns' impacts on the target quality metric. Just
+like in the main recommendation algorithm (Algorith~\ref{fig:recsys_alg}, Section~\ref{sec:recommendation_algorithm}), we can estimate whether a particular pattern has a positive or negative impact on the quality metric by considering modifications of source code where pattern count is decreased or increased. We perform such a procedure on a heldout set, which allows to estimate the average impact of a particular pattern on quality. 
+
+In Table~\ref{tab:pattern_analysis} we present a case study of the 34 patterns used at training of Aibolit (see Appendix for the pattern descriptions). On a separate test set (see details in Section~\ref{sec:dataset}), for each pattern, we considered increasing and decreasing the pattern's count by 1. We used the pretrained Aibolit's regression model to predict the corresponding change in quality metric.
+
+
+\begin{table}[ht]
+\footnotesize
+\begin{tabular}{lllllll}
+patterns                                 &   p- m-   &   p+ m+   &   p- m+   &   p+ m-   &   p- m=   &   p+ m=   \\ 
+\\ \hline
+Asserts                                	&	92	&	61	&	186	&	219	&	5	&	3	\\
+Setters                                		&	245	&	160	&	901	&	976	&	21	&	31	\\
+Empty Rethrow                          		&	77	&	65	&	25	&	36	&	0	&	1	\\
+Prohibited class name                  		&	617	&	311	&	218	&	522	&	0	&	2	\\
+Force Type Casting                     		&	2363	&	2313	&	1742	&	1790	&	14	&	16	\\
+Count If Return                        		&	969	&	883	&	214	&	298	&	0	&	2	\\
+Implements Multi                       		&	459	&	320	&	264	&	403	&	0	&	0	\\
+Instance of                            		&	1396	&	1374	&	151	&	173	&	6	&	6	\\
+Many primary constructors              		&	19	&	19	&	604	&	605	&	2	&	1	\\
+Method chain                           		&	573	&	574	&	2217	&	2214	&	43	&	45	\\
+Multiple try                           		&	371	&	239	&	259	&	391	&	0	&	0	\\
+Non final attribute                    		&	1249	&	1185	&	5835	&	5839	&	127	&	187	\\
+Null check                             		&	5863	&	5978	&	575	&	457	&	9	&	12	\\
+Partial synchronized                   		&	46	&	49	&	155	&	149	&	0	&	3	\\
+Redundant catch                        		&	84	&	46	&	38	&	79	&	4	&	1	\\
+Return null                            		&	1290	&	813	&	926	&	1401	&	4	&	6	\\
+String concat                          		&	1596	&	2089	&	1506	&	1012	&	23	&	24	\\
+Super Method                           		&	212	&	275	&	1012	&	949	&	10	&	10	\\
+This in constructor                    		&	15	&	42	&	72	&	45	&	0	&	0	\\
+Var declaration distance for 5 lines   		&	1976	&	1464	&	738	&	1244	&	19	&	25	\\
+Var declaration distance for 7 lines   		&	1190	&	959	&	733	&	949	&	16	&	31	\\
+Var declaration distance for 11 lines  		&	703	&	603	&	365	&	466	&	10	&	9	\\
+Var in the middle                      		&	3372	&	3117	&	2799	&	3046	&	37	&	45	\\
+Array as function argument             		&	882	&	768	&	351	&	464	&	1	&	2	\\
+Joined validation                      		&	232	&	226	&	14	&	26	&	8	&	2	\\
+Non final class                        		&	7141	&	3201	&	3723	&	7663	&	0	&	0	\\
+Private static method                  		&	529	&	541	&	667	&	648	&	2	&	9	\\
+Public static method                   		&	1166	&	1212	&	1142	&	1088	&	4	&	12	\\
+Null Assignment                        		&	1245	&	803	&	717	&	1150	&	6	&	15	\\
+Multiple While                         		&	120	&	97	&	16	&	39	&	0	&	0	\\
+Protected Method                       		&	868	&	402	&	1147	&	1613	&	7	&	7	\\
+Send Null                              		&	556	&	325	&	1510	&	1733	&	5	&	13	\\
+Nested Loop                            		&	580	&	527	&	29	&	81	&	1	&	2	\\
+\hline
+\end{tabular}
+\centering
+\caption{Empirical analysis of patterns. For each source code in the test set, we consider encreasing (\textbf{p+}) and decreasing (\textbf{p-}) pattern count by 1. And we recorded whether the maintainability metric increased (\textbf{m+}) or decreased (\textbf{m-}) as a result of that (where lower is better). In some cases the metric did no change (\textbf{m=}). The values in the cells are counts of cases.} 
+\label{tab:pattern_analysis}
+\captionsetup{font=scriptsize}
+% \caption*{
+%     We use the following notation into named columns:
+%     ($p-$): decrease pattern by $\frac{1}{ncss}$; 
+%     ($p+$): increase pattern by $\frac{1}{ncss}$;
+%     ($c-$): complexity has been decreased; $c+$: complexity has been increased;
+%     ($c=$): complexity has been not changed; 
+%     (\emph{-1(top1)}): decreasing of pattern shows best \emph{CogC} improvement;
+%     (\emph{+1(top1)}): increasing of pattern shows best \emph{CogC} improvement.
+% }
+% \caption*{
+%     We use the following notation into named columns: \\
+%     \\
+%     \centering
+%     \begin{tabular}{rl}
+%         $p-$ & decrease pattern by $\frac{1}{ncss}$ \\
+%         $p+$ & increase pattern by $\frac{1}{ncss}$ \\
+%         $c-$ & complexity has been decreased \\
+%         $c+$ & complexity has been increased \\
+%         $c=$ & complexity has been not changed \\
+%         \emph{-1(top1)} & decreasing of pattern shows best \emph{CogC} improvement   \\
+%         \emph{+1(top1)} & increasing of pattern shows best \emph{CogC} improvement \\
+%     \end{tabular}
+% }
+\end{table}
+
+Based on the statistics in Table~\ref{tab:pattern_analysis}, it appears that \emph{Prohibited class name}, \emph{Count If Return}, \emph{Instance of}, \emph{Null check}, \emph{Nested Loop}, 
+\emph{Array as function argument}, \emph{Joined validation} are \textbf{anti-patterns}, since their count decrease tends to improve  the metric. Another group of patterns are \emph{Setters}, \emph{Many primary constructors}, \emph{Method chain}, \emph{Non final attribute}, \emph{Super Method}, \emph{Send Null patterns}: for them we observe that decresing them causes the metric to deteriorate, and increasing causes the metric to improve. We consider them \textbf{pro-patterns}. The third group (the rest of the patterns) can both improve and deteriorate the metric. We restrain oursevles from calling them either anti- or pro-patterns.
