(;FF[4] CA[UTF-8] GM[1] GN[AGT + KataGo self-play - game 1] PC[https://online-go.com/review/431592] PB[AlphaGo Teach and KataGo] PW[AlphaGo Teach and KataGo] BR[9p] WR[9p] OT[Error: time control missing] RE[?] SZ[19] KM[7.5] RU[Chinese] ;B[qd] ;W[pp] ;B[dd] ;W[cp] ;B[nd] ;W[fp] ;B[qn] ;W[cf] ;B[nq] ;W[np] ;B[mp] ;W[no] ;B[oq] ;W[pq] ;B[pm] ;W[qo] ;B[jq] ;W[fc] ;B[fd] ;W[gd] ;B[fe] ;W[ec] ;B[dc] ;W[dg]C[White 55 % (AlphaGo Teach) LZ-258 says 57%, near enough ] ;B[be] (;W[ge] ;B[gf] ;W[hf] ;B[fg] ;W[hg] (;B[ce]C[Does black need to play here? My intuition (xela, Australian 2d) is a move on the top side instead? LZ-258 has white at 54% after this move, and thinks anything else is -3pp or worse. Best alternatives actually are on the left side, not the top. Looking at the variations, it's not that black's afraid of the D17 stones being cut off, the issue is that white can push at C15 and take territory in the corner. ] ;W[lc] ;B[dj] (;W[me]C[How bad would it be if white tried to run out with E12 instead? LZ-258 says bad but not terrible, -7pp. White doesn't get attacked strongly, it's just gote, so that black gets to reduce the top. ] ;B[ro] ;W[pn] ;B[qp] ;W[po] ;B[rm] ;W[ne] (;B[cm]C[I (xela) wondered if black could extend further. The game move is LZ-258's top choice by about 2pp ] ;W[ei] ;B[ej]C[Here I (xela) expected white to jump out instead, but LZ-258 thinks that's a mistake. It surprised me with its suggested alternative. ] (;W[fj] ;B[eh] ;W[dl] ;B[bk]C[I (xela) expected white's next move to be somewhere on the right, to break up black's area. But LZ-258 isn't interested in the right side at all. ] (;W[hq]C[LZ-258's first choice. White is still at +53%. ] ;B[hp] (;W[iq]C[Looks like a weird shape to me (xela), but LZ-258 agrees with this as first choice move. Why not hane instead? See variation. ] ;B[ip] ;W[gp] ;B[kp] ;W[jo] ;B[nm] ;W[ql] ;B[qm] ;W[rp] ;B[qq] ;W[so] ;B[rn] ;W[qr] ;B[rq] (;W[sq]C[LZ-258 initially isn't a fan of starting the ko, although it thinks white is winning either way. LZ needs 15k playouts before it admits that this is the best move, W+63% now. KataGo (42,353 playouts) has white at 55 % at this point. ] ;B[rr] ;W[rs] ;B[sp] ;W[or] (;B[jp]C[Why doesn't black resolve the ko instead? This position is W+66% (LZ-258) KataGo says white 56 %. ] ;W[rp] ;B[dq] ;W[dp] ;B[sp] (;W[cj]C[This wasn't on LZ-258's radar at all! But once it's on the board, LZ rates it as +2pp compared to N3. ] ;B[di] ;W[rp] (;B[ck]C[To me (xela) this doesn't feel like a ko threat, but LZ-258 agrees with it. White still at 62% here. White 57 % according to KataGo. ] ;W[sr] ;B[qj] ;W[ok] ;B[pl] ;W[pi] ;B[qi] ;W[hn]C[Another example of leaving the shape unfinished, but this time I (xela) can see the sense in it. Again, LZ-258 agrees with KataGo on this one. ] ;B[in] ;W[im] ;B[fr]C[Wow, can black really live in here? Probably not, but there is aji here: it springs to life later in the game... ] ;W[gr] ;B[cq] ;W[er] ;B[eq] ;W[fq] (;B[jm]C[Apparently black can't push through at E4, but the variations are pretty interesting. White at 66% now (LZ-258). White 60 % according to KataGo. ] (;W[bp]C[This is the interesting bit. K7 was threatening the E4 push! See variation. (LZ-258) ] ;B[od] ;W[pg] ;B[en] ;W[eo]C[Weird shape, but LZ-258 agrees with KataGo. ] ;B[ie] ;W[gg]C[For once the AI follows a human proverb, "force before defending"! ] ;B[ff] ;W[he] ;B[hb] ;W[ic] ;B[gc] ;W[ib] ;B[gb] (;W[jn]C[Looks weird to me, I didn't think saving K5 was all that big. But LZ-258 doesn't mind this move (not on LZ's radar, but +0.2pp when I put it on the board). White now at 67%. White 62 % according to KataGo. ] ;B[km] ;W[io] ;B[kn] ;W[kk] (;B[mk]C[I (xela) wondered if black can break into the centre with an attachment: see variation. ] ;W[mj] ;B[oi] ;W[ni] ;B[oh] ;W[pe] (;B[pd]C[Feels passive to me (xela). Can black make better eye shape by dropping down to the second line? Actually, LZ-258 wants to tenuki! It thinks the game move is -3pp, but white is still +69% ] ;W[ph] ;B[lj]C[This gets a bit complicated, but again LZ-258 agrees with KataGo. ] ;W[lk] ;B[oj] ;W[nk] ;B[pj] (;W[ml]C[The M10 sacrifice means that black's connected out P12 in sente. If white doesn't capture? See variation. (LZ-258) ] ;B[ho] ;W[ng] ;B[og] ;W[of] ;B[qf] ;W[qe] ;B[re] ;W[rf] ;B[rg] ;W[il] ;B[nh] ;W[mh] ;B[mg] ;W[nf] ;B[mb] ;W[mc] ;B[nb] ;W[eb]C[Weird shape, not on LZ-258's radar but actually OK, it's a nice semedori play, +2pp when I put it on the board for LZ to look at. White now at 68%. White 65 % according to KataGo. ] ;B[db] ;W[hc] ;B[fa] ;W[mq] ;B[lq] ;W[qg] ;B[sf] ;W[mr] ;B[ko] ;W[fi] ;B[fk]C[Nice move by black; White can't seal black in: see variation to next move. (LZ-258) ] (;W[gk] ;B[gh] ;W[fl] ;B[fh] ;W[hh] ;B[gi]C[A nice little reduction by black, but white is still ahead -- actually, the longer I (xela) let it run, the higher the winrate gets. LZ-258 initially assessed this at W+69% on a couple of thousand playouts, but at 30k it's saying 77%. White 66 % according to KataGo at 17,832 playouts. ] ;W[ek] ;B[lb] ;W[in] ;B[nr] ;W[lr] ;B[kr] ;W[mo] ;B[mm] ;W[ha] ;B[ea] ;W[lp] ;B[kc] ;W[kd] ;B[ld] ;W[md] (;B[jd]C[This