From a sketch to a playtest-ready game in five days
How Tide Pools went from an idea on a Thursday morning to a balanced card game, ready for playtesters, by Monday evening, without a line of code.
16 November 2026
On a Thursday morning I had no game. By the next Monday evening I had a card game with a base version and an advanced version. Both had been played tens of thousands of times by bots, every player count was close to even, and it was ready to put in front of people. I never wrote a line of code, and every card fits in one printed box.
This is how that week went, step by step.
The game
Tide Pools is a card game for 2 to 4 players about marine researchers.
- Photograph creatures. You don't collect the crabs, starfish and eels in the tide pools: you photograph them, and the photos go into your logbook. The ecosystem stays safe, and the research gets done.
- Take on research grants. Each grant pays for certain photos in your logbook: pairs of urchins, the most eels, at least four crabs, one of every kind.
- One thing per turn. On your turn you either take photos from the pool of four creatures, or apply for a grant.
- The tide ends it. When the Low Tide card turns up, the round finishes, and the best-funded researcher wins.
The Advanced game, Into the Deep, adds the kelp forest and the reef further out. You reach them with dive permits from the marine station, and the creatures there are worth more.
Day 1: from an idea to a game I could play
Start with a sketch
I wanted something small and new, "something similar to Point Salad", where cards score other cards. By lunchtime the sketch was a single page:
- 60 creature cards, 12 each of 5 kinds.
- About 18 scoring cards.
- Take two creatures, or take one scoring card.
The theme came over the day. Photographing instead of collecting. A logbook. Grants instead of "field notes". A portfolio for the grants you hold.
Put it in, no code
That afternoon I entered the game on Playtestable's guided screens. The screens ask the questions a rulebook answers, in the order you'd write one.
What's in the box. I made one component, a Creature card, with a value called kind: Crab, Starfish, Anemone, Urchin and Eel, 12 of each. That single form made all 60 cards. Later I changed counts and kinds in one place, and every card followed.
The grants, pasted in. My grants were a table in my design document, so I pasted them. Each grant's columns read like the card itself: scores each, points 3, per 2, counts Starfish. "i pasted the table and it filled in pretty good." The next morning I pasted the rest straight from the document. Playtestable matched my headings to its columns and didn't duplicate anything.
Where things go on the table. The tide deck, the row of four creature cards, the grant deck, the row of three grants, and each player's logbook and portfolio. Each place says how its cards sit: stacked in a deck, spread out in a row, or in no order, like a supply of cubes.
Numbers I wanted to test are named, not typed. The pool holds Pool size, the grant row holds Grant row size, a portfolio holds Portfolio limit, and each player starts with Logbook starting size. "everything goes in a variable for me that can be changed and playtested." Any rules option can change any of them later without touching the screens.
Setup reads like a rulebook. Shuffle the Tide deck. Deal from the Tide deck to the Tide pools until it has Pool size. Deal Logbook starting size to each player's logbook. Eight steps in all.
The turn is one choice. The current player chooses what to do: take photographs, or apply for a grant. Each option holds its own steps inside it. Taking photographs is one card that reads takes 1 to 2 cards, and any they don't take come from the top of the Tide deck. Applying for a grant is only offered while the portfolio has room.
Scoring reads like the cards. Grants score each, most, fewest, at least or sets, with the points and creatures from my table. A "most" grant pays only for strictly the most, never for a tie. The total is a value I named Funding. Ties go to the player with fewer grants.
The rulebook writes itself. While I entered the game, the rulebook beside the form wrote it out as a player would read it. Reading it shows a mistake, such as a wrong card count, before you ever play. A badge at the top said whether the rules could run.
Play it
Around midnight I played my first game by hand, with a test set of seven grants. On the play screen, clicking a creature means "take photographs" and clicking a grant means "apply". When only one move is possible, it's made for you. The move log names every card drawn from the deck, and at the end a scoring sheet shows every line: Starfish pairs: 5 for each 2 Starfish: 5 Starfish, 2 sets of 2 → 10.
