Mind Over Magnet

Planning the process cut his next game from three years to seven months. His second game, Word Play, shipped in seven months by cutting scope hard and building in layers. He credits better process planning, alongside having already learned Unity, as the difference.

Zach, who scored Mind Over Magnet, got first dibs on Word Play's music.

The whole game grew from one borrowed mechanic: a magnetic glove. Mark Brown built Mind Over Magnet around the magnetic glove from Zelda: Oracle of Seasons, used there to cross a platforming level. It was his first game, made while learning Unity.

He groups it with Braid, which began from Prince of Persia's rewind, and Shovel Knight, from Zelda 2's down thrust.

Skipping a production pipeline made the game slow and weaker. Not knowing what to do and when meant content had to be redone, and the game took over three years, announced at the end of 2021. He says it ended up less good than it could have been.

Afterwards he advises prototyping before committing, since "our brains are terrible video game simulators." He finished with help from composer Zach Jones, artist Grayson Evans, Rahul Gandhi on contracts, coders Sophie Stevens and Dom Harris, playtesters and Patreon supporters.

Content was frozen, with one exception: a post-game developer commentary mode. After the freeze he only polished existing content, except for a commentary replay unlocked after finishing the game, where speakers in some levels play voice notes about how they were made.

It was inspired by Valve's commentary in Portal and Left 4 Dead, which first got him interested in game design as a craft. Keeping it behind completion leaves the first run uninterrupted. Release polish also included making sure the game fills the Steam Deck screen fully.

Testers were told to report every problem, big or small. A volunteer team from the GMTK Discord played the whole game with a fine-tooth comb and reported almost 100 bugs.

Collecting everything first and triaging later kept testers from filtering out useful reports: "No matter how big or how small, just tell me about it."

Scoring every bug 1 to 5 got game-breaking ones fixed first. Each bug went into a spreadsheet scored from 1 (harmless visual glitch) to 5 (critical, game-breaking), then sorted so he worked down from the worst.

The aim was to fix the important problems before running out of energy and deciding the rest would be fine.

Fixes created new bugs, so testing repeated until nothing new appeared. Some fixes broke something elsewhere, so the game went through several QA rounds. After a few weeks no new reports came in, and only then did he build the release and upload it to Steam.

He notes the testers had by then either found everything or got bored.

The worst bugs came from mashed toggles, odd bindings and badly timed switches. Mashing the change polarity button could leave a platform stuck midway, and turning magnetism off just before Magnus connected bounced him off the ceiling at an awkward angle. A Next Fest demo player emailed that binding move to the mouse scroll wheel made Uni speed along like a jet engine.

The lesson is to stress-test every state toggle with rapid presses, flip states around contact points, and check what happens when inputs fire far faster than a thumbstick would.

Broad playtesting drove a genre change and a new art style. Over three years, feedback from friends, family, Discord members, strangers recording themselves and professional developers led him to change the game's genre and its entire art style.

Skipping a production pipeline made the game slow and weaker. Not knowing what to do and when meant content had to be redone, and the game took over three years, announced at the end of 2021. He says it ended up less good than it could have been.

Afterwards he advises prototyping before committing, since "our brains are terrible video game simulators." He finished with help from composer Zach Jones, artist Grayson Evans, Rahul Gandhi on contracts, coders Sophie Stevens and Dom Harris, playtesters and Patreon supporters.

Mark Brown cut the plan instead of moving the deadline. The game went from 50 puzzles to about 40, five worlds to four, and three magnets to two, with a simpler story. With the smaller plan it was playable from intro cutscene to credits within a couple of months.

He dropped Max, the throwable magnet, and brought the switcher magnet in earlier with polarity switching unlocked later, which saved assets and streamlined the story. The bug-prone scissor gate and the mighty magnet, which had no levels yet, were also cut. The game was already over three years old, and his private goal was to finish by December 31st, 2024: "If I'm not willing to change the deadline, I guess I've gotta change the plan."

A complete playable build showed exactly where the game was weak. Playing the whole game exposed a dull story and an anticlimactic section where players finally had every magnet but only a few levels. He added cutscenes with a small twist, more puzzles and a bigger finale.

He also found he missed the scissor gate, so he fixed its bugs and added a couple of levels using it. Playtests and a close read produced a long list of problems, including a bug when a magnet dies mid-dialogue and magnets sticking on corners.