+
+Attempting to interpret the results, we observe that ``true'' anti-patterns usually have \emph{if/else condition} or cycle in their definition.
+E.g., \emph{Null check} always checks for a null, \emph{Count If Return}, \emph{Instance of}, 
+\emph{Joined validation} have always \emph{if condition}, \emph{Nested Loop} has always at least 1 loop inside. Given that so far we have worked with the Cognitive Complexity metric, it is no surprise that those patterns affect it (\emph{if/else condition} or cycle are the main contributors into \emph{CogC}). Despite this limitation of the present analysis, we believe the proposed \textit{method} itself can be very useful in software engineering practice and research.
+
+
+
+
+% ORIG TABLE:
+% patterns                                & -1(top1)  & +1(top1)  &   p- m-   &   p+ m+   &   p- m+   &   p+ m-   &   p- m=   &   p+ m=   \\ 
+% \\ \hline
+% Asserts                                 &   0   &   100 &   92  &   61  &   186 &   219 &   5   &   3   \\
+% Setters                                 &   1   &   113 &   245 &   160 &   901 &   976 &   21  &   31  \\
+% Empty Rethrow                           &   1   &   4   &   77  &   65  &   25  &   36  &   0   &   1   \\
+% Prohibited class name                   &   80  &   24  &   617 &   311 &   218 &   522 &   0   &   2   \\
+% Force Type Casting                      &   69  &   24  &   2363    &   2313    &   1742    &   1790    &   14  &   16  \\
+% Count If Return                         &   311 &   26  &   969 &   883 &   214 &   298 &   0   &   2   \\
+% Implements Multi                        &   24  &   244 &   459 &   320 &   264 &   403 &   0   &   0   \\
+% Instance of                             &   211 &   6   &   1396    &   1374    &   151 &   173 &   6   &   6   \\
+% Many primary constructors               &   0   &   343 &   19  &   19  &   604 &   605 &   2   &   1   \\
+% Method chain                            &   3   &   203 &   573 &   574 &   2217    &   2214    &   43  &   45  \\
+% Multiple try                            &   156 &   180 &   371 &   239 &   259 &   391 &   0   &   0   \\
+% Non final attribute                     &   34  &   485 &   1249    &   1185    &   5835    &   5839    &   127 &   187 \\
+% Null check                              &   1573    &   14  &   5863    &   5978    &   575 &   457 &   9   &   12  \\
+% Partial synchronized                    &   1   &   93  &   46  &   49  &   155 &   149 &   0   &   3   \\
+% Redundant catch                         &   6   &   2   &   84  &   46  &   38  &   79  &   4   &   1   \\
+% Return null                             &   104 &   40  &   1290    &   813 &   926 &   1401    &   4   &   6   \\
+% String concat                           &   43  &   126 &   1596    &   2089    &   1506    &   1012    &   23  &   24  \\
+% Super Method                            &   1   &   174 &   212 &   275 &   1012    &   949 &   10  &   10  \\
+% This in constructor                     &   2   &   891 &   15  &   42  &   72  &   45  &   0   &   0   \\
+% Var declaration distance for 5 lines    &   396 &   14  &   1976    &   1464    &   738 &   1244    &   19  &   25  \\
+% Var declaration distance for 7 lines    &   25  &   95  &   1190    &   959 &   733 &   949 &   16  &   31  \\
+% Var declaration distance for 11 lines   &   16  &   686 &   703 &   603 &   365 &   466 &   10  &   9   \\
+% Var in the middle                       &   118 &   50  &   3372    &   3117    &   2799    &   3046    &   37  &   45  \\
+% Array as function argument              &   86  &   25  &   882 &   768 &   351 &   464 &   1   &   2   \\
+% Joined validation                       &   63  &   2   &   232 &   226 &   14  &   26  &   8   &   2   \\
+% Non final class                         &   1056    &   2370    &   7141    &   3201    &   3723    &   7663    &   0   &   0   \\
+% Private static method                   &   35  &   133 &   529 &   541 &   667 &   648 &   2   &   9   \\
+% Public static method                    &   37  &   408 &   1166    &   1212    &   1142    &   1088    &   4   &   12  \\
+% Null Assignment                         &   103 &   35  &   1245    &   803 &   717 &   1150    &   6   &   15  \\
+% Multiple While                          &   65  &   7   &   120 &   97  &   16  &   39  &   0   &   0   \\
+% Protected Method                        &   71  &   151 &   868 &   402 &   1147    &   1613    &   7   &   7   \\
+% Send Null                               &   13  &   109 &   556 &   325 &   1510    &   1733    &   5   &   13  \\
+% Nested Loop                             &   412 &   35  &   580 &   527 &   29  &   81  &   1   &   2   \\
\ No newline at end of file
diff --git a/wp/sections/related_work.tex b/wp/sections/related_work.tex
index d0f22f6d..f8ce4101 100644
--- a/wp/sections/related_work.tex
+++ b/wp/sections/related_work.tex
@@ -1,89 +1,85 @@
-There are many definitions of a defect.  
-\citet{5989519} says that defect is ``a fault, bug, inaccuracy or lack of expected 
-functionality in a project artifact.'' 
-\citet{Assurance} says that it is ``a problem 
-(synonym of fault) which, if not corrected, 
-could cause an application to either fail or to produce incorrect results.''
+\subsection{Quality and quality metrics}
+IEEE Standards define software quality as the array of features of
+a software product that represent its capability to satisfy 
+specific needs \citep{Youness2013ComparativeSO}.
+Software quality  is the extent to which  a  process,  
+component,  or  system fulfills customers' needs  or expectations 
+through product or service features, thus providing customer 
+satisfaction \citep{Iacob_Constantinescu}. 
 