That first game told me something about my design straight away: "somehow the grants pile ran out. oh! i need to add all the grants!" Seven grants weren't enough for a game.
Day 2: let the bots tell you what's wrong
The first report
The next morning I had bots play the four-player game. The report put the problem in the first line: the first player won 43% of games and the fourth won 10%, against a fair 25% each.
Comparing the seats showed why. My game ended the moment the creature deck ran out, which could be in the middle of a round, so the first player often got one more turn than the last.
The Low Tide card
Making the game finish the round gave everyone equal turns. But I didn't like ending a tabletop game on what's left in a pile: "it makes no sense to end the game based on what remains in a stack on a tabletop game… I guess you could put a 'marker card' on top of the final X cards, right?"
So the game got a Low Tide card:
- Setup: after the shuffle, Low Tide goes into the deck with twice as many cards under it as there are players.
- During play: when it is drawn, it is set aside on the Shore, and the next card is drawn in its place.
- The end: the round finishes, so everyone has had the same number of turns.
Random bots checked it on thousands of games: every seat won 24% to 26% at four players, and no game ever got stuck with nobody able to move.
Does skill matter?
I asked whether my game was tactical or strategic. A stronger bot studied the decisions in two-player games:
- Choosing which grant to apply for decided about 20 points of win chance.
- Choosing between photos and a grant decided about 8.
- Choosing which creatures to photograph decided about 1.
Strong bots beat weak bots in every match. So skill matters, and the grants are the heart of the game.
Which strategies win?
The paths to victory report sorts players by the kind of grant they go for, and shows how often each kind wins. In 400 four-player games:
- Volume (points for each creature): half the players went for it, and they won 37% of the time.
- Majorities ("most eels"): won 9%.
- Thresholds ("at least four crabs"): won 6%.
Volume grants were too good, and majorities were traps.
The same report found that players holding 3 grants beat players holding 4, 32% to 24%. I decided to keep it: "it's a meta players can pick up on their own." Only take the fourth grant if it pays more than more photos would. The report now says a result like this can be intended.
Grants priced by luck
How much should a grant pay? It depends on how often the shuffle hands it to you: "IF a player 'lucks into' a good draw, then some grants, which otherwise would be 'too hard' to fulfill, now become worth a lot of points!" The What each grant pays report measures exactly that, for every grant, over thousands of games.
To try different sets of grants, I had Playtestable make every grant its scoring rules allow, all set aside. Then I ticked the Active box on the ones I wanted in play. No copies of the game, and no variants to keep track of.
Day 3: balance it, side by side
Ideas, one click from a test
The report does more than describe. It suggests changes, such as Strongest grants pay less or Even out the grants, and Make it a rules option turns a suggestion into a version of the rules I can test. A comparison then plays both versions on the same shuffled games, so any difference comes from the rules, not from luck.
| Path's win share | My grants | Strongest grants pay less | Even out the grants |
|---|---|---|---|
| Volume | 36% | 33% | 29% |
| Sets | 19% | 27% | 26% |
| Thresholds | 12% | 14% | 25% |
| Majorities | 7% | 12% | 13% |
Two changes brought three strategies close to a fair share: three rounds of Even out the grants, and setting aside six grants that players scored without ever trying for them.
A win on zero points
One comparison listed a winner with 0 points. Every game replays exactly from its seed, so I could look at that one game out of a thousand. Four weak bots had photographed creatures all game, nobody took a single grant, and all four tied on 0. "omg that is hilarious!"
It gave me an idea for later: "a game where some photographs are worth more than others and there are fewer of those types of creatures."
Two players
Two players need slightly different grants. Rules options stack, and an option can say which player counts it is used with. So Two players turns itself on at two players. Today it simply sets six grants aside and leaves every printed card as it is.
After one more round with stronger bots the next day, the base game is even: two players win 50% each, and four players win 25%, 23%, 25% and 27%.