A short final world and transition stages fixed the pacing. World 4 had grown to about 20 levels against about 15 in the others, so he split the game back into five worlds. The last world, on the factory rooftop, is deliberately much shorter so the game ends on a high.

Players who already have all the magnets would not want another long world of hard puzzles. Moving between worlds felt flat, so he added a short transitional stage at each key point.

After locking content, small daily fixes improved the game faster. While adding content the game got longer but not much better; once content was locked, polish improved it with every update. Daily playtests led to fixes like a pop-up pointing to the hint system when a player is stuck for a set time.

He added dialogue a tester wanted, and changed levels whenever a playtester skipped a puzzle in an unexpected way. Testers began clearly having fun, and he says "the morale boost of a good playtest cannot be understated."

No one tells you a game is done, so call it yourself. He set the Steam release for November 6th, 2024, exactly ten years after the first Game Maker's Toolkit episode, and kept the last three months for polish, playtesting and marketing rather than new levels.

Calling it means accepting unfinished content and unexplored ideas. Other developers he spoke to said shipping less than planned is normal: "There is no external sign that the game is finished."

He planned launch about a month after the June Next Fest. Working back from a June showcase, he targeted a July or August release, treating Next Fest as the main marketing moment.

The model was Balatro, which launched a couple of weeks after its February Next Fest, and Pepper Grinder, about a month after. The game ultimately launched on November 6th, 2024.

Large pixel art could not scale cleanly to 1080p, the commonest PC resolution. Big sprites filled a 1280 by 720 room, which scales by whole numbers to 720p, 1440p and 4K but not to 1080p, which is 1.5 times larger. He had made the sprites long before the camera system and hoped they would fit.

Letting Unity round pixels made angles and curves look squiffy, anti-aliasing looked blurry, and borders made everything small. Unity's pixel perfect camera did not solve it; he called his assumption "dumb and naive."

Doing the real fix took less time than hunting for a compromise. He tripled every sprite in size and redrew it as 4K-ready vector art with Photoshop's pen and shape tools. The remake probably took less time than bouncing between workarounds.

Every button, door, decoration, magnet and UI element was redone, and he felt Magnus gained charm and the game stopped looking like a generic pixel art indie. "Sometimes it's just better to bite the bullet, knuckle down, and do the hard work to make an actual real solution."

The remake kept simple shapes and only blurred the old backgrounds. No extra detail was added, so objects stay readable on small screens like the Steam Deck. Backgrounds stayed pixel art under a depth of field blur instead of being redrawn.

The original art was already made of simple chunky shapes and basic colours, which made the remake quick.

A trailer faked from unfinished content pushed wishlists past 30,000. With only World 1 finished, he built a trailer in about a day from unplaytested levels, half-finished backgrounds and partial mechanics. It helped push the game past 30,000 wishlists before Next Fest began.

World 1 alone would have shown the same turquoise sewer for two minutes, and some shots still use the blueprint development background. Viewers congratulated him on finishing; his reply: "It's not actually done. It's all fake!"

The trailer moves from one simple mechanic to Magnus to a montage. It opens with magnet beams lifting blocks, slows to introduce Magnus and his mechanic, then runs a fast montage of later magnets, mechanics and worlds, ending on a call to wishlist.

He recorded in 4K for a 1080p trailer so he could crop without losing quality, and used a stock Epidemic Sound track with separate stems so cuts land on the drum beat while the composer was not yet ready.

The demo had to represent the whole game, not just World 1. Because World 1 alone would misrepresent the game, he planned to add World 2, or half of it, which meant building more content first.

He planned launch about a month after the June Next Fest. Working back from a June showcase, he targeted a July or August release, treating Next Fest as the main marketing moment.

The model was Balatro, which launched a couple of weeks after its February Next Fest, and Pepper Grinder, about a month after. The game ultimately launched on November 6th, 2024.

A fixed weekly routine alternated new content with bug fixing. Tuesday to Thursday went to new content, Friday to bug fixes from a Trello list and a new build, and Monday to watching playtest footage.

It was modelled on Valve's approach to building Portal. Friday builds meant testers played over the weekend, where midweek builds had left him idle waiting for footage. Monday notes covered bugs, polish gaps and levels that were too easy, too hard, too fiddly or exploitable, so early levels became heavily polished.

Each build went to at most three playtesters. Giving builds to too many people produced more footage than he could watch, and the same bug repeated across many videos. Capping it at three gave a manageable amount with more distinct issues.

The testers were recruited through Twitter.