Drop7 Research

Drop7 Research is built around a single question: what is the best strategy in a game with chance? It hosts a browser version of Drop7, a puzzle game many players regard as a classic, and documents an open-source attempt to build an autonomous player that performs as well as an experienced human. The project frames a million-point score as the long-term target and treats the work as ongoing rather than solved.

The site combines play and research in one place. You can drop discs on the board, watch games play out, and then move into sections covering approaches, engines, research notes, and a leaderboard. Reported results are specific: the strongest candidate so far, learned lookup tables inside a depth-4 search, averaged 516,155 points across 512 unseen games, while a depth-4 expectimax with a hand-written leaf averaged 382,567 on the same games.

This makes the site most useful to puzzle-game fans curious about strategy and to students or researchers interested in expectimax search, learned evaluation functions, and benchmarking under randomness. The vocabulary and learning pages help newcomers build up the concepts before reading the experiment records.

A practical path is to play a few games first, read the rules and the strategy ideas, then browse the approaches and leaderboard to see what has been tried and what remains open. Treat the reported averages as screening results rather than final proof, and check the research notes for context on each candidate.