Module lab 

open-chess additions to the Lichess API, used by the chess training app:

• POST /api/board/game/{id}/assist: the app declares that its engine help was switched on or off during one of our games (ground truth for the fair-play model);
• GET /api/esl/position?fen=: the position dictionary — the deepest stored evaluation of a position, with win/draw/loss odds, and the deepest analysis a training app sent in (a hint);
• POST /api/esl/position: send in the app’s own analysis of a position;
• GET /api/esl/position/moves?fen=: every legal move with its value, where the dictionary knows it;
• POST /api/esl/recognise: read a board from a photo;
• POST /api/esl/photos/{id}/confirm: say what a photo kept for review showed.
• GET /api/esl/tablebase?fen=: the exact result of a position with up to seven pieces, from the tablebase server, kept once fetched.

Structs

ConfirmBody

HintBody

HintMove

One move’s value as the app sends it.

PositionQuery

RecogniseBody

Functions

assist

confirm_photo

POST /api/esl/photos/{id}/confirm: the player who sent a photo says what it showed (they picked the move it matched, or finished setting the position up in the editor). A label for training, marked as the player’s: a reviewer’s label always wins.

position

position_hint

POST /api/esl/position: the training app’s analysis of a position, a hint for others looking it up. One per player and position (the deepest kept); never mixed with the worker’s evaluations. Refused while you’re playing a game here.

position_key

The dictionary key for a position (see chess_analysis::dictionary).

position_moves

GET /api/esl/position/moves?fen=: every legal move, each with the best value the dictionary has for it, from the mover’s view, and White’s odds after it: the worker’s scoring of this position’s moves first, then the evaluation stored for the position the move leads to, then an app’s hint. null where nothing is known.

queue_photo_positions

Queue a confirmed photo’s position(s) in the dictionary for the fair-play worker to evaluate (counting the photos of each).

recognise

Read a board from a photo: with our board reader if there is one, asking the vision language model only when it isn’t sure (or there is no reader). A photo that wasn’t read for sure is kept for the Review page (photoId in the answer) unless CHESS_VISION_REVIEW is off.

tablebase

GET /api/esl/tablebase?fen=: Lichess’s tablebase answer (category, distance to zeroing and to mate, and every move’s result), cached. Like the position dictionary, refused while you’re playing here.