News, dev logs, and game recommendations.
We let two identical computer players play 560 games of Mancala and forced the first move each time. Opening with pit 3 and then pit 6 won 77 games out of 80. Opening with pit 5 lost all 80. Moving first is a big advantage, but only if you spend it on the right pit. Here are the numbers.
Stuck on a sudoku and tempted to guess? We dealt 12,000 of our puzzles and solved every one with logic alone. Easy, Medium and Hard never needed so much as a what-if. Expert always needs more than the basics, and a small slice needs a structured what-if. Here is which technique each level really demands.
Ludo feels like pure dice, but it is not. We pitted simple playing habits against each other across 420,000 games. Checking whether you can be hit before you move nearly doubled a player's win rate; bringing a token out on every six cut it almost in half. Here is how much skill matters, which habit is worth most, and where luck still wins.
How much harder is Spider Solitaire with two suits, or four? We had one bot play the same 400 deals at every suit count, with the cards in identical places and only the suits changed. It won 85.5% on one suit, 39.5% on two and 2.3% on four. Here is what those numbers mean, and what they don't, for your own games.
We dealt 3,000 FreeCell games and set a solver on them. It won all 3,000, and every win was replayed move by move in the game itself. So if you lose, the deal almost certainly did not beat you. The interesting number is what happens when free cells run short: with two instead of four, about one deal in four stopped solving.
No, but nearly. A solver that sees every card won at least 84.6% of 500 Klondike deals on Draw 1 and 78.2% on Draw 3. A simple player who cannot see the face-down cards won 20.4% and 3.6% of the same deals. So most lost games were winnable, and Draw 3 costs a careless player far more than it costs a good one. Here are the measured odds.
Everyone knows the 2048 corner strategy. We tested it: across 900 games, three different versions of "keep the big tile in the corner" never reached 2048 once. A player that looked two moves ahead got there in 141 of 200 games. The corner is necessary, but it is not the strategy. Here is what the rest of it is.
We played 4,000 Yatzy score sheets with a strategy that weighs every hold, and wrote down everything. The median score was 216, only one sheet in ten passed 279, and the upper-section bonus turned out to be worth more than rolling a Yatzy. Here is what a good score looks like, and where the points actually come from.
Minesweeper is mostly logic, but not entirely. We ran a solver that only ever makes moves it can prove are safe across 9,000 boards. It cleared 90% of easy boards without a single guess, and only about 55% of the medium and hard ones. Here is where the luck lives, how big it is, and why guessing well still wins most games.
Pyramid Solitaire has a reputation for being mostly luck, and part of that is true: a solver that tries every line proved 55 of 200 random deals impossible. But the number of passes through the deck matters far more than people think — it moves the winnable share from 4% to over 60%. Here are the measured odds.
Our Mahjong Solitaire deals are built backwards, so every one of them can be cleared. We then played 1,000 of those guaranteed-winnable boards with a bot that picks legal pairs at random, and it got stuck on roughly six in ten. So "is it winnable?" is the wrong question. Here is the right one, and what the numbers say.
Melon Drop is our first game with resting contacts — fruit that piles up and stays piled. Getting it to settle took six changes, and the bug underneath them was an ordering mistake: the position-correction pass ran before positions were integrated, so it separated bodies that were touching, the next frame found no contact, and gravity was never cancelled.
Anyone can put our games on their own site. The route worked, the snippet worked, framing worked. And one grep for who links to it returned zero hits outside the embed pages themselves, so every embed had to be preceded by an email from us. A channel only an email can reach is not a channel.
A player told us our games UI was poor. Rather than start a redesign, we counted: 28 of 81 bundles draw with zero gradients, 16 are under 700 lines, and 38 never load the shared art layer — and they are almost exactly the same 38. The shared layer already existed. The games that most needed it are the ones that skipped it.
Bubble Shooter shipped two instances of the same bug class in one game. A row's parity was stored and also recomputed from its index. The aim guide traced the shot and the shot re-integrated its own flight. Neither computation was wrong on its own, and neither failure ever threw — because a wrong cell is still a legal cell.
Fairway proves each hole clearable at or under par before it deals it. It shipped four separate verification seams, and every one of them replayed the shot through a helper while the player's ball ran through a different function. Nothing asserted the two agreed — so deleting the stroke counter from the real game left all nine tests green.
We shipped three games in one batch and sabotaged each of their suites by deleting the rule the game depends on. All three stayed green, by three completely different mechanisms: the assertion tested a helper instead of the game, a second cause produced the same outcome, and one test inherited an environment where the bug is invisible.
