1. Start after the game
The intended workflow begins only once a game has ended. Go AI Coach is not designed to suggest moves during an active game, and it should not be used to bypass the rules of a server, club, event or tournament.
2. Reconstruct the intention
A numerical swing can show that a move mattered. It does not necessarily show why the move felt reasonable. Reasoning Replay is intended to help the player name what they noticed, feared or tried to accomplish at that moment.
That intention becomes a hypothesis to examine—not a fact inferred with certainty and not a judgement about the player.
3. Compare the idea with the position
The product design separates two layers:
| Layer | Role |
|---|---|
| Validated position evidence | Board state, legal moves, candidate comparisons and other deterministic facts remain authoritative. |
| Coaching language | Generated language may organise and explain validated evidence, but should not invent board facts or overstate close decisions. |
4. Ask “Why not my move?”
A natural-looking move can answer the wrong question. The intended comparison explores what the player’s move achieved, what it left unanswered and why another priority mattered more.
Where alternatives are close, the explanation should say so. A small difference should not become a dramatic mistake simply because dramatic wording sounds more decisive.
5. Build a reusable lesson
The review should end with something the player can recognise again: a whole-board priority, shape issue, reading habit, direction-of-play choice or another recurring decision pattern.
Patterns and drills are intended to turn that lesson into deliberate practice rather than leaving it inside a single game report.
What we are not promising
- No guaranteed rank improvement.
- No replacement for a qualified human teacher.
- No official relationship with a Go federation, server or tournament.
- No live-game assistance.
- No public release date on this website yet.