(;GM[1]FF[4]CA[UTF-8]AP[CGoban:3]ST[2]RU[Japanese]SZ[19]KM[6.50]PW[alphago]PB[alphago]C[very slow games 2 mins/movei.e. 8-10 hours/game"The art of defence"Fan Hui 2p (Deepmind)Shi Yue 9pLiu Xing 7pThis is from Fan's analysis of alphago's reading];B[pd];W[dd];B[pp];W[dq]C[Shi: alphago likes to invade 3-3, but still plays 4-4?Fan: It sees it as even, like a joseki];B[co](;W[fp]C[Liu: Nie is particularly fond of this moveShi: whereas we don't like the fact B has a lot of ways for sabaki with attach C3, double hane, tiger, or crosscutbut this is faster than E4, and is better if W spends another move at C4Fan: alphago strives for speed in the opening, and likes to play in the widest areas];B[cc]C[another of alphago's "playing cards"];W[cd];B[dc](;W[fc](;B[fb]C[but this game alphago directly attachesLiu: humans normally play here, and W takes sente or plays G17];W[gb]C[Fan: no W doesn't tenuki, W blocks](;B[eb]C[but in the game, alphago directly pulls back];W[gd];B[ed];W[ec];B[bd];W[be];B[bc](;W[cf]C[Fan: the corner shape is standard, but the outside isn't];B[ee]C[Fan: W asks B to run out, and B doesLiu: neither side is scared of the other];W[df];B[ef];W[eg];B[ge];W[fa];B[db];W[he]C[Shi: alphago plays very differently with itself compared to with us. The self-play is normally very violent, but with us, it doesn't fight, but just builds up a steady advantage.Liu: we always look for a fight, but it never works.];B[gf](;W[jc]C[shi: oh?...very logicalinitially it looks very strange, but it seems the only move];B[cj]C[keeping up the pressure on this group];W[di];B[fg];W[ci];B[lc](;W[jf];B[ig](;W[qn](;B[ie];W[if];B[hd];W[hf];B[hc];W[hb];B[gc]C[Shi: this explains why W didn't play K13 earlier as that move would be uselessso though it is locally bad for W, B spends an extra move];W[nq]C[liu: B is very solid now, what do you think?shi: I prefer W];B[np];W[mp];B[no];W[pq];B[qq];W[oq](;B[pn];W[op];B[po];W[qr];B[rp];W[rq];B[qp]C[liu: this is the first time I've seen thisfan: alphago still sees it as completely even];W[nc]C[settling the W group];B[nd];W[mc];B[md];W[lb];B[ld];W[kc](;B[qj];W[oc];B[od];W[cm];B[rr]C[not so big for points but huge as it is thick - if W got this, there would be problems with Q9](;W[qh]C[and alphago is very good at reducing the efficiency of opponent groups, such as with Q10](;B[lq]C[shi: tenuki is very strange we would normally try to attack when B has so much supportliu: W has two good points on the right at R15 with a hane and connect in sente and Q10, so perhaps it isn't so easy to attack, so harassing this is bettershi: and the invasion is probably good timing before B gets Q17, as W can make use of B not being aliveFan: if B tries to attack now, perhaps there isn't any use. If there isn't a good move locally, alphago will tenukishi: and i think this is not necessary a double attack, this is just a big point. W gets a big point on the right, so B needs to take this big pointliu: yes, perhaps it isn't so complicated (as double attack)];W[mq];B[kp]C[Fan: with students, we say M3 loses terribly much, and you can't play it, but now we can't dare to say that.Liu: we can't dare to teach anything anymore (laughter)];W[rs](;B[sq];W[ln];B[hq];W[qe]C[liu: now B really doesn't look like a thick group anymoreonly