DieterVerhofstadt/Patterns in mistakes

Sub-page of DieterVerhofstadt

Table of contents


In 2021 & 2022 I've analyzed close to 100 games with KataGo. In each game I have identified the 3 biggest mistakes, defined by the biggest shift or swing in points as indicated by KataGo. This results in about 200 "mistakes". See Dieter's Mistakes In 2021 and Dieter's Mistakes in 2022. Two assumptions inspired me to do this project:

  1. I will find patterns in these mistakes, which show a possible improvement to my game
  2. Patterns in mistakes AI finds in my game, are relevant to other amateurs or lower dans.

Indeed, patterns emerged, while you can argue the human mind is built to find patterns even where there are none.


One is a "non-pattern", where I found the sequences supporting KataGo's suggested move too hard to interpret. Usually it's about the relative strengths of multiple groups, which can move from healthy to being in danger, depending on cuts and connections around the board. With my conceptual understanding, the improvement just boils down to "reading very deeply" - which is the basic criticism on reviewing with AI. A related known concept is global thickness or atsusa.

Another is a "trivial pattern", where I missed a sequence on my level or below, mostly about the life and death status of a single group or the liberties in a 1-1 capturing race.

But there are more ...

Strategic patterns

Strengthen the corner in sente

The corner is important in the opening and remains so in the middle game. When facing a choice where to reinforce, if you can do so with the corner and especially if it takes sente, that's probably the better choice

Strengthen groups that build or erase

Cut & connect

Slow connections

One emerging pattern is the slow connection. I have a tendency to make hanging connections, table shapes, bamboo joints ... without too much thinking, even because of the visual shape, while it protects against a cut which is not necessarily harmful or urgent. They look like honte and maybe can be considered as such, but AI finds them slow to the point of losing 5-10 points.

Heavy cuts

Closely related are cuts or reinforcements of cutting stones, where the opponent's groups are already relatively strong and the cutting stones end up as a heavy (central) group.

Small connections or cuts

Vital connections or cuts

Applying pressure

Another, more difficult pattern, yet within my reach, is that of "applying pressure", which reminds me of John Fairbairn's promoting the term ijime, which he translates as "bullying". There are two subsets of this pattern.*

Harass before living

The second one applies to situations where one is under such an attack. Rather than being "bullied into small life", there's often an option to counterattack from the other side, "bullying the opponent into small capture". Or even better, bullying the attacker from the outside, before accepting small life. You can think of this as a flavor of forcing moves.

Surround in sente

A major strategic principle I (re-)discovered is to surround in sente. The sente part means that if the opponent doesn't answer, the group can be killed. If surrounding means it can be harassed next to gain some points, that may be offset by the opponent's move elsewhere. So if surrounding doesn't threaten to kill, then it might be better to play elsewhere myself. It requires life and death skills to make such a judgment so it's not a trivial decision process.

Sacrificing the small, attacking the whole

When attacking, threaten the whole not just a part. Attack the bigger group. When defending, sacrifice the small regain the initiative.

Technical patterns

Killing techniques

In the following cases, I was aware of the opportunity to kill (or live) but didn't play the proper move.

Cutting or connecting techniques

In some cases the cut (or connection) is valuable but the choice for the cutting technique is wrong. Usually I'm employing a move that looks like good shape but in this case is not working as well as another move, which is more crude at the surface.

Invasion points

The last pattern is related to (my failure in) dealing with moyos. I tend to invade too deeply, or too soon. I'm pretty good at following AI's opening gospel, but when my opponent builds a moyo from consecutive exchanges of influence for territory, I'm not as good in reducing or invading those, or timing an invasion. KataGo is pretty comfortable at "high invasions": creating a living group in the middle of a huge moyo or with the option of escaping. It will often prefer those to the deeper invasions, which can be more easily killed or "bullied into submission". It will create such central groups fairly late in the game, later at least than my (premature) side invasion. When the moyo is smaller, KataGo will often choose not to invade and rather build elsewhere to match the potential territory formed by the moyo.

Take the mutual shape point (in sente)

See local ko threats

Shortage of liberties


Raw peep

KataGo is not so fond of peeps because they remove the possibility of an atari later. The extra influence acquired doesn't usually offset this. So KataGo has a lower tolerance for raw peeps.

Endgame (sente)

Nearly every big mistake in the endgame is about losing sente


Many of my mistakes or even blunders were due to a lack of awareness about a situation changing due to activity in its vicinity. In other cases I was just too lazy to even consider quite obviously better moves than the one I played.

Killing opportunities or threats


Given these patterns I'm going to play my next 10 games with this compass:

  • North: surround in sente / avoid being surrounded in sente
  • West: be aware of changing conditions to stable situations when activity spreads into their vicinity
  • East: cut and connect vulnerable groups & figure out the best way of doing so
  • South: (practice more tesuji like problems)

DieterVerhofstadt/Patterns in mistakes last edited by Dieter on March 18, 2022 - 16:41
RecentChanges · StartingPoints · About
Edit page ·Search · Related · Page info · Latest diff
[Welcome to Sensei's Library!]
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Login / Prefs
Sensei's Library