A history of joseki habits · 19
AI-era joseki feels strangest when it exchanges less, tenukies early, and returns later
Early 3-3 changed the definition of completion, not merely one corner answer. Omitting three moves, preserving cuts, and taking another corner can be more efficient than cleaning locally—if you keep managing the unfinished aji.
Early 3-3 changed finished into no move needed now
After a bare 4-4 is invaded at 3-3, the deep-search line reaches Q17–R16–Q15–S14 and spends Black 7 on an empty corner. This is not a seven-move tenuki rule. It asks local R14 to bid against Q4 and other global points after the same six-move prefix. The figure reports endpoint means and ranges across eight board symmetries at 2,048 visits each, so search-direction noise remains visible.
After one identical six-move prefix, compare Black 7 at R14 and Q4 with move count, turn, and starting conditions held fixed. The gain from not exchanging is where the next move lands
Put the reply on the board and opportunity cost becomes concrete. If Black 7 continues at R14, White 8 is nearly forced to answer S13, committing one move from each side locally. If Black 7 turns to Q4, White 8 takes D4 and both moves expand whole-board coverage. Both lines contain eight moves and leave Black to move, so their endpoints are comparable; the delta is still a whole-line difference, not R14's fixed price.
Two eight-move endpoints: local R14–S13 versus global Q4–D4, with win rate, score, and symmetry ranges together. Playing fewer moves is not tenuki skill: one wrong cutoff can lose ten points
The previous page called Q16–R17–Q17–R16–Q15–R15–P16 a seven-move pause. This audit does not rescue it with prose: its eight-symmetry mean leaves Black about 10.6 points behind, while the deep-search six-move exchange followed by Q4 is down roughly one point. Both lines use seven moves and leave White to play. The shape—not brevity—causes the difference, so the retired line remains visible as a falsifying counterexample.
Two seven-move pauses: A is retired by audit and B follows deep search. Shorter is not itself efficient. A 2026 world event treats five local moves then tenuki as ordinary language
In Shin Jinseo–Liao Yuanhe at the 2026 Chunlan Cup, Black 5 invaded the upper-left 3-3. Only five local moves followed before White 10 turned to Q4 in the lower right. Read19's existing study also checked that pause with KataGo. The comparison places the game beside a hypothetical White 10 at G18: one is more locally complete, the game claims another corner. The efficiency gap is simply where White spent the same move.
The game turns White 10 to Q4; the teaching line stays at G18. One coordinate makes the whole-board efficiency ledger visible.
AI explanation, data, and practice
Before tenuki
- How many moves are omitted, and who keeps the turn?
- Can the opponent exploit the unfinished point now?
- Does outside influence have an immediate use?
- Reread after every nearby stone