-Defects and software quality are directly related. There are multiple studies 
-of defect types, their impact, complexity, 
-root causse and other characteristics,~\citet[e.g.][]{10.1145/69605.2085, 
-10.5555/256664.256773, 10.1145/390016.808455, Glass1981PersistentSE, 
-10.1145/1353535.1346323, 10.1007/s10664-013-9258-8, catolino2019bugs}.
-There are a few common preventive ways to deal with defects, like 
-static code analyzers, testing software, or peer code review. 
+Functional and  structural qualities are the key  aspects  of  software 
+quality  \citep{Liu2006AnIE}.  \citet{Farhan} describe 
+functional  quality  as  the capability  of  the  software 
+to  properly  perform  its  tasks according  to  user  needs  and  
+intended  objectives.  Structural quality refers to 
+the resilient structure of the code itself and is difficult to test
+compared to functional quality. The main difficulty is that this 
+notion is quite subjective.
 
-Static code analyzer 
-is a tool that helps find defects before a program is executed. 
-Such an analyzer inspects various program representations, for example 
-Abstract Syntax Tree (AST), Control Flow Graphs (CFG), or
-Program Dependency Graph (PDG), 
-and search for handcrafted defect patterns. 
-These tools are popular among developers and are often embedded into
-Integrated Development Environments (IDE) such as IntelliJ IDEA or NetBeans. 
-There are more than 40 static analyzers currently on the market, including
-very famouns open source projects, such as
-PMD, Rubocop, PHPCS, Sparse, CLion, cpplint, FindSecBugs, ESLint, and Checkstyle.
-There are also many commercial tools, like 
-IBM Security AppScan, PVS-Studio, SonarQube, and Parasoft.
-Software companies like Google and Facebook have their own open source
-static analyzers: Error Prone~\citep{10.1109/SCAM.2012.28} and 
-Infer~\citep{10.1007/978-3-319-17524-9_1} respectively. 
-Static code analyzer may vary in supported languages, 
-supported defect types, and their integration workflow.
+In general, structural code quality is a multi-faceted concept, which covers
+different  attributes of software engineering, for example, maintainability  and
+readability \citep{Mohammadi2013AnAO}. To evaluate them, various metrics of
+software structure were proposed. For instance,  McCabe’s software complexity
+metrics \citep{McCabe1976ACM} and cognitive  complexity metric
+\citep{Cognitive}, which are intended to measure readability  aspects of the
+code. Also, for object-oriented systems, a popular  set of metrics is the CK
+suite \citep{Chidamber1994AMS}. Many approaches apply such metrics suite to
+distinguish parts of the source code with good  or bad quality
+\citet{Fil2015ACO}, \citet{Shatnawi2010FindingSM}  or to identify code smells
+(problematic properties and anti-patterns  of code)
+\citet{Ouni2011MaintainabilityDD}. However, in general, the software engineering
+community has not yet reached a consensus as to what exactly structural quality
+or maintainability is \citep{Broy2006DemystifyingM}.
+% (\todo: this is quite an old reference...).
 
-Usually static code analyzers are rule-based. 
-In order to detect a new kind of defect their developers have to
-design a new pattern. To address this extensibility problem 
-there were attempts to learn pattern
-from data, as explained by~\citet{bielik2016learning, wang2019learning}.
 
-Recent empirical study by~\citet{10.1145/3238147.3238213} demonstrates that 
-state-of-the-art code analyzers miss more than 90\% of defects. Most of those missed 
-defects are inconsistencies with the specification or programmer's intent. 
-In order to catch such defects it is necessary to reason about 
-possible behavior depending on the input data, 
-which is difficult or impossible for the rule-based approach. 
-Such defects are known as \emph{semantic} defects. 
-It has been demonstrated by~\citet{10.1007/s10664-013-9258-8} 
-that semantic defects are the dominant root cause for the majoirity of security issues, 
-when attacker may get unauthorized access to some resources.
+\subsection{Software patterns and code smells}
 