groups with two eyes can be considered thick, so even the S2 group can't be considered thick.](;B[kr]C[fan: this looks very strangeI spent some time checking if this was a misprintShi: but this?hmm, this is better for B's eyespaceLiu: if playing online, you would assume this is a misclick](;W[sr]C[wow!Shi: normally if B was scared of this ko, B would play T2.liu: it doesn't seem reasonable for B to connect at S3 because of the problem at Q9but B also can't really afford to lose this ko as then the efficiency of the L2 is called into questionso this is a very good way to harass BFan: but actually if you look, W doesn't have ko threats];B[ss];W[em];B[pr]C[locally, W loses the eyespace and points, as the T2 throw in removes W's Q1 ko, but W forces B to spend a move on the 2nd lineW is very fast, getting 4 moves on the outside and the bottom right is lightB spent a lot of moves on the 2nd line to attack.];W[pj];B[pk];W[oj];B[qk];W[of]C[Fan: W is truly attacking B nowFan: before we see the brillliant upcoming fight, let's take some time to judge the position and what both sides have as their purpose.... after countingFan: what is W's aim?Shi: W is aiming to save the N4 group in sente via attacking B and forcing B to run on dame and getting sente to protect the left side, say C4. Then B will find it hard to pay komi.Liu: B's territory is about 46 with no potential, it will be hard for it to surpass 50.but if W spends a move on the left, the left may already be 40 points.](;B[pg]C[shi: still trying to find W's weaknessesliu: W already has 5 stones hereFan: the best defence is attackshi: in a real game you wouldn't necessarily have the courage to play here as B is in such danger and heavy](;W[og]C[shi: it is easy for B to get harassed];B[ph];W[qg]C[B's two stones are basically dead];B[pc](;W[qb]C[not even allowing an eye in sente];B[pf];W[pe];B[rd](;W[re];B[oh];W[nh];B[ni](;W[oi];B[mh];W[ng];B[mg];W[mf];B[lf](;W[mi]C[very violent];B[nj](;W[qf];B[pb];W[rc];B[rb];W[sb];B[ra](;W[qc];B[ok]C[very painful for W];W[pi];B[kg];W[ke];B[qd];W[sd];B[qa];W[sc];B[le];W[nk];B[mj];W[li];B[lj];W[ki](;B[hg]C[fan: somehow B connects up with one move](;W[kj]C[sente on the cut as there is the ladder](;B[kb];W[jb];B[mb];W[id];B[fd];W[ka];B[nb](;W[lp](;B[nl]C[fan: it seems like this is a big B success, not only living with points, but not really losing anythingliu: so it is now a big advantage to B?fan: alphago still thinks it is a close game and tight];W[ko];B[jp];W[gr]C[if B responds and W gets C4, while W didn't really profit much from the attack, W can get the last big point.Shi: and W succeeds in the plan we set for it - to save N4 in sente and to take C4];B[in];W[ll];B[mk];W[il];B[cq];W[cr];B[br]C[fan: from here, gradually W feels the position isn't so good and clarity appears in B's favourliu: this sort of advantage, do you mean a large onefan: no, I still mean very very small and close. And alphago thinks it is increasingly likely that B wins by 0.5 point];W[jg];B[ma];W[lr];B[kq];W[hr](;B[go];W[cp];B[bq];W[iq];B[ip];W[jr];B[mr]C[B can't accept W connecting under with lots of points];W[jo];B[io];W[hm];B[km];W[lm];B[hn];W[gm];B[gq];W[fq];B[gn]C[W does profit and P2 is sente];W[dr];B[cn]C[not scared of W killing the corner, saying the side is