A thousand levels, and nobody placed one of them by hand

Swarm Kingdoms has a thousand levels across ten regions. Nobody sat down and placed a single one of them, and that is not a shortcut — it is the only way a campaign that long can be fair.

A hand-placed campaign of a thousand levels has a thousand opportunities for one level to be accidentally impossible, and no way to find out except a player getting stuck. It also drifts: level 400 is balanced against the upgrade curve as it stood when 400 was written, and the curve moved twice after that. Generation from a difficulty model turns a thousand judgement calls into one, and makes the one checkable.

What the model actually produces

A hundred difficulty tiers across ten regions, with fifty boss families drawn from five archetypes, twenty-two upgrades, eleven enemy types and eight army unlocks to spend them on. Each level resolves to an encounter table: a time, a hit-point pool, a damage rate, a count, a kind and a formation, per wave. The first level opens with sixteen melee attackers in a line at six seconds, then twenty at twenty-two seconds, then twenty-four in a wedge. It reads like a spreadsheet because it is one, and that is the feature: a spreadsheet can be swept for a row that cannot be beaten.

Every figure in that paragraph is counted out of the files the shipped build loads, by the pipeline that builds this page, rather than typed by anybody. It throws if a region file is missing. A number about a thousand-level campaign that a person types is a number that will be wrong within a release.

Winnable is a claim about every route, not about the game

The useful property is not "the game is balanced". It is that every combination of routes a player can take through the campaign is winnable under defined conditions — which is a different and much larger statement, because an upgrade path a player might reasonably choose can leave them under-equipped at tier 60 while every other path is fine.

So the project runs its own acceptance pass over the whole campaign and reports exactly that, with the conditions stated. Settling it took a documented rebalance. The reason it could be settled at all is that the campaign is a function of a model: you change the model, you re-run the sweep, and you learn in minutes whether you broke tier 60 for one route.

Why it is in the Lab and not on a store

This is the part a portfolio page usually leaves out. Swarm Kingdoms is a complete, balance-verified game with no production plumbing around it. Windows and Android builds run and the campaign is playable from the first level to the thousandth. The store layer is declared mock in its own data file and labels itself a development store inside the game. The application identifier is still a local development string. The project itself states that physical-device testing, production billing and production ads are unpassed gates.

So it carries a LAB badge, and the badge is the honest one. A finished game and a shippable product are different things, and the distance between them is almost entirely the work nobody demos.

One more thing that sounds like a detail and is not: it ships ten languages, with the game’s own locale identifiers kept exactly as the game names them — including the ones that are a region and not just a language. The demo on this site speaks those strings rather than a translation written for a website, falling back to English the same way the game’s own localisation code does.