-Another issue of most static analyzers is their high 
-percent of ``false-positives,'' when they find a defect, which
-in reality is not a defect. The amount of these false signals grows 
-when the size and complexity of the project increases. 
-This leads to developers loosing trust to the tool and stop
-using it. 
+We understand the term \textit{software patterns} in the most general and
+abstract way, namely, as any observed code subtructures and software solutions.
+Patterns can be of different scale (from variable and method-level to project
+level). The term \textit{designed patterns} refers to patterns that are
+recommended solutions to commonly occurring programming goals and problems
+\citep{gamma1995design}. Despite their popularity, there is much controversy about the
+usefulness and universality of such recommended ways of implementation
+\citep{mcconnell2004code}. The software engineering community
+also identifies patterns that are detrimental to the resulting code. Such
+patterns are often called \textit{code smells}. These are parts of the
+source code that contain violations of  fundamental design principles
+and negatively impacts maintainability in terms of the
+ability of the product to evolve, quality  of end-product, and developer
+productivity \citet{Reeshti2019MeasuringCS}. \citet{Din2012AntipatternsDA}
+identified 22 types of  code smells in object-oriented
+design. \citet{Kessentini2019UnderstandingTC} found strong correlation  between
+several code smells and software bugs.
 
-The efficiency of static analyzers is yet another problem. It was
-empirically shown by~\citet{10.1145/3188720} that 
-a better performance metric for a static analyzer 
-is the amount of fixed, rather than found defects. 
-Static analyzers must provide the right information 
-at the right time doing everything possible to not annoy 
-software developers. 
+For the latter reason, a lot of tools and methods have been designed to avoid
+code smells. \citet{Kreimer2005AdaptiveDO}  proposes a decision tree based
+approach to identify code smells, e.g, long method  and large class.
+\citet{Vaucher2009TrackingDS} apply Bayesian  networks to detect God class.
+\citet{Palomba2015MiningVH} propose  to consider changes of repository history
+as an input to the code smell detector  for computing the list of code
+components affected by the smell. \citet{Liu2019DeepLB} propose a deep learning
+based approach  to detect code smells.
 
-Usually, static code analyzers use fixed in time code versions as their
-input. However, one can also use the history of code changes and 
-the information from the Issue Tracking System (ITS) 
-to enhance the quality of prediction, as done by~\citet{Gupta2018IntelligentCR, kapur2018estimating}.
+Code smell detection has been integrated into code inspection tools. 
+\citet{MurphyHill2010AnIA} integrate software metrics  visualization with a
+source code view.  SonarQube\footnote{https://www.sonarqube.org/} controls and
+manages  the code quality in several ways, such as continuous  inspection and
+issue detection. The platform shows issues  like code smells, bugs, using
+lightweight  visualizations. 
+%It also helps developers to collaborate with a
+%shared vision of code quality.
+Checkstyle\footnote{https://checkstyle.sourceforge.io/} and
+PMD\footnote{https://pmd.github.io/}  work similarly to SonarQube.  
 