bigger];W[bm];B[dp](;W[do];B[dn];W[ep];B[en];W[dm];B[bp]C[liu: it is unfortunate W didn't get the chance to atari P2 in sente](;W[kh];B[gj];W[fi];B[ii];W[gh];B[gg];W[lg];B[fj];W[ej];B[eh];W[dg];B[kk];W[kl];B[jk](;W[im]C[fan: very interestingliu: this looks like it clearly loses a point as K6 is sente...liu: very precise, and we know this is a 0.5 victory, so every point counts];B[fh];W[bn];B[ei]C[fan: B can't let W get this in senteliu: it does look like B has a lot of ko threats whereas W only has 1];W[bo];B[dp];W[eo];B[fn];W[ek];B[cp];W[ri];B[rj];W[bs]C[comfortable];B[jn];W[jl];B[ls]C[preventing P2 sente];W[or];B[dh]C[shi: very precise and necessaryfan: critically important for the ko threats this gets];W[ch];B[ic];W[ib];B[qs];W[js](;B[ks]C[very precise, not clear whether to play O1 or not - fan: this is for a few extra ko threats];W[oo];B[nn];W[ae];B[jd];W[ik];B[ir];W[is];B[jq];W[ir];B[me];W[nf];B[kd];W[la];B[jh];W[kf];B[ij];W[ml];B[je];W[kg];B[jj];W[hk];B[fl];W[fm];B[gl];W[om];B[on];W[nm];B[pm];W[hj];B[el]C[shi: interesting, so these two pushes are miai];W[dl];B[ji];W[lh];B[hi](;W[ol];B[de];W[ce];B[pl];W[nk];B[dj];W[dk];B[nl];W[ns];B[ms];W[nk];B[mo];W[lo];B[nl];W[gp];B[hp];W[nk];B[bj];W[bi];B[nl];W[ob];B[oa];W[nk];B[bg];W[cg];B[nl];W[mm]C[liu: B has too many ko threats];B[ea];W[ga];B[qi];W[sh];B[ai];W[ah];B[aj];W[bh];B[ao];W[ad];B[ac];W[an];B[sj];W[nk];B[lk];W[ap];B[si];W[rh];B[nl]C[B wins this ko and W is unable to throw in at Q1so B wins by 0.5 pointnote W needs another move at B9shi: its endgame is so precise])(;W[de]C[shi: isn't this 1/3 point better as W no longer has to spend another move inside the arealiu: probably the 1/3 point doesn't affect the game result]))(;B[ns];W[os];B[nr]C[B gets 2 points not 1, but the R2 potential is gone from ko to nothingbut B doesn't have the J2 etc ko threats]))(;W[jl]C[normal];B[fl]C[shi: oh, very interesting];W[fm];B[ek];W[ei];B[dk];W[hj];B[hk];W[hi];B[ik];W[gi];B[lk];W[gk];B[gl];W[fk];B[hl]C[now this becomes relevant];W[im];B[el]))(;W[eh]C[liu: like this W seems to make a lot of points, B has 60, W has close to 60];B[kk]C[fan: I was surprised too, isn't W alive, but alphago was going to play here];W[kl];B[jk];W[jl];B[ii]C[shi: nice move];W[jh];B[gj]C[and B is threatening J17](;W[ic];B[fk])(;W[fk];B[ic];W[jd];B[ib];W[la];B[ia];W[ja])))(;W[eo](;B[bn];W[do];B[bp];W[ao];B[an];W[bs];B[dm];W[ck]C[liu: I find Fan is a bit unsure now when faced with Shi Yue's famous killing ability];B[dl];W[dj];B[dn];W[en];B[fl]C[probably B can utilise weaknesses at H5 and L9])(;B[dm]C[shi: oh directly here is even better];W[ck]C[necessary];B[do]C[W in trouble])))(;B[ir]C[liu: what about this?];W[hp];B[mr];W[iq]C[this is still senteshi: ];B[];W[bq]C[shi: this is powerful if W is very solid];B[cp];W[bs]))(;B[ko];W[lk];B[lo];W[mo];B[mn];W[nn];B[mm];W[oo];B[nm];W[lr];B[kq];W[ms]))(;W[lk]C[fan: why not here];B[mo];W[lo];B[lp];W[oo];B[nn]C[fan: oh]))(;B[mb]C[fan: why not this side?...liu: perhaps it thinks this loses too many points];W[nb];B[ml];W[jg];B[jh](;W[id];B[fd];W[kh];B[kb];W[ma];B[kd];W[jd];B[jb];W[ih]C[laddered];B[ic];W[je];B[ib]C[W is dead])(;W[je]C[liu: perhaps alphago is scared of this variation - there is still a cut])))(;W[jg];B[jh];W[id];B[fd];W[kh];B[mb](;W[nb];B[kb];W[ma];B[jb]C[W