The strategy classes

AI notice: This text is created with the support of AI systems; it is reviewed editorially and taken responsibility for before publication.

Out of 16 instructions, four decades have produced a handful of basic forms. Almost every warrior you will find on a hill is one of them — or a mixture of two.

This chapter introduces them. And because the same sentences about their matchups have been passed along for thirty years, we check at the end whether they hold. Two out of three do not.

The basic forms

Stone — the bomber

The dwarf from step 1 is one. A stone stands still and throws DAT bombs blindly at regular intervals through the core. It does not search, it scatters. Small, fast, uncomplicated.

Its strength is speed: it is at full effect within a few cycles. Its weakness is that it cannot tell whether it is hitting anything — and that an opponent reproducing faster than the stone can bomb simply keeps running.

Paper — the replicator

A paper copies itself to ever new places in the core and starts further processes there. Instead of attacking, it becomes a population. A bomber does keep destroying copies — but they appear faster than it can hit them.

The best-known construction is called silk: it copies itself before starting the new process, and needs remarkably few instructions to do so.

Papers rarely win quickly. They win by not dying.

Scanner — the searcher

A scanner searches the core for deviations from the empty state. When it finds something, it strikes precisely instead of scattering.

Here it pays to recall step 1: empty core is DAT.F $0, $0 everywhere. So a scanner looks for cells that appear different. That has an immediate consequence which surprises many people — bombing with DAT 0, 0 leaves no trace such a scanner can find. The bomb is lethal and invisible at once.

A quickscanner is a particularly fast opening: in the very first cycles it checks a few widely spaced locations. If it hits, the battle is often over before the opponent has properly started. If it misses, the warrior switches to its actual strategy. That is why strategy descriptions so often read qscan -> something: the quickscanner is not a warrior of its own but an opening.

Imp — the immortal

You know it from step 1. Copies itself forward and runs after the copy. Cannot win, but is hard to kill. Today it almost never appears alone but as a fallback: if the main body dies, the imp keeps running and at least salvages the tie.

Vampire — the slaver

A vampire bombs not with DAT but with JMP instructions pointing into a pit inside its own code. An enemy process running onto one is not killed but redirected — and works for the attacker from then on, until it starves in the pit.

Elegant, and very effective against individual processes. Not against a flood of them: you cannot enslave faster than the opponent creates new processes. The numbers below show this starkly.

Oneshot — the sniper

A oneshot scans once, very thoroughly, and stakes everything on that one hit: if the opponent is found, a devastating attack follows immediately. If not, the warrior usually falls back on an evasive strategy.

Clear — the bulldozer

A core clear runs systematically through memory and overwrites everything. No searching, no aiming — just area. Rarely used alone, usually as an endgame after a scanner has roughly located the opponent.

And now the truth

About the relationships between these classes there is one sentence the scene has been passing along since the early nineties:

Scanner beats paper, paper beats stone, stone beats scanner.

A clean rock-paper-scissors. It appears in tutorials, in forum posts, in source comments.

We computed it. The basis is 349,866 pairings of 1000 rounds each from the Koenigstuhl field — 837 warriors whose class could be determined reliably from their authors' own strategy notes.

Of the triangle's three edges, exactly one holds.

Why the folklore no longer holds

Not because it was wrong back then. Because the field has changed.

"Paper" today almost never means a pure paper. The most common construction in the field is the combination paper with stone — a replicator that bombs on the side. And the comparison shows why that prevailed:

The hybrid wins not just on average but in every single column. That also solves the puzzle from the previous table: the anti-scanner strength of the "paper" class comes largely from the stone components of hybrids. The folklore describes pure forms that barely exist in the field any more.

The quiet king

A finding that appears in no tutorial: the oneshot is the only class without a weakness. It wins against every other class — against paper by +43.9, against clear by +46.9, against imp by +61.8.

That does come with an honest qualification, or you will draw the wrong conclusion: the oneshots in this field are also its elite (best class average, best median rank). So the matrix measures the oneshots that were actually submitted — not the class as an idea. Oneshots leading does not mean your first home-built oneshot will win.

Other clear edges

• Paper crushes vampire: +75.0. The most pronounced effect in the whole matrix. Pits do not scale against floods of processes — precisely the theoretical weakness of the class, here as a number.
• Scanner eats imps: +54.3 and clears: +42.4.
• Paper against paper: only 117.7. Paper wars almost always end in ties. Two populations that cannot exterminate each other — just like the two imps from step 1, only larger.

What to take away

First: pure forms are teaching models. What wins on hills are hybrids. When you build your first serious warrior, the question is not "which class" but "which two, and how do I combine them".

Second: check received wisdom. The rock-paper-scissors sentence has been passed along for thirty years and is now two-thirds wrong. Not through sloppiness — the field moved out from under the rule and nobody measured.

You can do exactly that yourself today. How not to fool yourself with randomness in the process is the topic of the next step.

All figures from the class matchup matrix of the Koenigstuhl field (status July 2026): 837 classified warriors, 349,866 pairings of 1000 rounds each. Classification comes from the authors' own ;strategy notes; 267 warriors without a reliable classification were excluded rather than guessed.

← Core War overview