How hard it is

What makes a journey hard, what we measured and what moves it, on a phone and on a computer. Every number here comes from a tool you can run yourself (the last chapter says how). Measured on 29 September 2026; the levels measured again on 7 October 2026, after the new rules for seals, stones and covers (version 0.3.0).

A journey should be won by most people on Relaxed and by few on Pharaoh, and it should feel about the same on a phone and on a computer. Nobody can play a thousand journeys to check, so we let two pretend players do it and count.

Contents

1. The short version
2. How we measure
3. What makes a stack hard
4. The levels now
5. What moves it and by how much
6. A phone and a computer
7. What we learned along the way
8. Measuring it yourself

1. The short version

• Since the covers changed (7 October 2026), careless players win clearly less, and much less on a computer. Careful players are still on aim almost everywhere. On Normal a careless player wins 52 journeys in a hundred on a phone (aim 70) and 35 on a computer; on 29 September it was 64 on both (chapter 4).
• A phone and a computer have drifted apart again. A cover now lifts only once both its sides are gone, and a computer’s wide stacks have long rows: there covers hold much longer (chapter 6). Bringing the two together again is the next thing to decide.
• Shuffles, lives and the Gods move it most, the same on every screen. Most other settings act differently on a phone and a computer, or not at all (chapter 5).
• A phone’s stacks are easier by nature. The same stack drawn tall for a phone is cleared much more often than drawn wide. So a phone is dealt a little stiffer and a computer a little gentler, to meet in the middle.
• Players end rich. They have 70 to 100 deben before the last great stack, where we once aimed at 30 to 60. Since the deben left over became the journey’s score, that may no longer matter.

2. How we measure

The rules run without a browser, so a bot can play them. The bot takes the pair that frees the most amulets. It doesn’t plan ahead, which makes it play much like a person. For measuring, it plays as two people:

Each plays whole journeys: the map, markets, shrines, the river, every stack, the lives. It uses a shuffle when stuck, then a boon that helps. The last two amulets are dealt again for it and it asks the Gods like anyone would. It never uses undo or a hint.

Each level has an aim in content/settings.jsonc ("wins"): how many journeys in a hundred the careful and the careless player should win. We call a level on aim when it is within 7 of it. At 200 journeys, a result wobbles by about 5 either way from one run to the next, so a difference smaller than that is chance.

A real person is neither of our two. They plan a little, they use undo and hints and they learn. So the tools tell us which way a change moves the game and roughly how far. Playing it tells us how it feels.

3. What makes a stack hard

Many settings touch how hard a stack plays. They come in this order, from the map down to the last tap:

1. Which stack. The first two rows deal only a few small, forgiving stacks ("start"). After that the stacks grow along the journey and each season keeps to so many layers: Akhet 2 to 4, Peret 4 to 5, Shemu 5 to 6 ("seasonLayers").
2. How many kinds and which special amulets: the row’s "depth" (more kinds and more special amulets as the map climbs), the level’s kinds on top and an omen’s.
3. The screen’s shape: a phone’s stacks get two kinds more and more buried twins, a computer’s four kinds fewer (never below six) and fewer buried twins ("shapes", chapter 6).
4. Which deal. Eight deals are tried and the one whose careless plays get stuck nearest the row’s trap is kept ("choose"). Every deal can be cleared.
5. The help in play: the level’s shuffles, lives, undo and warning; the last two amulets dealt again ("mercy"); the Gods, once a journey ("askGods"); and the boons and relics that rescue a stuck stack.

4. The levels now

200 journeys at each level, careful / careless, against the aim.

Measured on 7 October 2026 (version 0.3.0), with 29 September beside it.

On a phone (the levels are tuned here):

On a computer:

In bold: off aim (more than 7 away). Careful players are where they were, on both screens. Careless players lose more: on a phone Normal and Hard are below aim (by 18 and 8), on a computer every level is, by 15 to 35. They also need the game’s help far more: on a phone the last few are dealt again 100 to 170 times in a hundred journeys (70 to 110 before).

What did it, measured one change at a time on 6 and 7 October with both screens’ shapes mixed (careful / careless on Normal): seals on tiles you can see, 94 / 59 to 91 / 57; stones on the middle layers, no change; a cover lifting only once both its sides are gone, 91 / 55 to 90 / 34; a quarter fewer covers, back to 95 / 46. The new cover rule is by far the strongest; the mixed measurement hid how differently it falls on the two screens. Every step is in the decisions.

Where it gets hard. The first rows are cleared almost every time without getting stuck (row 1: 100 in a hundred on Normal). From the second season on, only a third to two thirds of the stacks are cleared without once getting stuck, careful or not. The journey’s difficulty is in the second and third seasons, as intended.

5. What moves it and by how much

Measured on 29 September 2026, before the new rules for seals, stones and covers. The direction of each setting will still hold; the sizes may have changed.

One setting changed at a time, everything else as it is, 200 journeys on Normal. The change in wins, careful / careless (Normal is 94 / 64 as it is, on both):

A difference of 5 or less can be chance.

What this tells us:

• Shuffles, lives and the Gods move it most, by 12 to 21 careless wins each, on both screens.
• The same setting can do different things on a phone and a computer. A harsher trap makes a computer much harder but a phone hardly at all. Two kinds more matter only on a computer, two fewer only on a phone: most of a phone’s stacks already have as many kinds as a set holds (12), while most of a computer’s are already at the fewest allowed (6).
• Dealing the last two again matters more on a phone, where stacks more often get stuck with only a few amulets left.
• Undo shows as nothing only because our pretend players never use it. A real person does.
• More buried twins do nothing measurable on either.