dies])(;W[kb];B[nb];W[ma];B[ib])(;W[jb])))(;B[jg]C[this would be a disaster]))(;W[qa]C[Q19 is sente for N18];B[pa];W[sa];B[mb];W[nb];B[na];W[ma];B[ob]))(;W[lg]C[liu: W just doesn't have the ladder due to H3];B[lh];W[li];B[kh];W[nk]))(;W[qf]C[W could have just protected without giving up the P9 sente];B[kh]C[and then alphago would play hereLiu: this is tewari of];AE[pe][lf][mf][pf][qf][mg][ng][og][pg][qg][kh][mh][nh][oh][ph][ni][oi];W[];B[mg];W[mf];B[lf];W[ng];B[mh]C[this is unreasonble as there is the M13, but the game is as if W responded];W[nh];B[ni];W[oi];B[kh];W[];B[pf]C[these exchange lose a few points, but not too much];W[pe];B[pg];W[qg];B[ph];W[og];B[oh];W[qf]C[as now P10 also loses libertiesso this result looks good for BFan: and perhaps why W could have gotten this variation so easily but sacrificed shape with the N11 atari.]))(;W[mg]C[shi? why not here?];B[nj];W[rc]C[and then here?Fan: alphago agrees];B[qi](;W[kg]C[defending against L15 and then M13];B[km]C[alphago thought this was also a reasonable gameboth sides gain about 40 points, M12 is sente, W gets sente, and there is still aji with the M6 groupshi: this is why B poked at Q13 rather than just running as B could sacrifice the wallliu: that poke looks simple, just one move, but contains a myriad of variations behind it to read and judge])(;W[lh]C[shi: why not here, this helps the centre a lot morefan: perhaps F16 is now sente?Liu: W will give up L18 thoughfan: and p15 is senteshi: oh, yes, ...];B[oe](;W[];B[qd];W[sd];B[pb];W[rh];B[pa]C[double sente on N18 and S18])(;W[qd];B[kg]C[fan: this seems to give B even more sente on the outside]))))(;W[pb]C[fan: I seriously wondered why not here];B[pi];W[qi];B[oi]C[wow];W[ni];B[nj];W[ok];B[nh];W[mi];B[oh]C[shi: this looks unacceptable for WB's wall looks thick again]))(;W[rc]C[normally you would think here ];B[ob]C[alphago would play here];W[nb];B[rb](;W[qb];B[rd];W[sb];B[ra];W[qc];B[qd];W[pb];B[pf];W[pe];B[qf];W[re];B[rf];W[sd];B[oe];W[se](;B[mg])(;B[oh];W[mg]C[shi: even like this B can be content perhaps]))(;W[rd];B[oa]C[B gets an eye in sente])))(;W[qg]C[if here];B[lg]C[alphago plays here, and it is hard for W to keep up the attack](;W[];B[pf];W[pe];B[og]C[this is severe])(;W[ng];B[kh];W[lf]C[shi: if here?];B[mf]C[liu: B is probably ok...];W[le](;B[pf];W[mg];B[pe])(;B[nf]C[not ok];W[mg];B[og];W[me];B[ne];W[oh]))))(;B[mg]C[shi: my first feeling is still to run];W[mf]C[fan: alphago would play here];B[lf];W[ng]C[W attacks while fixing the weaknesses];B[lh];W[ke];B[le];W[lj]C[liu: this is how W will get sente to protect the left and win like we said before])(;B[lg]C[if directly ];W[ok](;B[ng];W[mh];B[mg];W[nm]C[B gets sente, alphago thought this was also possible])(;B[pg]C[no longer possible];W[og];B[ph];W[qg])))(;W[pk]C[fan: in checking many of the variations such as if L2 was at M2, W is aiming to play here, so T2 probably really does matter]))(;B[jn]C[shi: to follow up here, I would normally consider this or])(;B[lr]C[this would mean W immediately has only half an eyeactually alphago thought this was also possible ];W[nn]C[I checked what alphago was thinking];B[kn];W[ko]C[this is now possible]))(;B[sr]C[liu: it is hard to choose between these movesthe difference is very tiny, and it is very hard to judgeshi: if W connects at Q2, this is slightly better for points, but if B is surrounded T3 is more flexible shape to liveand if W tries to make life with Q1, then capturing may be better to leave the move at T1 to attack W's eyespace.Liu: but really, all these possibilities are very unlikely, so it is very hard to choose.]))