feels to me (xela) like an overplay, "looking for a place to resign". But It's in LZ-258's top two. And the game's not over yet, in fact white is down to "only" 70%. White 72 % according to KataGo. ] ;W[jc] ;B[ke] ;W[kb] ;B[ka] ;W[kd] ;B[lh] ;W[mi] ;B[kc] ;W[pk] ;B[qk] ;W[kd] ;B[dr] ;W[fs] ;B[kc] ;W[jb] ;B[kg] ;W[lf] ;B[jh] ;W[ii] ;B[lg] ;W[kd] ;B[jj] ;W[ji] ;B[kj] ;W[li] ;B[ki] ;W[jk] ;B[kc] ;W[kf] ;B[jf] ;W[kd] ;B[ep] ;W[do] ;B[ij] ;W[hj] ;B[kc] ;W[jg] ;B[ig] (;W[kd]C[Actually, white doesn't need to take the ko, can win more simply (to my mind) by squeezing: see variation. (xela) Interestingly, LZ-258's first glance at this position says that taking the ko is a mistake, it takes about 10k playouts to read out the ko fight and decide that white is actually winning by lots (over 80% now). White 90 % according to KataGo. ] ;B[es]C[See comment way back at move 87! (xela) ] ;W[le] ;B[fr] ;W[gs] ;B[bq] ;W[er] ;B[br] ;W[nc] ;B[oc] ;W[hi] ;B[dm] ;W[fn] ;B[go] ;W[fo] ;B[ds] ;W[pf] ;B[rf] ;W[bn] ;B[dk] ;W[al] ;B[bm] ;W[am] ;B[ak] ;W[em]C[White 94 % (KataGo at 141,684 playouts) LZ-258 on 10k playouts has 93% ] )(;W[le]C[LZ-258 finds this easier to read out, W+94% on 10k playouts. ] (;B[kd]C[(LZ-258 says black will play some forcing moves (starting with H11) before connecting, but the idea is the same.) ] ;W[id] ;B[je] ;W[ik] ;B[oe] ;W[kh] ;B[pf] ;W[if]C[White had to give up some stones on the right, but is still miles ahead. (LZ-258) ] )(;B[hi] ;W[ih] ;B[id] ;W[hd]C[KataGo thinks that 216. wM15 gives white 69 % winrate (-22 pp). ] )))(;B[hi]C[First choice by LZ-258: same result, different move order. ] (;W[ii] ;B[jd] )(;W[kb] ;B[ji] ;W[ii] ;B[ij] ;W[ih] ;B[jk] ;W[jh] ;B[ki] ;W[kh]C[KataGo thinks that 177. bH11 gives black 13 % winrate (-15 pp). ] )))(;W[fh]C[This is a disaster, with white's winrate dropping to 7% even though the result doesn't look that dramatic to my eyes.. (LZ-258) KataGo doesn't give this move any playouts (out of 14,948). ] ;B[hk]C[White has a lot of choices here, all of them bad. (LZ-258) ] (;W[hj] ;B[gj] ;W[gi] ;B[gk] ;W[ij] ;B[ik] ;W[jk] ;B[fn] ;W[jl] )(;W[ik]C[Actually the best response according to LZ-258 ] ;B[fm] ;W[hj] ;B[gj] ;W[hi]C[Similar result, black has swallowed up D8 in sente. (LZ-258) ] )(;W[dh]C[LZ-258 finds this crazy variation where white can make a ko in the corner, but still can't turn it into a living shape. ] ;B[eg] ;W[ek] ;B[fl] ;W[dk] ;B[ci] ;W[de] ;B[ed] ;W[bi] ;B[bj] ;W[ch] ;B[cj] ;W[bg] ;B[ah] ;W[em] ;B[dm] ;W[fn] ;B[el] ;W[ai] ;B[ag] ;W[af] ;B[bf] ;W[bh] ;B[ag] ;W[cd] ;B[ee] ;W[bd] ;B[df] ;W[ae] ;B[cg] )))(;W[il]C[This