Inkfield's fairness assertions all passed while an attentive player making zero mistakes died 23 rounds out of 24. The death rate alone said only "this is hard". Splitting it by cause — edge 0, self 0, rival 23 — named the problem in one line, and the fix was arithmetic rather than tuning.
Kingsroad's reference build never constructs a Barracks and never collects a gold crate. That is deliberate: it turns the guarantee from "beatable if you play well" into "beatable even if you ignore both of these", and it converts two mechanics from hopefully-balanced into provably-upside. The exclusion is not a comment. It is two assertions.
Kingsroad proves every map beatable before showing it to you, with a three-rung ladder that repairs a map that fails. Then we read the diagnostics: repairs 6, trims 12, health scaled to 0.43. Every rung firing on the same map means none of them is aimed at the actual problem — and the bottom rung was hiding a generator bug by making the game 57% easier.
Star Grid's generator threw away 99.5% of the boards it made, hunting for one with a unique solution. Instead of generating more boards, we started repairing the ones we had: enumerate the rival solution, then make a single edit that provably kills the rival and provably cannot kill the answer. Yield went from 0.5% to 100%, at 2.9 to 29.8 milliseconds a board.
Building difficulty bands for Bubble Shooter, we used a reference solver to estimate par. It reported par 7 for boards a real player needs thirty-odd shots to clear, and a shot budget built on that number would have made the game unwinnable. The solver was not buggy. It was playing a different game — one where you get to choose which colour comes next.
We wanted Tile Order to show a proven fewest-moves number, not a guess. On a 4x4 that meant proving optimality, and a fully shuffled board cost up to 22 seconds to prove — unusable in a browser. The fix was not a faster solver. It was generating boards the solver could afford to prove, and being honest in the copy about what that costs.
Some puzzle genres are structurally incapable of being hard — not because the designer went easy on you, but because of the rule that defines them. There is a one-question test that sorts them, we ran it across our whole catalogue, and it explains why a game marked "Expert" can still feel like nothing. Here is which of ours are which.
Most people learn board games in whatever order a relative happened to own them, which is why so many bounce off chess. There is a better order — each game teaching one idea the next one assumes you already have. Six games, six ideas, all playable free against the computer with the rules a click away.
Most game lists ignore the only constraint that actually decides what you open: how long you have. A game you can finish in three minutes is frustrating for an hour, and a game worth an hour is the wrong thing to start on a station platform. Four time budgets, and what makes a game fit each one.
Aeroplane wifi is either expensive or fictional. These games are built as a progressive web app, so adding the site to your home screen before you leave means they keep working with no connection at all. Here is how to do it, which puzzles suit a long flight, and the one thing that does not work offline.
"I'm bored" covers at least four different feelings, and the game that fixes one of them makes another worse. Restless bored wants something fast. Fidgety bored wants something with no lose state. Stuck-waiting bored wants something you can abandon mid-move. Here is which of ours to open for each.
A game shrunk into an aspect-ratio box is not a smaller game — it is a wrong one. We measured every bundle we ship at phone size and found stretched boards, controls pushed somewhere with nothing to scroll to, and a median board area of 21% of the screen. Three of our five predictions from reading the CSS were wrong.
Our first difficulty rater asked "did this technique fire while solving?" One technique fired on a median of one board in one, and almost no board actually needed it. Rating by necessity instead — is the board still solvable with the technique switched off? — produced a ladder we could trust, and two findings we could not have guessed.
"Every row is passable" sounds like a simple promise. It took five wrong versions of it to keep, and the interesting part is that the road generator was never the thing at fault — the proof was. Including one bug that did not break fairness at all, and so survived scrutiny by making the game impossible to lose.
Most puzzle games ship a bank of pre-computed levels because their solver is too slow to run while you play. We found a heuristic that made ours run per move, which deleted the bank, the build step and the shipped asset — and then let us put something on screen no other ball sort game has: the shortest finish still available from where you are.
We abandoned Kakuro once. Fill-then-test found a uniquely solvable board 0% of the time; the fix we wrote down was "find a construction that forces uniqueness". That framing was the trap. Aiming at a cheaper property — a board a guessing-free solver can finish — got us to 100% yield in milliseconds, and it implies uniqueness anyway.
Before building Arrow Escape we tried to work out how hard the genre could get. The answer was: not very, and not for want of trying. Five levers, all neutralised, and a ceiling that comes straight from the rule every one of these games shares. Here is the measurement, including the part we got wrong.
