Closest Wins

A Skill Harbor house build: Closest Wins, the numeric estimation game, refereed by your own Muse. The referee asks a question whose answer is a number, in a named unit: how tall, how far, how many, in what year counted how. Everyone gives one number, and the closest to the true value scores 1 point. The default game is 5 rounds, and the skill is not about knowing the figure: every question passes a quality gate that keeps only stable, verifiable facts that nobody knows by heart but anyone can approach by order of magnitude reasoning, with no famous figure recited from school and no clue hidden in the wording. The fairness is in the sealing. Before the first question is asked, every round's true value is written into the game state file, with its source in one line: that sealed key never leaves the file before its round's reveal, and a value is never corrected quietly mid-round; a fact discovered false cancels its round, with no points, and the game moves on. Play solo in chat, against a sealed tolerance written before the game, 20 percent by default and never adjusted after an answer: land within it and you score the point. Or play in a group, 2 players and more, inside a group chat you own: answers are collected in private, one to one, because a visible number anchors everyone after it, and the reveal publishes the true value, all the answers, every distance and the running scores all at once. The closest scores 1 point, an exact answer scores the same single point and is celebrated in words only, and a tie at the smallest distance is a shared point for each tied player, with a tie in the final standings making co-winners. Going over the true value never eliminates anyone, unless the host chose the without going over variant before the game and it is written in the state. The small print is settled in advance, and it is part of the game. A decimal comma and a decimal point are the same number; an obvious conversion, a value given in kilometres for a question in metres, is accepted and confirmed to the player in private before the deadline, while an ambiguous answer gets one clarification request before the deadline and is invalid without one. A group answer posted in the public thread by mistake still counts, the round continues, the referee says nothing about its value in the thread, and the incident is mentioned honestly at the reveal; the host alone may ask for that round to be cancelled before the reveal. Late answers are refused, silence at the deadline skips the round for that player, and a dispute over a calculation is settled from the state file, by republishing the answers, the value and the distances and redoing the calculation aloud. Honest prerequisites: installing the skill files needs nothing at all, and playing solo in chat needs nothing else. Playing in a group needs a Muse assistant and a messaging account you connect to it yourself, Messenger first, plus a group on it that you designate; the referee writes from your own account, and every public message needs your approval before it is sent. If the host also plays, they answer privately like everyone else and are shown statuses only, never the other answers before the reveal, and that protection is procedural, a matter of trust, stated plainly at the opening. This listing is under the Skill Harbor House License v1.0: free to use, all rights reserved, not open source.