-There are studies trying to use ML in order to detect defects. 
-For example, \citet{Dam2018ADT} first trained vector representations of 
-an AST in an unsupervised manner and then used it as a feature vector 
-to train the binary classifier. 
-\citet{kapur2018estimating} 
-combined the information extracted from the code programming style and ITS 
-and built a predictor to estimate the defectiveness of an input source code 
-file. Using an idea that the names of identifiers (variables, classes, functions) 
-convey useful information, which might be used to understand programmer's 
-intent,
-\citet{Pradel2018DeepBugsAL} proposed a method that first learned 
-vector representation of identifiers 
-and obtained a fixed length vector for a code snippet to train a binary classifier 
-with feed-forward neural network. 
-\citet{vasic2019neural} used pointer-network to do joint prediction of 
-both the location and the possible fix for variable misuse bugs. 
-\citet{briem2019using} used attention-based neural network  to model binary 
-classifier to detect off-by-one defects.
+%iPlasma
+%evaluate the quality of  %object-oriented programs. It can be used %to calculate
+%different metrics and to identify the most %critical points that can be improved
+%through code smell %detection.
+
+All in all, there is no uniform agreement about which patterns are good and
+which are bad. We made it our ideology while developing Aibolit: we do not
+decide what is good or useful \textit{a priori} but let it be inferred from
+data. By customizing the dataset, quality metric and pattern set, the end user
+of Aibolit is able infer which patterns are good for his own end goals.
diff --git a/wp/sections/risks.tex b/wp/sections/risks.tex
deleted file mode 100644
index fffe210e..00000000
--- a/wp/sections/risks.tex
+++ /dev/null
@@ -1,26 +0,0 @@
-There is a number of risks we identified and expect to mitigate.
-
-\textbf{Survey Validity}.
-While designing the survey procedure we have to 
-consider the following threats to its validity:
-\begin{enumerate*}[label=\arabic*)]
-    \item Set of interviewers must be representive. 
-    Will the results be different if we take a different set of interviewers?
-    \item Set of code snippets must be representive. 
-    Will the results be different if we take a different set of code snippets? 
-    \item How to identify bad interviewers? 
-    Some interviewers may give answers that are not correlated with actual readability.
-\end{enumerate*}
-To mitigate this risk we have to think about the diversity of interviewers, 
-for example by age, skills, experience, education, and so on. We have to select code snippets
-that vary in length and project domain. The more data we collect the better. 
-We have to preprocess survey results to exclude outliers.
-
-\textbf{Feature Limits}.
-To predict the readability we are going to use features 
-calculated with static code analysis tools. 
-These features mainly reflect the structure, the syntax and 
-size properties of a particular code snippet.  
-However, we are not considering semantic properties of the code. 
-In the future, we can design semantic related features and 
-add them to our dataset.
\ No newline at end of file
diff --git a/wp/sections/saved.tex b/wp/sections/saved.tex
deleted file mode 100644
index af50cd7f..00000000
--- a/wp/sections/saved.tex
+++ /dev/null
@@ -1,63 +0,0 @@
-% The refactoring recommendation
-% is a pointer to specific line code with description what to do to improve code readability. 
-% To do this we studied a relationships between the code readability and mannualy designed AST patterns usually 
-% encountered in the code. Knowing that the pattern presented in code and has negative impact on readability
-% we do our recommendations, pointing to code line where this pattern presented in the code.
-
-% In our work we created a dataset of Java code snipets features. The features may be devided into 
-% three groups: code metrics, AST patterns and code readability. The code metrics calculated using existing
-% tools: CheckStyle, RefactoringMiner, ChangeDistiller, and SourceMeter. The AST patterns are manually designed
-% features that reflect a presense of some syntax structures or other static code preperties, for example
-% number of nested FOR loops of length 2 or number ternary operators. To obtain the code readability we conducted 
-% a survey that described in the next section.
-
-% \subsection{Dataset}
-
-% Next, we enriched each pair with metrics calculated 
-% based on static code, for example \acrfull{sloc}, \acrfull{cyclo}. Eventually we 
-% have the dataset, where for each snippet of code there are readability score and the set of metrics.
-
-
-% % \begin{itemize}
-% %     \item Is it possible to predict code readability using static metrics of code? 
-% %     \item Which metrics do have more importance to predict code readability?
-% %     \item Which ML model predict code readability better?
-% %     \item Is it possible to recommend particular code refactoring to improve code readability?
-% % \end{itemize}
-
-% \subsection{Methodology}
-
-% We split our work into two parts. The first one is readability score prediction. 
-% The second part is code refactoring recommendations to improve code readability.
-
-% \subsubsection{Readability prediction}
-
-% Having unseen code snippet we have to estimate its readability score, because we are
-% not going to give recomendation if the readability score is acceptable.
-% We train a \acrshort{ml} model using gathered dataset.
-% We stated the problem as a regression problem of predicting the readability score:
-
-% $$
-%     r_{i} = f(X_{i}, \theta) + e_{i}
-% $$
-
-% where $i$ is the index of rows in our dataset, $r$ is the readability score, $X$ is 
-% the set of the code snippet's metrics, $e$ is some error or noise. And our goal is to choose 
-% the parametric function $f$ and finds its parameters $\theta$.
-
-% We tested a set of well-known regression models, like Linear Regression, \acrfull{cart}, 
-% to predict readability. We selected the best model among 
-% the considered set.
-
-% \subsubsection{Refactoring recommendations}
-
-% We measured a features importance with respect to readability score and made an ordered
-% list of all features by importance.
-% Having unseen code snippet with unacceptable readability score we do our refactoring recomendations.
-% We rank our recommendations based on feature importance list. For AST related features we are able 
-% to point to specific place in the code. 
-
-
-% We formulate our business goal following way: having code snippet we want 
-% get list of refactoring recommendations indicating line and what to do to improve code quality. 
-% List should be ordered by refactoring impact on quality.
\ No newline at end of file
diff --git a/wp/sections/selected_task.tex b/wp/sections/selected_task.tex
deleted file mode 100644
index 141e82fd..00000000
--- a/wp/sections/selected_task.tex
