Handbook

game_move_rank v1

Parsed JSON subset returned to callers:

Invariants

Every move_id in ranked_moves must be copied from caller legal_moves (subset allowed).
ranked_moves is non-empty; rank integers are 1..k each exactly once for k rows.
confidence values are numeric in [0,1] (not boolean / NaN).

Input

Field	Type	Notes
`player_view`	JSON object	Seat-filtered projection
`legal_moves`	non-empty list	Minimum `move_id` per row
`objective`	string	Strategic goal description
`heuristic_scores`	JSON object	Keys ⊆ legal `move_id`; numeric values ; may omit keys
`style`	string	`balanced` \| `aggressive` \| `defensive` \| `teaching`
`temperature`, `timeout_sec`	optional	Knobs

Output

Parsed JSON subset returned to callers:

Field	Content
`ranked_moves`	`{move_id, rank, confidence, reason}` rows
`overall_plan`	string summary
`risk_notes`	list of strings

Hidden information

Only encode information in player_view + legal_moves that the active seat may know; heuristic keys must likewise avoid undisclosed futures.