Days 4 and 5: going deeper
For the Advanced game, every idea became its own version of the rules, tested against the current one. Within the first day I switched to judging everything with Hard bots, which play out the rest of the game 400 times before every choice. Easy bots had been fine for finding the obvious problems, but they weren't playing the way people would.
One coastline, three zones
My notes had rare creatures and something you earn to reach harder cards. I wanted five creatures in each zone, plus a few uniques, and zones that make sense on a real coast. The layout became one coastline, from the shore outward:
| Zone | How you get there | Creatures | Unique |
|---|---|---|---|
| Tide pools | walk | Crab, Starfish, Anemone, Urchin, Eel | none |
| Kelp forest | swim, 1 permit | Sea otter, Rockfish, Abalone, Kelp crab, Sea hare | Giant octopus |
| Reef | dive, 2 permits | Moray, Garibaldi, Nudibranch, Spiny lobster, Wolf eel | Giant sea bass |
- Each zone has its own pool and grants. It plays like a small Tide Pools of its own.
- Reports earn permits. You can report on a grant early for half its points. If the marine station sponsors that study, it gives you dive permits.
- Photos are worth points. Every photo scores its creature's value.
It's a rules option of the same game, so it stacks with Two players, and the bots can compare it with the base game.
The biggest find: permits are spent
At two players, whoever reached the reef first almost always won. The obvious fix was a catch-up rule, such as a free permit for the second diver. I didn't want one. A catch-up rule with no reason inside the game only hides the problem, so I asked what was really keeping the second diver from a fair chance.
Then the real cause came out. The rules as written let you keep your permits forever, and I had never meant that. A dive should spend its permits, and they go back to the station to be issued again. A sponsored report gives two.
| Two players, Hard bots | Permits kept forever | Permits spent and returned |
|---|---|---|
| First to the reef won | 92% | 53% |
| Reef dives per player | 6.0 | 1.4 |
Same game, same cards, one rule read the way I had always meant it. The bots showed in one afternoon that it mattered.
One printed set of cards
Balancing Advanced was going to mean new points on some tide pool grants, which means a second set of cards. On a screen that's a switch; in a box on a store shelf it's a second print run. So my rule became one set of cards. Advanced can only use what does nothing in the base game:
- The value printed on each creature. It scores nothing in the base game.
- A permit icon on some grants. Only Advanced has permits.
- Cards that only Advanced uses.
That turned out to be enough. The tide pool grants are balanced for Advanced by which ones carry the permit icon, not by their points. When kelp forest divers lagged behind reef divers at two players, the fix was to sponsor every kelp forest grant. And Two players no longer changes any printed number: it only sets cards aside.
The last seat
In Advanced the last player won too often: 32% at four players. A final dive is worth a lot of points at once, and only the players after the Low Tide card know their turn is their last. I tried four turn-order rules:
| Idea | Four players, by seat |
|---|---|
| One more round for everyone | 25 / 23 / 24 / 28 |
| The final round in reverse | 29 / 27 / 22 / 23 |
| Everyone picks their final action in secret | 32 / 24 / 22 / 23 |
| The first player passes left each round | 30 / 20 / 27 / 23 |
Each one just moved the edge to another seat, and each added a rule to learn. The game has to stay simple, and if a small bias has to remain, so be it.
Then I tried the cards instead of the turns: every tide pool creature worth 1, the kelp forest 2 to 3, the reef 4 to 5. Each kelp forest kind is now printed with some cards worth 2 and some worth 3, and each reef kind half worth 4 and half worth 5. Four players went to 24 / 25 / 25 / 26 in 1,200 games. The last seat still gains about as much at the end, but those points decide fewer games. I hadn't expected a change to the cards to fix a problem with the turns.
