Cheery Chimp

Inventing a solitaire is easy. Finding out whether the one you invented is any good is the hard part, and no computer could tell me.

Most of the card games on this site are adaptations — Klondike, FreeCell, Spider, games with a century of received wisdom behind them and a clear right answer about how they work. Convergence is not one of those. It started from an idea and had to be argued into a game.

The idea was small: what if some foundations counted up and others counted down?

Both ends at once

In every standard solitaire the foundations climb. Ace, two, three, up to the king, one pile per suit, and every card has exactly one place it eventually belongs.

Convergence gives each suit two foundations instead of one. An up pile that starts from the Ace and climbs, and a down pile that starts from the King and descends. Eight foundations in all. A suit is finished when its two piles meet in the middle — when the highest card on the up pile is one rank below the lowest on the down pile, and between them they account for all thirteen. Win when all four suits have converged.

The consequence, and the whole reason the game exists: most cards have two homes. A seven of hearts might be the next card the up pile wants, or the next the down pile wants, or both, or neither — and which you choose changes what the rest of the suit needs from you afterwards. Feed a card to the wrong end and you have not broken anything, you have merely made a commitment you will regret in about six moves.

That fork is the game. Everything else is a delivery mechanism for it: fifty-two cards dealt face up into eight columns, four free cells, no stock, no hidden cards. Nothing is concealed, so nothing that goes wrong is bad luck.

The game had an opinion I did not give it

Here is the thing I did not design and could not have predicted: the two directions are not equally hard, and the reason is structural.

Tableau columns build downward. So do the down foundations. That means a descending run you have carefully assembled in the tableau is already in exactly the order the down pile wants to eat it — the two halves of the game are pulling the same way, and down foundations fill almost by themselves.

Up foundations get no such help. They want the Ace first, then the two, then the three, and low cards have a habit of being buried under everything else you have built. To feed the up pile you have to dig, and every dig costs you a free cell or an empty column. The ascending side is the bottleneck, permanently, in every deal.

I like this more than anything I deliberately put in. It means the two foundations per suit are not a symmetrical gimmick — they are two different jobs that happen to share a suit, and the tension between an easy half and a hard half is what gives the game its shape.

It also fed straight back into the controls. Tapping a card sends it to a foundation, and often it could legally go either way. On a tie it goes to the up pile — because up is where the difficulty lives, and the automatic choice should be the one that relieves the pressure rather than the one that adds to it.

What the solvers could not tell me

Before building any interface I wanted to know whether the thing was worth building an interface for. The failure I was most afraid of is a solitaire that plays itself: a game where the legal moves are so constrained that following the obvious one wins most deals, and the player is just a hand doing what the cards already decided.

So I set bots on it. The results were the most useful ambiguous answer I have had from a test.

A naive auto-player — take any legal foundation move, always — wins essentially none of the time. A greedier one that digs for buried cards wins something like two to ten deals in a hundred. Meanwhile a proper solver, given time to search, failed to prove a single deal unsolvable. Not one.

Read those together and they say: almost every deal can be won, and almost no deal can be won without thinking. Which is precisely the profile I wanted, and is also FreeCell's signature — nearly every FreeCell deal is winnable and nearly nobody wins them all.

What the bots could not tell me is how it feels. Automatic players are notoriously poor at the kind of forward planning FreeCell-family games reward; they are bad in a way that does not resemble how humans are bad. A bot's 4% win rate is not evidence that a person will find it hard, only that this bot did. There is no substitute at that point for sitting down and playing it a few dozen times.

Two rules that came from playing, not from thinking

Both of the changes that made it good were made after the game already worked.

The tableau originally built down in alternating colors, the way Klondike and FreeCell do. It now builds down in suit. Same-suit building is stricter, which sounds like it should make the game harder, and in isolation it does — but a same-suit descending run is exactly what a down foundation can swallow whole. Tightening the tableau rule made the two halves of the game click together instead of merely coexisting.

And the free cells settled at four. They are adjustable — four, five, six or seven — and that is the difficulty dial, rather than an easy/hard label that would have to be explained. Each extra cell is one more card you can put down while you dig. Four is where it lands as winnable-but-working-for-it; seven is where it becomes a pleasant way to spend ten minutes not being challenged. Both are legitimate things to want.

A win worth watching

When the last suit converges, all fifty-two cards launch out of the eight foundation slots, arc down under gravity, bounce off the bottom of the screen and drift off the sides — the celebration old Windows FreeCell players will recognise instantly, with the difference that these cards do not leave smear trails behind them. The trails were a side effect of how that program drew to the screen in 1995, and I decided I liked the clean version better.

It honours the reduced-motion setting, and it cancels the moment you start a new game. It also ended up in the shared card library rather than in this one game, so anything else on the site can borrow it.

The hint and auto-play buttons got the same treatment in miniature: auto plays one move at a time on a short delay rather than resolving the board instantly, and it stops dead the moment you touch anything yourself. A feature that finishes your game for you before you have seen it happen is not doing you a favour.

Where it stands

Playable, and the one I have put the most of my own hours into — which is either a good sign or the least trustworthy possible review.

Two things I would still change. The converging foundations are eight slots across the top of the screen, and eight slots is a lot to fit above a tableau on a phone; that layout is not finished. And the auto-play is legal-move-driven rather than safe-move-driven, so it will occasionally send a card to a foundation you would rather have kept in play. Since every card in this game has two possible destinations, "safe" is a harder question here than in a game where it only has one — which feels like the correct problem for this particular deck to leave me with.