The first version of BELTRUNNER was not very good.
Matt Sephton built an orbital shooter with numbered targets sitting on fixed shells around a sun. It came together quickly in his new Jinks system, but the targets became sitting ducks and the interesting dodging disappeared.1
He threw away the orbital structure.
He kept the numbers, the chain and a few visual ideas, then moved back toward Asteroids.
That is what makes the BELTRUNNER postmortem useful. It is not a story about one perfect idea executed cleanly. It shows a new game borrowing lessons from decades of arcade design, testing them, discarding some and turning the survivors into its own rules.1
Start with a game the player already understands
The first wave deliberately looks like Asteroids.1
One large rock, familiar thrust and drift, splitting asteroids. The player gets time to learn movement and shooting without a heavy tutorial.
Clear the field and a black hole opens. From wave two, the actual race appears: numbered gates have to be collected from 5 down to 1 while the asteroid field keeps moving.1
Sephton cites Donkey Kong '94 as inspiration for the fake familiarity, starting with something recognizable before revealing the different game hidden underneath.1
It is also a teaching device. Players who know Asteroids can begin from muscle memory. Everyone else still gets a full wave with the core movement before another demand arrives.
The tutorial is the first piece of play.
Introduce a mechanic, then leave it alone long enough to learn
BELTRUNNER has sixteen waves in four acts of four.1
Each act introduces a new gate behavior, then gives the player several waves to practice before moving on. Sephton calls the rhythm “introduce → practice → practice → close” and describes it as a Nintendo-like way of teaching mechanics.1
The important part is not that four is a magic number.
It is the refusal to stack novelty immediately. A mechanic shown once has not been learned. The player needs time to be surprised, understand the failure and start predicting it.
Waves that keep the previous setup can even be mostly empty in the game data because only changes need to be described.1
Make randomness into a course
BELTRUNNER uses a deterministic pseudo-random stream.1
Rock positions, headings, gates and power-up scatter all come from a seed. Given the same seed and inputs, the run can be reproduced.1
Sephton prefers that because the field becomes a place players can learn, closer to memorizing Pac-Man patterns or Donkey Kong routes. It also helps with replay and tuning: change a rule while replaying the exact same situation.1
Even drops avoid a simple probability roll. Every eighth rock kill arms a power-up following a fixed sequence.1
The game can feel chaotic without being arbitrary.
Teach a secret by removing a sound
From wave two, one large rock is highlighted. Breaking it adds a sustained voice to a chord: root, third, fifth, octave. The player has to follow the highlighted sequence.1
Hit the wrong rock and the voices stop.
There is no explanatory text. The mistake becomes audible.1
Complete the sequence and a clover-shaped rock eventually releases a secret EXTRA bonus. Before discovery, the legend shows ????; afterwards it shows EXTRA.1
It is a neat use of audio as feedback. A small visual marker can disappear inside a crowded asteroid field. A chord suddenly collapsing is much harder to miss.
The game does not explain the secret. It gives the player enough evidence to form an hypothesis.
Bad ideas teach the engine what it needs
The postmortem contains many smaller changes created by actual play.
Round gates felt unfair on side grazes, so they became elliptical. The STOP bonus initially froze rocks but left the race clock running, turning the supposed breather into a quiet punishment; the clock was frozen too. A WARP shortcut originally appeared on the same wave that introduced a mechanic, exactly when the player needed to stay and practice, so it moved later.1
Other problems forced Jinks to gain polygon colliders, oriented ellipses, more truthful torus wrapping and better physical deflection.1
Sephton says those changes do not remain BELTRUNNER-only hacks. Useful mechanisms become general engine features for later games.1
That is a practical way to build a creative tool: do not invent every feature an engine might theoretically need. Make a complete thing, let the thing break the tool, and generalize only what the work actually demanded.
Jinks is still unfinished
Jinks itself is not released yet.2
Sephton describes it as a lightweight engine, interactive development environment and game-focused language. A platform-independent core produces a per-frame contract while thin platform shells handle drawing, sound and input.2
He says the same core already runs on web, Apple platforms, Linux and older hardware including Wii and Dreamcast.2 We did not independently reproduce those platform claims.
He also reports small source sizes: 475 lines for his Asteroids recreation and roughly 1,278 for BELTRUNNER.2 3
Line count is not a quality metric. Here it indicates the design goal: keep game rules readable and push platform work elsewhere.
BELTRUNNER is a useful test because it forced that system to grow in concrete ways.
A creative tool does not prove itself by having the longest feature list. It proves itself when a complete project survives long enough to discover which features were actually missing.