+++ /dev/null
@@ -1,84 +0,0 @@
-%Here will be paragraph linking previous sections with "Code Readability".
-
-%The research question is the following: is it possible, using static code analysis to do 
-%a code refactoring recommendations to improve a code readability? 
-
-We are looking for an answer to the following research question: 
-Is it possible,
-combining static code analysis and ML, 
-to detect defects in a Java class and 
-to give specific recommendations for its refactoring? 
-
-To answer this question we are doing the following research:
-\begin{enumerate*}[label={\alph*)}]
-    \item We take a set of Java \emph{classes};
-    \item We collect a number of static analysis \emph{metrics} per each class;
-    \item We locate \emph{code patterns} inside each class;
-    \item We ask volunteer programmers to review 
-      each class to estimate their \emph{readability};
-    \item We put collected data together into a \emph{dataset}; 
-    \item We find \emph{relations} between metrics and patterns in the dataset; 
-    \item We take an unseen Java class and \emph{locate code patterns}, which impact the quality more then others.
-\end{enumerate*}
-    
-\textbf{Classes}.
-We parse Java projects from GitHub to obtain training Java classes. 
-All parsed projects must have more than 100 stars, and more than five collaborators. 
-Collected Java classes must have more than 50 lines of code and less than 300.  
-
-\textbf{Metrics}.
-The code metrics are calculated using existing open-source
-tools like CheckStyle and SourceMeter. Examples of metrics are Lines of Code (LoC), 
-Cyclomatic Complexity (CC), and Number of Incoming Invocations (NII).
-
-\textbf{Code Patterns}.
-AST patterns are the features that reflect the presence 
-of some syntax structures in the code, 
-for example number of nested \texttt{FOR} loops 
-or the amount of ternary operators. The AST patterns are manually designed 
-and always point to a specific place in the code. 
-
-\textbf{Readability}.
-To gather readability characteristics we conduct
-a survey, where volunteering programmers are asked to estimate
-the readability of training Java classes, giving answers on a $[0..9]$ scale, 
-where 0 means not readable at all and 9 means perfectly readable.
-Each programmer is asked to review some snippets from 
-the entire training dataset, which means that each
-snippet is reviewed by a few programmers. The readability ``score'' per
-snippet is a mean of all answers collected.
-
-\textbf{Dataset}.
-We combine and put the data into the single dataset. 
-In the \autoref{tab:table1} you can see the example how dataset looks
-(here CC, CBO, LCOM, and NMD are acronyms for software metrics, 
-while RS is the Readability Score collected from volunteers).
-
-\begin{table}[H]
-\begin{center}
-\begin{tabular}{rrrrrr}
-\hline
-CC & CBO & LCOM & NMD & \dots & RS \\
-\hline
-3 & 6 & 34 & 2 & \dots & 4.3 \\
-4 & 5 & 55 & 1 & \dots & 2.4 \\
-3 & 5 & 22 & 0 & \dots & 5.2 \\ 
-\hline
-\end{tabular}
-\end{center}
-\caption{Example of collected dataset}
-\label{tab:table1}
-\end{table}
-
-\textbf{Relations}.
-Using ML methods we are going to learn how code metrics, 
-AST patterns and readability are related. 
-We are planning to use ML techniques to investigate the importance of features 
-with respect to the readability and find combinations that have stronger impact.
-
-\textbf{Locate Code Patterns}.
-Finally, for an unseen Java class we can calculate all features except the readability. 
-Using known features and the knowledge about relations between features we are planning
-to give refactoring recommendations and provide links to the lines of code 
-where most problematic patterns are located.   
-
diff --git a/wp/sections/selection_criteria.tex b/wp/sections/selection_criteria.tex
deleted file mode 100644
index abfea3af..00000000
--- a/wp/sections/selection_criteria.tex
+++ /dev/null
@@ -1,83 +0,0 @@
-There is a number of crucial expectations we have for the
-method and the tool under design:
-
-\textbf{Novelty} is crucial and has a few important aspects:
-\begin{enumerate*}
-\item 
-Novelty of the \emph{method} is one the most important values of novelty. 
-E.g.,~\citet{Akiyama1971AnEO} was the first who 
-tried to solve defect detection problem using 
-linear regression as it was mentioned by~\citet{7476771}.
-Sometimes authors can use some new idea in the 
-algorithm, e.g,~\citet{XIAO201917} used convolutional neural network
-together with word-embedding and feature-detecting techniques.
-
-\item
-The \emph{effectiveness} of the model is another important 
-value of novelty. If authors published new model which 
-is more effective, it means that we can solve a problem 
-much better. All evaluation measures can be described later.
-
-\item
-Collecting new dataset is one of the most important and hard work in Data Science. 
-Main results can vary depending on method used when gathering a dataset. 
-It is necessary to do it in a correct way, since there are many issues related 
-to data problems like \emph{Outliers}, \emph{Class Imbalance}, \emph{Data shift Problem}, 
-\emph{High Dimensionality of Data}, etc. mentioned by~\citet{10.1007/s10462-017-9563-5}.
-Also, \emph{granularity} is another problem in the 
-defect detection problem. Defect can be found in a module, 
-function, line of the code, etc. That is why seems 
-it is not an easy task to compare the models with a
-different level of granularity.
-E.g,~\citet{6464273} analyzed five papers published by 
-IEEE and mentioned that there are some data 
-quality issues in these papers.
-\end{enumerate*}
-
-\textbf{Evaluation} is the essential of ML. It is impossible 
-to say whether a model is good or bad, not evaluating it. 
-A lot of model performance evaluation measures were used 
-for defects detection, but we can 
-classify them the following way \citep{10.1007/s10462-017-9563-5, Jiang2008}:
-\begin{enumerate*}[label=\arabic*)]
-\item \emph{Numeric} performance evaluation measures
-are mostly used for defect detection models~\citep{Jiang2008}. 
-They can include accuracy, F1-score, G-means, specificity, 
-f-measure, and so on. E.g., they are used by~\citet{6349519}. 
-\item \emph{Graphical} measures are graphs derived from 
-the confusion matrix~\citep{Jiang2008}. 
-They can include ROC curve, precision-recall curve, cost curve etc. 
-E.g., they are used by~\citet{SHATNAWI20081868}.
-\end{enumerate*}
-
-\textbf{Performance} is another important value for defect detection model. 
-The selection of the best performance evaluation measure 
-is not a trivial task.
-For example,~\citet{Jiang2008} compared different alternatives
-and demonstrated that no single performance evaluation measure is able to evaluate the
-performance of a defect prediction model fair enough. The authors also added 
-that it is better not only to measure model classification performance 
-(like accuracy, recall, etc.), but to minimize the misclassification cost.
-\citet{ARISHOLM20102} also compared different types of measures 