Two players: a head start of two photos
At two players the second player won 56%, because when they turn up Low Tide the game ends with no reply. What should the first player get to make up for it? I asked whether a creature would do better than points: "that would represent the points of the end, right?" I tested five head starts:
| Player 1 starts with | Player 1 wins |
|---|---|
| Nothing | 44% |
| 2 points | 48% |
| 3 points | 43% |
| 1 chosen tide pool creature | 46% |
| 1 chosen kelp forest creature | 48% |
| 2 chosen tide pool creatures | 50% |
Creatures beat points because they also count toward grants. 1,200 more games confirmed it: 48.9%, and the typical gap between the two players' scores was 0. So at two players, the first player chooses two tide pool creatures before the first turn. The base game didn't need any of this: it was already 50/50.
Three players: a bias I keep
At three players the last seat wins 38% against a fair 33%. Every head start I tried swung it too far the other way, because three-player games are so close that a single creature decides many of them. A rule worth a point or two isn't worth an extra step at the table, so three players plays as printed.
Near to far
Last, I wanted the screen to read like the coastline: "from close to me to far, or in the screen orientation, from the bottom to the top." Each place can say where it's drawn. Advanced shows the reef at the top, then the kelp forest, then the tide pools nearest to you.
Ready for the table
| Mode | Wins by seat (Hard bots) | Fair share |
|---|---|---|
| Base, 2 players | 50 / 50 | 50 |
| Base, 4 players | 25 / 23 / 25 / 27 | 25 |
| Advanced, 2 players | 49 / 51 | 50 |
| Advanced, 3 players | 31 / 31 / 38 | 33, the small last-seat edge I kept |
| Advanced, 4 players | 27 / 23 / 24 / 26 | 25 |
Other checks came out even too:
- Grants: no grant's holders win much more than any other's.
- The reef race: at two players, the first to the reef wins 52%.
- The kelp forest: a player who never dives the reef still wins 53%.
What's left is what only people can tell me:
- Do players understand permits, dives and reports after one explanation?
- Does the kelp forest feel worth diving?
- Does the two-player opening pick feel natural?
- Does the game end at a good moment?
A designer's work is never done
There's plenty I could still try, and each one is a single changed number, because every number I might want to test has a name:
- A pool of 5 creatures instead of 4. More choice every turn, and maybe less luck.
- A grant row of 4 instead of 3.
- A portfolio limit of 3 instead of 4. Fewer, bigger decisions about grants.
- Two starting photos instead of one.
- More or fewer cards under the Low Tide card, for a longer or shorter game.
Each is a rules option I can play against the current rules on the same shuffled games, at every player count, while I get on with something else. The game is close, but there's always another question to ask, and now each one is cheap to answer.
What I'd tell another designer
- Start thin. All the components, a handful of scoring cards, one game by hand. You'll have a running game the same day, and the bots can start working on it.
- Name every number you might change. Pool size, hand size, starting cards. Then a variant is one changed value, not an afternoon of edits.
- Let the bots find the obvious problems. The 43% first seat, the dominant strategy, the grant that pays every game: you'd need dozens of playtests to see them, and the report showed each one in minutes.
- Compare on the same shuffles. Only then do you know that a difference came from the rules and not from luck.
- Judge balance with strong bots. Weak bots don't play your game the way people will.
- Look for the in-game reason. Spent permits, a Low Tide card and creature values fixed things that catch-up rules would only have hidden.
- Keep one box. If a fix needs a second set of cards, find a different fix.
- Keep the biases players can learn. Three grants beating four, the reef race at two players, a small last-seat edge at three: those are things to discover, not bugs.
- Then hand it to people. Bots tell you whether it's fair. People tell you whether it's fun.
Tide Pools by the numbers
| From the idea to ready for playtesting | 5 days |
| A first playable version, entered on the guided screens | about 9 hours |
| Cards | 60 tide pool creatures, 30 grants and a Low Tide card; Advanced adds 72 creatures and 18 grants |
| Versions of the rules tested for Advanced | about 30 |
| Games played by thinking bots | about 115,000 |
| Lines of code written | none |