The Ring of Five

How to start

1. Open your AI. Start a new chat in ChatGPT, Claude, or Gemini. No paid plan needed, but the chat must be able to open a link — the Steeple-Keeper reads the sealed page.
2. Paste the card. It opens the ringing chamber, prints the twelve remarks and an empty board, and reads out the sealed hash once it has the sealed page.
3. Work the notebook. Cross out freely, enter only what you can prove, and buy a hint when the chain runs cold.
4. Repaint the board and check the seal. Fifteen inked cells finishes it. Then check the revealed page against the seal at aigamecards.com/verify.

Questions

How do you stop the AI from settling the answer at the end?

It has no say in it. Which founder, which casting year and which duty belongs to each of the five bells was written out, put next to the twelve remarks, and hashed with SHA-256 as one block before the card shipped. That commitment is printed in the card's sealed notice. The Steeple-Keeper reads that page to check every line you paint and has no authority to move one. When you finish, you paste the published reveal into aigamecards.com/verify, which rebuilds the hash in your browser and shows whether it still matches the number you were shown on turn one — and the twelve remarks sit inside that same block.

Is it solvable by elimination, or is there a guess buried in it?

No guess. The twelve remarks force one arrangement and only one, and every step follows from remarks you already have plus what is already on the board. It was checked two ways before release: an exhaustive pass over all 1,728,000 possible boards found exactly one that fits all twelve, and a propagation-only pass — no trial assignments, no backtracking — closed all fifteen cells in four sweeps. If you are stuck, you have missed a remark, not run out of information.

Does the chat need to be able to open a link?

Yes. The Steeple-Keeper fetches the sealed page from aigamecards.com at load, because without it there is nothing honest to check your entries against. Pasting it in yourself would hand you the whole answer. ChatGPT opens links out of the box; in Claude, switch web search on first; in Gemini, ask it to open the link before you start. No paid plan either way.

What does a wrong answer cost, and can I lose?

Yes, by your own hand. A line that does not match the sealed page is a smudge on wet paint, and three smudges spoil the board and end the run at zero. The Steeple-Keeper says the line is wrong and nothing else, so a smudge buys almost no information. Crossings-out are different: they are your working, free and unlimited, and the Keeper never checks one against the page or tells you whether it is right. Say unstrike to take one back. The one thing it will flag is a cell or a code that has gone blank on the board — that is arithmetic you can see for yourself, not a reading of the page.

I have never solved a logic grid. Is this a fair first one?

It is built for it. Five bells, three things to learn about each, and the Steeple-Keeper redraws the whole board every turn so you never need paper. One bell falls out of three remarks before you have painted anything, which is the on-ramp. If the chain goes cold, buy a hint: the cheapest names one remark worth another look, and the dearest names every remark in that step and spells out what they prove together, without ever naming a founder, a year or a duty. Each grade contains the one under it, so paying more never tells you less. You can never run out of information. What is at stake is the score — and the three smudges.