-and mentioned that it is hard to draw general conclusions about the 
-best performance evaluation measure.
-
-Since the tool will be actively used with Continuous Integration (CI), 
-it must have a reasonable performance: it must predict in a few seconds. 
-If calculation takes a lot of time, a developer may stop using it.
-\citet{humble2010continuous} also mentioned that speed 
-is very important for CI. The sooner you release the software, 
-the sooner you get a return on your investment. 
-Performance is also important because it is necessary 
-to know whether bugfixes are useful. That is why
-we need to minimize the delay between the releases and 
-thus, accelerate the feedback, as explained by~\citet{humble2010continuous}.
-
-State-of-the-art papers use Neural Networks (NN) for defect 
-detection problem since it can give better results~\citep{XIAO201917,10.1145/3360588}. 
-Training takes a lot of time and demands TPU or GPU resources if we use NN. 
-Word embedding and other feature-detecting techniques can only increase training time. 
-E.g., \citet{10.1145/3360588, 8616596} noticed that 
-the training time matters for defect detection problem and 
-demonstrates the training time of their models.
-
diff --git a/wp/sections/threats_to_validity.tex b/wp/sections/threats_to_validity.tex
new file mode 100644
index 00000000..ac2bd907
--- /dev/null
+++ b/wp/sections/threats_to_validity.tex
@@ -0,0 +1,4 @@
+% consider only frequencies not inter location
+% not consider context
+% top OSS datasets
+% quality metrics 
\ No newline at end of file
diff --git a/wp/sections/usage_scenarios.tex b/wp/sections/usage_scenarios.tex
new file mode 100644
index 00000000..f3317d06
--- /dev/null
+++ b/wp/sections/usage_scenarios.tex
@@ -0,0 +1,68 @@
+\subsection{Aibolit Index}
+
+In addition to the recommendation functionality, we designed a feature to measure the overall quality of developer's source code, \textit{Aibolit Index}.
+It is is a single number, with following properties: 
+\begin{itemize}
+\item[(i)] the more patterns are suggested to fix, the higher Aibolit Index;
+\item[(ii)] the higher negative impact factor (see Section \ref{sec:recommendation_algorithm}) 
+of detected patterns the higher Aibolit Index.
+\end{itemize}
+Therefore, the lower Abolit Index the better the project's code from the point 
+of view of Aibolit.
+
+Let $P(C)$ be a set of all patterns Aibolit recommends to fix for Java class $C$.
+% As we know from the section ~\ref{sec:recommendation_algorithm}, each 
+% recommended anti-pattern $p \in P(C)$ has an associated impact factor $I_{p}$ and a count $count(p, C)$.
+We define the Aibolit Index $A(C)$ of Java class $C$ as a sum of the products of
+impact factors $I_{p}$ and scaled counts $count(p, C)$ for all of the 
+patterns occurring in the class (Eq.~\ref{eq:aibolit_index}). We use log-scaling for smoothing purposes, because some patterns are a lot more common than others.    
+
+
+\begin{equation}
+    A(C) = \sum_{p \in P(C)} { I_{p}(C) \cdot \ln{(count(p, C) + 1)} } \label{eq:aibolit_index}
+\end{equation}
+
+
+The Aibolit Index of a project is defined as average Aibolit Index
+across all Java classes in the project. In Table~\ref{table:aibolit_index_repos} 
+you can see a calculated Aibolit Index of some GitHub Java repositories with 
+more than 4500 stars. Aibolit Index is supposed to be a convenient instrument
+to get a first estimate of the project code's quality. 
+
+\begin{table}[t]
+\footnotesize
+    \begin{tabular}{|l|l|l|l|l|l|l|}
+    \hline
+    Repository & Aibolit Index & Total files & Total NCSS & GitHub stars \\ 
+    \hline
+    ReactiveX\textbackslash RxJava& 6.66& 1493& 25270 & 42972  \\
+    bumptech\textbackslash glide& 5.81& 465& 5078 & 29364 \\
+    JakeWharton\textbackslash butterknife& 6.31& 74& 935 & 25347  \\
+    greenrobot\textbackslash EventBus& 7.48& 51& 651 & 22637  \\
+    skylot\textbackslash jadx& 6.69& 602& 12658 & 22602 \\
+    alibaba\textbackslash fastjson & 9.97& 144& 20175 & 21891 \\
+    alibaba\textbackslash druid & 6.36& 822& 28852 & 21581 \\ 
+    Netflix\textbackslash Hystrix& 6.74& 292& 2920 & 19888 \\
+    ReactiveX\textbackslash RxAndroid& 5.47& 9& 59 & 19010 \\
+    google\textbackslash gson& 6.87& 160& 2941 & 18084 \\
+    square\textbackslash picasso& 6.21& 26& 687 & 17514 \\
+    libgdx\textbackslash libgdx & 5.02 & 1981 & 46409 & 17105 \\
+    nostra13\textbackslash Android-Universal-Image-Loader& 7.84& 62& 1059 & 16722  \\
+    qiurunze123\textbackslash miaosha& 3.57& 197& 841 & 16224 \\
+    wuyouzhuguli\textbackslash SpringAll& 13.41& 467& 463 & 15221 \\
+    justauth\textbackslash JustAuth& 4.36& 46& 241 & 8916 \\
+    heibaiying\textbackslash BigData-Notes& 11.17& 54& 204 & 7166  \\
+    crossoverJie\textbackslash cim & 6.70& 96& 574 & 5841 \\
+    wildfirechat\textbackslash server& 6.51& 285& 4671  & 5140 \\
+    febsteam\textbackslash FEBS-Shiro & 5.66& 90& 453 & 4777 \\
+    \hline
+    \end{tabular}
+    \centering
+\caption{Aibolit Index of some popular Java repositories. \label{table:aibolit_index_repos}}
+\end{table}
+
+
+
+
+
+
diff --git a/wp/wp.tex b/wp/wp.tex
index aa6fcea9..93c1d040 100644
--- a/wp/wp.tex
+++ b/wp/wp.tex
@@ -1,8 +1,7 @@
-\documentclass[12pt]{article}
+\documentclass[10pt]{article}
 \usepackage{natbib}
 \usepackage{pgf}
-  \bibliographystyle{plainnat}
-  \setcitestyle{citesep={,},aysep={}}
+\setcitestyle{citesep={,},aysep={}}
 \usepackage[colorlinks,citecolor=blue,linkcolor=black,bookmarks=false,hypertexnames=true]{hyperref} 
 \usepackage{url}
 \usepackage[inline]{enumitem}
@@ -12,9 +11,55 @@
 \usepackage{graphicx}
 \usepackage{xcolor}
 \usepackage{setspace}
-
 \usepackage{enumitem}
+\usepackage{caption}
+%\usepackage[margin=1.3in]{geometry}
+\usepackage{xcolor}
+\usepackage[top=1.3in, left=1.4in, includefoot]{geometry}
 \setlist{nosep}
+\usepackage{xcolor}
+\usepackage{algorithm}
+\usepackage{algpseudocode}
+\usepackage{multicol}
+\usepackage[T1]{fontenc}
+\usepackage[utf8]{inputenc}
+\usepackage{charter}
+\usepackage{listings}
+\usepackage{authblk}
+
+\definecolor{codegreen}{rgb}{0,0.6,0}
+\definecolor{codegray}{rgb}{0.5,0.5,0.5}
+\definecolor{codepurple}{rgb}{0.58,0,0.82}
+\definecolor{backcolour}{rgb}{0.95,0.95,0.92}
+
+\lstdefinestyle{mystyle}{
+	backgroundcolor=\color{backcolour},   
+	commentstyle=\color{codegreen},
+	keywordstyle=\color{magenta},
+	numberstyle=\tiny\color{codegray},
+	stringstyle=\color{codepurple},
+	basicstyle=\sffamily\footnotesize,
+	breakatwhitespace=false,         
+	breaklines=true,                 
+	captionpos=b,                    
+	keepspaces=true,                 
+	numbers=left,                    
+	numbersep=5pt,                  
+	showspaces=false,                
+	showstringspaces=false,
+	showtabs=false,                  
+	tabsize=2
+}
+
+\lstset{style=mystyle}
+
+
+\newcommand{\pattern}[3]{
+	{\bf #2. #1} \\
+	{\it Description:} #3 \\
+}
+
+\newcommand{\todo}{\textcolor{red}{\textbf{TODO}}}
 