So to move a level a lot, change its shuffles or lives: they work the same on every screen. Most of the rest act differently on a phone and a computer, or not at all, which is why the next step is to simplify them.

6. A phone and a computer

The finding. The same stack is much easier drawn tall for a phone than drawn wide for a computer. Dealt the same way and played by the same bot with no shuffles:

The worst are the wide, flat stacks: the mastabas are cleared 23 times in a hundred wide and 90 times tall, the three pyramids 24 against 81. A phone needs short rows to keep the amulets large and short rows split a stack into pieces that come apart on their own. A wide stack is hard because it is one piece across its whole width.

What we tried. Redrawing the tall shapes so no more amulets are free at the start than on the wide ones; rubble at the ends of the rows; towers; cutting a wide stack in two. None of them closed the gap: the mastabas stayed at about 80 in a hundred however they were raised.

What we do instead. The levels are tuned on a phone, where we play most. A phone’s tall shapes are dealt stiffer: two kinds of amulet more and buried twins more often. Tuned like that, a computer came out far harder (careful / careless on Normal: 78 / 36). So the wide shapes are dealt gentler: four kinds fewer, but never fewer than six, with fewer buried twins. The two now meet in the middle:

Apart again since the covers changed (7 October 2026). With covers that lift only once both their sides are gone, careless players win 52 on Normal on a phone and 35 on a computer (careful: 93 and 96). The likely reason is the same as above: a wide stack is one piece across its whole width with long rows, and in a long row a cover waits until the tiles on both sides of it are gone, while a phone’s short rows free it sooner. Bringing them together again (fewer covers on a computer’s shapes, as the kinds already differ by screen, or a gentler computer in general) is for the owner to decide.

What still differs is how a journey is lost. On a phone the bot is often stuck with only a few amulets left, so the last two are dealt again about once a journey (70 to 110 times in a hundred journeys). On a computer that happens about a quarter as often (18 to 27) and the Gods are asked a little more instead. The wins come out the same.

7. What we learned along the way

• 27 September: Normal felt hard. Hex found Normal hard and Relaxed “just right”, so every level moved down one and Pharaoh was added at the top. A careless player winning about half on Normal was fine.
• 27 September: what matters most. Tuning the lives, the strongest settings were the shuffles at the start, the lives, the kinds of amulet and the boons and relics that rescue a stuck stack. A row’s trap hardly mattered while there were rescues.
• 29 September: a journey had become too easy. Anubis played a whole journey on Normal in half an hour and ended with shuffles and boons to spare. Only the last great stack needed thought. The bot won all its journeys on Normal too. Four kinds of help had been added one after another (the last few dealt again, the Gods, undo and the one-pair warning). Together they undid what the levels asked. With less help alone, careless wins on Normal went from 98 to 90 and on Hard from 82 to 66.
• 29 September: no tombs. Tombs could only appear on a season’s fourth row, which a stack usually took: on 56 maps in a hundred no tomb could be reached at all. Mended, with eight rows a season and at least one tomb on every way up from the second season.
• 29 September: the phone’s shapes (chapter 6).
• 29 September: the vultures. A new player found far too many of one kind in a stack. Buried twins copied a kind again and again until it crowded out the rest. Now no kind gets more than its fair share and a pair.
• 6 and 7 October: seals, stones and covers. After a play, seals went onto tiles you can see (before, they sat on the bottom layer and had opened long before anyone reached them), stones onto the middle layers, and a cover now lifts only once both its sides are gone. Each change was its own commit and measured on its own (chapter 4).
• 7 October: covers that wait for each other. Two covers in the same row could never lift: the tiles between them waited for each other. Most deals with covers were impossible, so the deal tried hundreds of times (a stack took 0.5 to 6.7 seconds to deal instead of 4 ms) and then quietly placed fewer covers. That made the first measurement look easier than the game really was. Now a row gets at most one cover.
• 7 October: one number hides two screens. The playtests of the changes mixed a phone’s and a computer’s shapes. Measured apart, the computer turned out far harder than the phone. Measure both, always.
• 29 September: a player like us. A third pretend player learns from our play logs (engine/tools/fit-player.js). On the stacks in our two logs, weighing only the pair in front of it cleared about half, but we cleared 10 of 12 with no help: we see traps coming. Looking ahead 8 times in 10, it clears them as often as we did. Fitted to those logs, which were played before the longer map, it won every journey on Normal, each taking it about 71 minutes. It will be fitted again to a journey played on the game as it is now.

Every decision is in the decisions, with who made it.

8. Measuring it yourself

The tools need Node.js. Build first (python3 build.py).

SHAPES=phone node engine/tools/difficulty.js 200
SHAPES=wide node engine/tools/difficulty.js 200

The level table above, with each row of the map, how stacks are lost and how often the help was needed. Each takes about twenty minutes to half an hour and keeps every core of the computer busy.

node tools/levers.js 200
SHAPES=wide node tools/levers.js 200

Chapter 5: one setting changed at a time, on Normal. Give a level after the number to try another (node tools/levers.js 200 fierce). It works on a copy of the content, so your own files are never touched.

node engine/tools/shapes.js

Chapter 6: every stack’s wide shape against its tall one.

After changing a level, run the first two again and update this page.