Runekeeper

Runekeeper is a js13k 2023 entry from Ariel Wexler, and its design notes in the repository are refreshingly blunt about what it's trying to do: "the drawing mechanic should driving everything — it's the fun part," with mobile thumb-drawing listed as a bonus goal. That framing turned out to be the right lens for reviewing it, because the drawing mechanic really is the whole game, and it really is the fun part — when it works.

A trained classifier in 13 kilobytes

Open the build and a wizard's tower sits beside a 6x6 grid populated with a runestone, potions and slowly advancing enemies with HP bars. Below the grid is a drawing pad, split into a left half and a right half, with the same six abstract symbols (an upward chevron, a circle, a lightning bolt, a downward chevron, a wave, and an X) listed as column and row headers on the grid itself. Draw one of those six shapes with a mouse and the game genuinely recognizes it: a hand-drawn circle in testing was correctly classified and moved the runestone one row, changing the header text from "DRAW SYMBOLS TO MOVE THE RUNESTONE" to "0 ENEMIES SLAIN" — confirming the turn had actually advanced, not just that a stroke was drawn. Repeating that circle five times in a row and then drawing a zigzag on the other half of the pad kept producing consistent, correct results: the runestone advanced through the grid, enemies with visible HP bars closed in, and a colored trail animation played that matched the drawn shape's assigned color. Reading the source confirms what's happening under the hood is not a simple shape-matcher — it's an actual small neural network, weights encoded and shipped inline in the page, that resamples the drawn stroke onto a 28x28 canvas and classifies it, voting across twenty jittered resamples to settle on an answer. Shipping a trained ML model inside a 13KB competition build is a genuinely unusual and impressive piece of engineering, and it works.

The one platform it names as a goal is the one where it didn't work

This shelf exists because Flash games cannot be played on a phone, so every review here treats touch as the test that actually matters — and this is where Runekeeper could not be verified as playable at all. Five separate tests were run, each varying the method or parameters to rule out a one-off tooling fluke: CDP-level touch event dispatch (the same mechanism a real phone playtest uses, with proper touch radius and force values), a synthetic TouchEvent construction, two different circle radii and point densities, a slower stroke cadence, and two different viewport sizes. Every one of them, run under standard mobile-browser emulation, produced a runestone that never moved and a header that never advanced past "DRAW SYMBOLS TO MOVE THE RUNESTONE" — despite directly confirming, via an injected diagnostic listener, that the touch events reached the game's own event handlers with the correct on-screen coordinates every time. The moment the identical gesture was retried with mobile-device emulation turned off — same touch dispatch mechanism, same coordinates, same shape — it worked immediately. That comparison points at something specific to how the game (or the browser's mobile-mode touch handling) processes a touch-drawn gesture rather than a mouse-drawn one, though it should be said plainly that this was tested in a headless emulation environment, not on a physical phone, so a device-specific quirk on either side can't be fully ruled out. What can be said with confidence is that touch drawing did not work in any configuration tried here, and there is no keyboard or button alternative to fall back on — drawing is the only way to play, and it was only ever confirmed to work with a mouse.

Verdict

Runekeeper deserves real credit for its idea and its engineering: a working, trained gesture classifier packed into a 13KB browser game is not something this shelf sees often, and the turn-based loop built around it — cast a rune, watch the runestone move, watch enemies close in — holds together well once you're actually able to draw. But the review has to be honest about what was and wasn't confirmed, and on a shelf built entirely around phone playability, the one thing that could not be confirmed working, across a genuinely thorough set of attempts, was touch. A game whose author explicitly hoped it would work on mobile, and which currently has no way to play it at all without a mouse, is not ready for this shelf regardless of how clever the mechanic underneath it is.