Saltfire: one target a day, four elements to start from, a number to beat

01  A daily puzzle instead of a sandbox

Saltfire is an element-merging game, which is a mechanic most people have met: tap two things together, get a third. Ember and Dew make Steam. Salt and Ember make Ash. Salt and Salt make Crystal. What is different here is that it is not a sandbox. There is no infinite space to wander and no session that never ends — every day names exactly one target, and the whole world is trying to reach the same one on the same UTC day.

That single constraint changes the game more than the tree does. In an open-ended crafting game the question is "what can I make?" and the answer is always yes, eventually. Here the question is "how few moves does it take me?" and the answer is a number you can compare with a friend’s. Recent targets have run from Coal at par 4 up to Window, Basalt and Hope at par 8. Everyone starts from Salt, Ember, Dew and Breath and nothing else, so the only thing separating two scores is what the player remembered and how well they read the puzzle.

It is free, it runs in a browser tab, and a day takes about five minutes. The whole tree ships with the page — 339 elements built from 733 recipes, authored rather than generated while you play — so it loads instantly and plays the same on every visit.

02  The one rule that makes it a puzzle rather than a toy

A pair of elements you have not tried today costs you one merge, whether it fuses or not. A pair you have already tried costs nothing, ever again. That is the whole scoring system, and it produces an unusual kind of pressure: guessing is expensive and remembering is free.

So the game writes everything down. Every pair you have tried sits in a register in the order you tried it, with what it produced, and the duds struck through so you can see at a glance what has already failed. Tapping any line re-opens that pair — it costs nothing, because you already paid for it. A misclick or a forgotten experiment can never punish you twice.

The subtle half of the rule is that par counts distinct pairs, not branches. If you make an intermediate once and then reach for it twice, that is one merge, not two, because it is already sitting in your tray. The short way through almost any day reuses something, and noticing which element is doing double duty is most of the skill.

03  Where the par number comes from

Par is the smallest number of merges the day can be solved in, and it is computed rather than chosen. Before a puzzle ships, a breadth-first search walks the entire recipe tree outwards from the four primordials, level by level, and the first level the target appears on is the floor. The build then replays the chain that search found against the game’s own merge rules; if the replay spends one step more or one step less than the search claimed, the build fails and that day never runs at all.

The practical consequence is that par is a real limit and not a target someone eyeballed. You cannot beat it, so a clean run is the ceiling of the game rather than a lucky guess at it — and two or three over par is a perfectly good day for someone playing the tree cold.

04  What carries over between days

Two things persist, and both live in your own browser. The first is the cabinet: every element you have ever made, kept permanently, with the four primordials already lit on your first visit so the shelf never opens empty. The second is a streak that counts consecutive days solved. The cabinet is a museum and deliberately not a head start — every day begins from the same four elements, because a veteran starting with two hundred elements in hand would make two players’ scores incomparable.

What actually accumulates is knowledge of the tree, and the game is generous about letting you acquire it. Each day closes at midnight UTC and moves into an archive of the 31 most recent closed days. A past day’s page spells out that day’s shortest chain step by step and then explains why nothing shorter exists: which links came straight out of the primordials, whether any intermediate was wanted twice, whether the target has only one recipe in the whole tree. Today is never in that list, so the archive is not a way to cheat at the puzzle in front of you — it is the fastest way to learn how the tree is shaped.

05  How to get a good score

Once you know that par counts distinct pairs, most of the technique follows from it.

• Build the shared middle early. Ash, Brine, Dust, Steam, Mist and Stone appear in an enormous number of chains, and one is usually on the short path to the day’s target.
• Read the target as a recipe. Most names tell you their own ingredients if you say them out loud — obsidian is cooled lava, a lantern is a flame you can carry.
• Self-merges are legal and often necessary. Salt and Salt, Ember and Ember, Flame and Flame, Breath and Breath: on one recent par-8 day, four of the eight merges were an element fused with itself.
• Watch for the element doing double duty. If something you have already made is needed twice, the second use is free, and that is usually where the spare step is hiding.
• Check the tier under the target name — it tells you whether the answer is a couple of merges above raw salt or sits five levels up.
• Do not be afraid to spend merges. After twelve steps the register quietly names one element that appears somewhere in a shortest solution — one element, never the route.

06  What it deliberately does not have

There is no account, no sign-up, no app to install, and no text field anywhere. The only inputs are two taps between elements that already exist, which means there is nothing to type, nothing to post that carries words you wrote, and nothing on the site fed by a player.

It is free and ad-supported: one ad slot below the game, and none inside the workbench where you actually play. The only thing that reaches our server is the number of merges a finished run took, counted anonymously into that day’s totals with no identifier attached to it — no name, no device, no history. Until a day has enough runs behind it to be worth quoting, the result screen prints no percentage at all rather than a figure nobody measured.

The puzzle itself does not need the network — the target, the par and the scoring are all in the page, so the bench genuinely works offline. Only the day’s tally waits, and the screen says exactly that instead of showing an error for something that is not broken. Leave a tab open past midnight UTC and a notice tells you the puzzle underneath you is yesterday’s, rather than yanking a run in progress.

07  Questions people ask

Is Saltfire like Infinite Craft or Little Alchemy?

Same mechanic, opposite shape. Those are open-ended sandboxes with no finish line. Saltfire gives you one named target a day and a computed par, and everyone starts from the same four elements, so scores are comparable. Nothing is generated while you play — the tree is a fixed file of 339 elements and 733 recipes that ships with the page.

Do I need an account, or an app?

Neither. It is free, it runs in a browser tab, and there is nothing to install or sign up for. You never type anything: the only inputs are taps between elements.

Can I play yesterday’s puzzle?

Yes. Past days move to an archive of the 31 most recent closed days, and each one shows that day’s full shortest chain with an explanation of why nothing shorter exists. Today’s answer is never shown while the day is still being played.

Is par really the shortest solution, or an estimate?

It is computed. A breadth-first search walks the whole recipe tree outwards from the four primordials to find the true minimum, and the build then replays that chain against the game’s own merge rules — if the replay is off by even one step, the build fails and that day does not run. You cannot score under par; matching it is the best possible round.

Does Saltfire work offline?

The puzzle does. The target, the par and the scoring rules are all local, so the bench plays fine with no connection. Only the day’s tally of other players’ results needs the network, and the screen tells you that rather than showing an error.

How long does one day take?

Around five minutes for most targets. Recent ones have needed anywhere from four to eight merges, and you can keep playing after you solve — the bench stays open, further merges cost nothing, and anything new you find is kept in your cabinet.

08  More from the shelf

• Lumesweep — Never lose to a 50/50.
• Gridmark — One grid. Four tiers. Beat your time.
• Isobath — Drop a pin. Find out how wrong you were.
• Dewkeep — Keep a creature. Grow a little world.

Browse every app in the catalogue →