   \setstretch{1.1}
 \setlength{\bibsep}{0.0pt}
@@ -30,88 +75,65 @@
 \renewcommand*{\bibfont}{\footnotesize}
 
 \title{
-  \includegraphics[height=48pt]{logo}\\
-  \textsc{Aibolit}:\\
-  Style Checking\\
+  \includegraphics[height=100pt]{logo.png}\\
+  \vspace{10pt}
+  \textsc{Aibolit:}\\
+  Static Analysis
   Using Machine Learning}
 
+\author{Yegor Bugayenko, Anton Cheshkov, Ekaterina Garmash, Andrey Gusev, Yaroslav Kishchenko, Pavel Lukyanov, Evgeny Maslov, Vitaly Protasov}
+\affil{Huawei Technologies \\
+		System Programming Lab, Russian Research Institute, Moscow}
+
 
 \begin{document}
 
 \maketitle
 
+\pagebreak
+
 \begin{abstract}
-Quality of code is crucial for the stability and robustness
-of software systems. Defect detection at early stages of
-software development lifecycle is the cheapest and the easiest
-way to increase the quality of code. There are well-known
-instruments such as static analyzers and linters, which are
-actively used by programmers. However, none of them use Machine
-Learning (ML) to detect defects more effectively. 
-We managed to create such an instrument and empirically demonstrate 
-its effectiveness\footnote{This is statement is yet to be confirmed}. 
+
+Aibolit is a next generation static analyzer powered by machine learning.
+Aibolit gives recommendations to developers to avoid specific software patterns  
+in order to improve quality of the source code. Aibolit can be extended by adding 
+custom patterns and quality metrics of choice. In this paper, we explain 
+how Aibolit works and how it differs from other static analyzers.
+
+
 \end{abstract}
 
 \pagebreak
 
 \section{Introduction}
+\label{sec:intro}
 \input{sections/introduction}
 
-\section{Related Work}
+
+\section{Software quality and patterns}
 \label{sec:related}
 \input{sections/related_work}
 
-\section{Categories of Methods}
-\label{sec:categories}
-\input{sections/considered}
-
-\section{Available data}
-\label{sec:data}
-\input{sections/available_data}
-
-\section{Selection Criteria}
-\label{sec:criteria}
-\input{sections/selection_criteria}
-
-\section{The Method}
-\label{sec:method}
-\input{sections/selected_task}
-
-\section{Risks}
-\label{sec:risks}
-\input{sections/risks}
+\section{How the Aibolit recommender works}
+\label{sec:how_aibolit_works}
+\input{sections/how_aibolit_works}
 
-% \section{Implementation Details}
-% \label{sec:implementation}
-% \input{sections/implementation}
+\section{Other usage scenarios}
+\label{sec:usage_scenarios}
+\input{sections/pattern_emp_analysis}
+\input{sections/usage_scenarios}
 
-\section{Empirical Results}
-\label{sec:results}
-\input{sections/empirical_results}
+\section{Conclusion \& Future Work}
+\label{sec:conclusion}
+\input{sections/conclusion}
 
-\section{Conclusion}
-We will write this section later.
 
-\section{Future Work}
-We will write this later.
-
-\section{Acknowledgements}
-The tool was designed in System Programming Lab. Many thanks
-to the contribution of (in alphabetic order of the last name):
-Yegor Bugayenko,
-Anton Cheshkov,
-Vadim Chibiriev,
-Lu Jianhua,
-Evgeny Maslov,
-Yu Jiayuan,
-Alexey Zorchenkov.
-
-%\section{Notes}
-%\input{sections/notes}
+\section{Appendix}
+\label{sec:appendix}
+\input{sections/appendix}
 
+\bibliographystyle{apa} 
 \bibliography{references}
 
 
-
-
 \end{document}