(;B[kp];W[mo];B[lq];W[nn]C[shi: peeping first is a shape and very good order as if you attack first, W can get shape with this ataribut if you peep first, O5 is light]))(;W[mo]C[Shi: I would probably play here as W is very wide as welland as W has Q17, invading the right is not so urgentLiu: me tooFan: for alphago, this was another option it could have chosen, both very bigprobably it invades because it is still eyeing the life and death of the b wall above]))(;B[oc]C[Fan: we were always taught this was very big so why not here?];W[qi]C[Shi: perhaps it is as W is completely alive, so P17 is purely points and not so urgent]))(;B[qp]C[Liu: perhaps B avoids this because F4 makes W more solid];W[qk]))(;B[nq];W[nc]C[if B responds, W would start a fight]))(;W[jg]C[Fan: because it isn't so easy for W to play locally];B[ie];W[if];B[hd];W[hc];B[hf];W[je];B[ih]C[B can peep at L18 and then at L16, very painful]))(;W[je]C[more standard];B[hf]C[but alphago would play here])(;W[ig]C[Shi: but then why not here?](;B[hf];W[if])(;B[hd];W[hc];B[id];W[je]C[Liu: W is happy])))(;W[id]C[normally you would think about thisLiu: but you fear the peepShi: yes true];B[hc];W[gc];B[hf])(;W[jd]C[this might be even worse];B[hf];W[ie];B[hc]))(;W[df]C[Shi: If I played, it would be here as I think B is completely alive, and it is not as if W has a helping stone at C10 for the fightFan: I think it is because in the game W can protect nicely with one move, so it expects B to pull out immediately, but like this, W doesn't have a good local follow up either. My understanding that this isn't as good as forcing B out for a fight because];B[ck]C[Liu: for example, though W doesn't need to protect yet, there is still the aji of F15so W still owes a move, but if W protects it is worse than if D14 was at C14]))(;B[ec];W[fd]C[W is preparing this];B[ed];W[ee](;AE[fb][gb][cc][dc][ec][fc][dd][ed][fd][ee];B[ed]C[Liu: this is very similar to];W[ee];B[cc];W[dd];B[dc];W[fd];B[ec];W[fc];B[fe];W[ef];B[bd];W[be];B[bc];W[cf]C[Liu: in this joseki, F18 is a bad exchange](;B[fb];W[gb])(;B[ge]))(;B[fe]C[Fan: and alphago will also ];W[ef];B[bd];W[be];B[bc](;W[he]C[but W will play hereLiu: that's a very strange move!];B[ce]C[if here];W[ff]C[alphago just defendsLiu: very strange!Fan: alphago thinks W is slightly better. it emphasises centre influenceLiu: and it uses it very well])(;W[ge]C[Liu: we should emphasise that if W has the ladder it should still be here])))(;B[ed]C[shi: so why doesn't it play here in the first place?liu: ah that's a good idea];W[ee]C[shi: probably because W is not scared of E17 as that is the same as the other variation];B[eb];W[fd]C[W won't tiger anymore]))(;B[ec]C[normally alphago plays here];W[ed]))(;W[ed]C[Shi: I'm not sure what the difference with this move isFan: probably it is dependent on the surroundings](;B[ec](;W[fc]C[the same])(;W[fd]C[or here]))(;B[fc])))(;W[ep]C[it almost never plays here, sometimes it tenukis and even less it pincers]))