Bugged AI review in SGF Library (9 stone handicap game)

The AI review says that White is ahead in the final position, which is untrue (as shown by my local KataGo as well as the Fox Go server’s AI). When I try to play it out, it seems that the AI has some confusions about the top right corner, as evidenced by some invalid invasion attempts as well as suggesting suicide moves for White in the end.

I’ve tried adding a level IV review to see if more playouts makes any difference.

I have seen in other games the estimate score be incorrect in final positions. So I’m wondering if something changed recently.

Definitely something wrong with the OP’s game ai review, it’s more consistent with the AI not getting the board state properly.

Do @mprshblnght have the sgf file, maybe there’s a different formatting being used for the handicap stones?

I’ve overwritten the original sgf file while debugging, but choosing “Download SGF” in the uploaded game on OGS should give the original SGF file, except that the file name is changed.

73724727-sgf-星阵-bbbbbbbbba.sgf (634 Bytes)

1 Like

Troubleshooting it also I think it’s to do with the AB[.][.][.]… to add black stones in the beginning

If they’re placed like B[.];B[.]; as below the AI seems like it does a better job

(;CA[UTF-8]
AP[Sabaki:0.52.0]
GN[name]
PW[white]
PB[black]
HA[9]
KM[9]
RE[B+R]
SZ[19]
ST[17]
RU[chinese]
B[pd];B[dp];B[dd];B[pp];B[dj];B[pj];B[jd];B[jp];B[jj];W[qd];B[pc];W[qf];B[pe];W[rg];B[qe];W[re];B[rd];W[pf];B[se];W[mg];B[of];W[og];B[nf];W[ng];B[ph];W[oi];B[pg];W[pi];B[rf];W[qj];B[rh];W[pk];B[oj];W[ri];B[qh];W[qi];B[ok];W[pl];B[ol];W[pm];B[nn];W[oo];B[on];W[pn];B[po];W[mk];B[nj];W[ro];B[oh];W[ni];B[mj];W[mi];B[lj];W[nq];B[np];W[oq];B[op];W[lq];B[lp];W[kq];B[kp];W[ir];B[jq];W[jr];B[mq];W[mr];B[mp];W[cn];B[co];W[dn];B[fq];W[dl];B[pq];W[cc];B[nr];W[lr];B[or];W[dc];B[ed];W[ec];B[fc];W[fb];B[gc];W[gb];B[hc];W[be];B[dg];W[fl];B[fj])

The original say looks like

(;CA[utf8]
AP[golaxy]
GN[自由战]
PW[星阵]
PB[bbbbbbbbba]
WR[3300]
BR[900]
HA[9]
RE[B+R]
KM[9]
SZ[19]
ST[17]
RU[chinese]
AB[pd][dp][dd][pp][dj][pj][jd][jp][jj];W[qd];B[pc];W[qf];B[pe];W[rg];B[qe];W[re];B[rd];W[pf];B[se];W[mg];B[of];W[og];B[nf];W[ng];B[ph];W[oi];B[pg];W[pi];B[rf];W[qj];B[rh];W[pk];B[oj];W[ri];B[qh];W[qi];B[ok];W[pl];B[ol];W[pm];B[nn];W[oo];B[on];W[pn];B[po];W[mk];B[nj];W[ro];B[oh];W[ni];B[mj];W[mi];B[lj];W[nq];B[np];W[oq];B[op];W[lq];B[lp];W[kq];B[kp];W[ir];B[jq];W[jr];B[mq];W[mr];B[mp];W[cn];B[co];W[dn];B[fq];W[dl];B[pq];W[cc];B[nr];W[lr];B[or];W[dc];B[ed];W[ec];B[fc];W[fb];B[gc];W[gb];B[hc];W[be];B[dg];W[fl];B[fj])

Might be something one could ask @anoek or @GreenAsJade for support for.

If we’re making the parser slightly more flexible, there’s also

3 Likes

Actually, I just realized that OGS handicap games with the Chinese rules always ask Black to place the handicap stones manually, while with the Japanese rules the stones are placed automatically. Has this always been a workaround for this issue?

Some Rulesets use free handicap, I guess Chinese is one of them