is a game-losing mistake, white's winrate drops to 11%. (LZ-258) KataGo doesn't give this any playouts (out of 22,965). ] ;B[ml] ;W[nj] ;B[nh] ;W[li] ;B[oe] ))(;B[oj]C[LZ-258's first choice KataGo doesn't give this move any playouts (out of 15,176). ] ;W[pd] ;B[pc] ;W[qc] ;B[qb] ;W[rc] ;B[ob] ;W[rb]C[It's starting to look like a 4-4 attach joseki played out in the wrong order! (LZ-258) ] ;B[qe] ;W[mb] )(;B[qb]C[LZ-258 says this is a terrible blunder, -22pp -- but it doesn't go and kill the top group, it just tenukis again. KataGo doesn't give this move any playouts (out of 15,176). ] ;W[ph] ;B[oj] ;W[db]C[I don't understand go! Now white is at 92%. (LZ-258) ] ))(;B[jk]C[LZ-258 says no, this doesn't work, -10pp ] ;W[jj]C[KataGo only devotes 2 playouts (out of 19,903) to 113. bK9 and stops reading here. It says black 30 % winrate (-8.5 pp). ] ;B[il] ;W[qh] ))(;W[db]C[LZ-258's first hoice. ] ;B[fa] ;W[eb] ;B[cb] ;W[hc] ;B[da] ;W[io] ;B[kn] ;W[jn] ;B[km] ;W[kk] ;B[mk] ;W[mj] ;B[oi] ;W[oh] ;B[pj] ;W[ni] ;B[ph] ;W[og] ;B[mb] ;W[il] ;B[lb] ;W[mc] ;B[nc] ;W[kb] ;B[oe] ;W[qf]C[KataGo thinks 108. wD18 gives white 59 % winrate (-3 pp). ] ))(;W[jl]C[-4pp (LZ-258) ] ;B[ep]C[(Actually, when I looked again with more playouts, LZ-258 preferred a different move order, black H7 first, but ended up with the same position.) ] ;W[eo] ;B[do] (;W[co] ;B[dn]C[Notice this time black doesn't force at E2... (LZ-258) ] (;W[bq]C[If white tries to play here as in the previous variation... (LZ-259) ] ;B[hm] ;W[il] ;B[fo] ;W[en] ;B[go]C[The H7 stone sets up a ladder. White collapses. (LZ-258) ] )(;W[km]C[Actually, white should bail out wiht this move, 58% now. (LZ-258) ] ))(;W[bq] ;B[bp] ;W[co] ;B[bo] ;W[dn] ;B[cn] ;W[do] ;B[dr] ;W[br] ;B[es] ;W[cs] ;B[hm] ;W[il] ;B[gn]C[KataGo thinks 94. wK8 gives white 59 % (-1.5 pp). ] )))(;B[ep]C[-8pp (LZ-258) ] ;W[eo] (;B[do] ;W[co] ;B[dr] ;W[fs] ;B[dn] ;W[bq] ;B[br] ;W[ds] ;B[cr] ;W[bp] ;B[ar] ;W[cs]C[Actually, LZ-258 says white should tenuki instead of this move, but either way white is winning. ] )(;B[bp] ;W[bo] ;B[bq] ;W[ln] ;B[jm] ;W[io] ;B[nr] ;W[do] ;B[dr] ;W[fs] ;B[ir] ;W[jn] ;B[nb] ;W[mb] ;B[ph]C[KataGo thinks that 93. bE4 gives black 32 % winrate (-7 pp). ] )))(;B[gc]C[LZ-258 rates this as -14pp KataGo doesn't give this move any playouts (out of 50,643). ] ;W[hc] ;B[sp] ))(;W[mq]C[LZ-258 prefers this ko threat ] ;B[mr] ;W[rp] ;B[fr] ;W[gr] ;B[cq] ;W[lq] ;B[lr] ;W[sr] ;B[eq] ;W[fn] ;B[ep] ;W[eo] ;B[do] ;W[dn] ;B[co] ;W[qj] ;B[pk] ;W[pj] ;B[oj] ;W[oi] ;B[nj] ;W[ni] ;B[mj] ;W[cn] ;B[bp]C[KataGo