TengenGoRules & game modes
Which Go does TengenGo play?
Normal games use Chinese area scoring. Experimental modes are separate choices.
Go, Baduk and Weiqi are different names for the same board game. National rulesets change details such as scoring and repetition. Board size and opponent difficulty are separate choices.
The rules in normal play
| Setting | Implemented behaviour |
|---|---|
| Scoring | Chinese area: living stones plus surrounded empty intersections. White receives the chosen komi. Captured stones are not added a second time. |
| Repetition | Positional superko: a move cannot recreate an earlier board position. |
| Suicide | Not allowed. |
| Board sizes | 9 × 9, 13 × 13 and 19 × 19. |
| Komi and handicap | Configurable in game setup. Normal games default to 7.5 komi. |
| Finishing | Two consecutive passes enter scoring. KataGo proposes dead stones; accept the proposal or resume play. Resignation also ends a game. |
Japanese territory rules, Korean rules and AGA rules are not selectable in this release. A Japanese scoring helper in the source does not make a complete Japanese ruleset available.
Choose a mode
| Mode | What changes | Status |
|---|---|---|
| Play KataGo | Normal Go against an engine profile. Ask for hints or review when you want them. | Available |
| Play a friend | An untimed game between accounts using a single-use invitation. Engine help is disabled during play. | Available |
| Review / SGF study | Study a game and try separate variations. Supported SGF imports use Chinese rules; unsupported rules are rejected. | Available |
| Fog of War | A custom hidden-stone practice variant, with collision and sonar clues. | Experimental |
| Multiverse | A custom mode with up to four active parallel boards that can be branched and pruned. | Experimental |
Difficulty is not a ruleset
KataGo is the AI engine. Its software version is separate from TengenGo's app version. Human-SL profiles use human-style move selection; other profiles use search. Their labels are not certified ranks. The hosted service limits search visits to keep a small shared server responsive, so it does not promise unrestricted engine strength or instant answers.
Board Oracle facts, ownership estimates, group-shape overlays and optional language-model explanations have different evidence limits. An estimate or a shape heuristic is not a proof of life.