Learned evaluator makes progress on Drop7 search strategy
What happened
An autonomous strategy study for Drop7, a puzzle game with randomness, pairs an n-tuple lookup-table evaluator with search to raise scores. On September 5, 2026, the setup first beat the frozen fair-evaluation leaf at the screen level; the next day the team reproduced the result using the same table, verified by SHA-256. An October 24 report shows the evaluator averaging 487,066 over 512 unseen public development games at depth-3 search, against 326,717 for a hand-written evaluation function under the same search, about 160,000 points higher. Results remain far below the million-point target, and the work has not entered qualifiers.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageShow HN: Searching for optimal strategies in a puzzle game with chance (Drop7)
一项针对 Drop7 的自主策略研究显示，基于 n-tuple 查表的评估器在深度 3 搜索下、512 局未见过的公开开发局中平均得分 487,066，比同一搜索内手工评估函数（326,717）高出约 16 万分。该结果于 2026-09-05 首次在 screen 层级击败冻结的公平评估叶，次日以 SHA-256 校验的同一查表复现，但仍远低于百万分目标，尚未进入资格赛。
Heat over time
Not enough continuous observations to draw a trend yet.