thinks 74. wN3 gives white 54 % (-2 pp). ] ))(;B[sr]C[Actually only 1pp worse... (LZ-258) ] ;W[mq] ;B[mr] ;W[lq] ;B[lr] ;W[nr]C[LZ-258 suggests a few different moves here (M6 or nearby), but you get the general idea: black's stones are now pretty weak. ] ;B[in] ;W[op] ;B[ln] ;W[pd] ;B[pc] ;W[oc] ;B[pe] ;W[od] ;B[qc] ;W[jp] ;B[kq] ;W[jn] ;B[nq] ;W[mn] ;B[mm] ;W[on] ;B[ob] ;W[nb] ;B[nc] ;W[oe] ;B[mb] ;W[qe] ;B[re] ;W[pf] ;B[rf]C[KataGo thinks 69. bT2 gives black 40 % (-4 pp). ] ))(;W[or]C[LZ-258's first choice on a small number of playouts. ] ;B[rr] ;W[rs] ;B[ko] ;W[sq] ;B[sp] ;W[ol] ;B[sr] ;W[om] ;B[jn] ;W[mq] ;B[mr] ;W[lq] ;B[lr] ;W[nr] ;B[nb] ;W[pe]C[KataGo thinks 64. wP2 gives white 53 % (-2 pp). ] ))(;W[ip]C[LZ-258 still has white at 48% on 10k playouts, but it would be saying 10% if it could see what's at the end of this variation... ] ;B[iq] ;W[ho]C[The ladder's broken, but even on 10k playouts LZ-258 wants to play the first couple of moves. ] ;B[gp] ;W[gq] ;B[go] ;W[di] (;B[gn] ;W[fi] ;B[fq] ;W[eq] ;B[fr] ;W[er] ;B[ep] ;W[fo] ;B[eo]C[Now the ladder works for black! (LZ-258) ] )(;B[fo] ;W[ep] ;B[fi] ;W[gi] ;B[fh] ;W[cj] ;B[ci] ;W[io] ;B[eo] ;W[dp] ;B[dn] ;W[lp] ;B[mq] ;W[mo]C[KataGo thinks 50. wJ4 gives white 34 % winrate (-9.5 pp). ] )))(;W[ql]C[-5pp (LZ-258) ] ;B[qm] (;W[qq]C[LZ-258 suggests this next (actually it wants to go back to H3, but this is second or third choice). Still not at all interested in the area around R10. ] )(;W[rp] ;B[qq] ;W[so] ;B[rn] ;W[qr] ;B[rq] ;W[or] ;B[gq] ;W[fq] ;B[fr] ;W[er]C[KataGo thinks 48. wR8 gives white 48 % winrate (-5 pp). ] ))(;W[qe]C[LZ-258 doesn't like this! -6pp, white down to 47%. ] (;B[pd] ;W[pe] ;B[nm] ;W[ql] ;B[qm] ;W[rp] ;B[qq] ;W[so] ;B[rn] ;W[qr]C[And so on. LZ-258 keeps ignoring the hole at P15/P16. ] )(;B[pe] ;W[qf] ;B[pc] ;W[pf] ;B[qq]C[KataGo thinks 48. wR15 gives white 41 % winrate (-12 pp). ] )))(;W[gi]C[This loses about 4% according to LZ-258. ] ;B[eh] ;W[hq] ;B[hp] ;W[iq] ;B[ip] ;W[gp] ;B[jr] ;W[mq] ;B[mr] ;W[lq] ;B[lr] ;W[lp] ;B[jo] ;W[ck]C[According to KataGo, 44. wG11 gives white 48 % winrate (-5 pp). ] )(;W[di]C[Best according to LZ-258, even though it's classic bad shape!, W+53%, about 0.5pp better than the game move. ] ;B[dp]C[Not sure (xela) why black forces here before continuing, but that's what LZ-258 says. ] ;W[do] ;B[fi] ;W[cj] ;B[ck] ;W[bk] ;B[dl]C[LZ-258 much prefers this rather than either atari. ] ;W[bi]C[Looks like white's group is safe now. Black will play a forcing move then tenuki. (LZ-258) ] (;B[ch] ;W[ci] ;B[dq] )(;B[dq] ;W[cq] ;B[cr] ;W[eo] ;B[bl] ;W[er] ;B[ak] ;W[pd] ;B[pc] ;W[qe] ;B[rd] ;W[oc]C[KataGo's variation for 44. wD11. It gives white's winrate as 50 % (-2.5 pp). ] )))(;B[cn]C[KataGo does not consider this move at all (50,887 playouts). ] ;W[cl] ;B[dl] ;W[ck] ;B[dk] ;W[cj] ;B[ci] ;W[bi] ;B[cm] ;W[ch] ;B[di] ;W[bg]C[Black can tenuki now. W+56% (LZ-258) ] )(;B[co]C[Not on LZ-258's radar (why not? I (xela) picked it as "an AI move"!), scores -1pp ] ;W[dp] (;B[cl] ;W[ei] ;B[di] ;W[eh] ;B[gi] ;W[gh]C[Black tenuki. W+55% (LZ-258) ] )(;B[bp] ;W[bq] ;B[cl]C[KataGo devotes 11 playouts (out of 50,887) to 41. bC5 and gives black 33 % winrate (-15 pp). ] )))(;W[eh]C[White at 48% here. (LZ-258) ] (;B[fh] ;W[ei] ;B[id] )(;B[ro] ;W[pn]C[KataGo only devotes 4 playouts (out of 344,143) to 34. wE12. It gives black 48 % winrate (+ 3 pp). ] ))(;W[pd]C[This is actually LZ-258's top choice ] ;B[pe] ;W[qc] (;B[re] ;W[pc] ;B[oe] ;W[nc] ;B[ld] ;W[mc]C[Black tenuki now (most likely S5) W+52% (LZ-258) ] )(;B[qe] ;W[pc] ;B[mc] ;W[ld] ;B[oe] ;W[ob] ;B[lb] ;W[kb] ;B[pi] ;W[ol] ;B[ro] ;W[pn] ;B[om] ;W[rn] ;B[qm] ;W[rm] ;B[on] ;W[po] ;B[rl] ;W[rp] ;B[sm] ;W[so] ;B[cm] ;W[ql] ;B[pl] ;W[qk] ;B[pk] ;W[qj] ;B[qi] ;W[di] ;B[ei] ;W[cj] ;B[ci]C[KataGo's variation for 34. wQ16. It gives black's winrate as 46 % (+0.5 pp). ] )))(;B[cj]C[Second-best option (LZ-258) KataGo doesn't give this move any playouts (out of 104,256). ] (;W[ce]C[W+58% (LZ-258) ] ;B[ed] ;W[bc] ;B[ch] ;W[bd] ;B[gc] ;W[hc] ;B[gb] ;W[hb] ;B[eb] ;W[kd] )(;W[ed]C[Again, this is terrible, loses 21pp (LZ-258) ] ;B[ee] ;W[de] ;B[gc] ;W[db] ;B[cb] ;W[eb]C[Interesting that white plays the continuation differently with C10 in place. (LZ-258) ] ;B[he] ;W[cc] ;B[hd] ;W[cd]C[Black can tenuki now. B+84% (LZ-258) ] ))(;B[kd]C[-9pp (white at 64% here (LZ-258)) Again, white wants to continue at C15, not push through and cut, similar variations. KataGo doesn't give this move any playouts (out of 104,256). ] ))(;W[ed]C[This is -24pp (LZ-258), terrible move for white! It turns out that white can split black into two groups, but B15 is in the right place to make a living shape in the corner, and white comes off worst. ] ;B[ee] ;W[de]C[After this, black is at 72% (LZ-258) ] (;B[ce] ;W[df] ;B[gc] ;W[gb] ;B[hc] (;W[db] ;B[cb] ;W[hb] ;B[ic] ;W[ib] ;B[jc] ;W[eb] ;B[bc] ;W[jb] ;B[he] ;W[dj]C[Black 75 % (LZ-258) ] )(;W[hb]C[Close second choice (LZ-258) -- end up with the same result, different move order ] ))(;B[gc] ;W[gb] ;B[hc]C[KataGo only devoted 10 playouts (out of 158,750) to 26. wE16. It gives black 55 % winrate (+ 10 pp). ] )))