Word Play

Launch revenue tails off but Steam sales and festivals keep it going. Word Play launched in summer 2025 on PC, Mac and Steam Deck with a 10% first-week discount and a free demo. It made about $44,000 in 2026, roughly $400 a month.

Most money came at launch, then small spikes every time there was a Steam sale or festival.

Blend many influences so the game is more than one source. Word Play draws on Balatro, Bookworm Adventures, Scrabble, Slay the Spire and Vampire Survivors, with music inspired by The Sims and Wii menus.

Mark keeps a deliberate stance as an accessible, breezy casual game and rejects complexity like buffable perks even where rivals add it.

Assume AI can rebuild your game's core in minutes. Two hobbyists cloned Word Play with Claude Code in 11 days; from a finished spec, Claude had a playable core in 12 minutes, against seven months for the original. Given a list of 21 modifiers, Claude added them right away.

Mark says writing those modifiers was months of his life. The clone sits online behind an access code for friends and family.

Polish in UI, sound and music is what a clone can't match. Mark expects many players to still choose the original because it has better UI, sound design and music. He calls the clone's vibecoded UI bad-looking.

Unfinished post-launch features leave a gap a clone can fill. The clone added a timer mode, daily challenge, multiplayer and meta progression. Mark had half-finished daily and speed-run modes in the code but was burnt out and drowning in technical debt.

He admits the clone is superior in some ways, and that buyers don't care about a developer's struggles.

Modding support lets fans extend a game and still pay for it. Mods require buying the original, while the clone gives players Word Play without paying Mark. Word Play only had a janky way to add dictionaries and letterbags.

Unity games are easy to decompile, so AI can mod them anyway.

Guardrails and simple-mechanic warnings won't stop cloning. Anthropic added two sentences to Claude Code's prompt to steer it away from recreating specific games, but prompts can be circumvented and rival models may not care about IP.

Mark rejects Valve's advice to avoid easily copied mechanics, citing Tetris, Fruit Ninja and Wordle.

The clone makers asked first, and Mark put the decision to his audience. The developers emailed Mark before release and said they would keep it private if he objected. He paused the release, made a video to work out his view, and asked viewers for comments and a poll.

"We won't do anything you dislike." He is glad the game inspired them but needs to pay his mortgage.

Perks that react to the word spelled create real strategy. The first word-specific modifier boosted the score when the first and last letters match. Mark says these mechanics proved very important to strategy and balance.

Another example multiplies the final score by 1.5 if the first letter is a vowel.

Ship the jam with if-statements, then rebuild for scale. The jam prototype checked each modifier with if statements inside the score script, which worked to prove the idea. For the real game Mark needed a system that scaled to 120+ modifiers and was easy to test, change and debug.

The if-chain grows with every perk, and changing "first letter" to "last letter" means hunting through all the code. Balatro-style ordered triggering was a requirement from the start.

Give every perk its own script inheriting one abstract base class. An abstract ModifierClass has a virtual onWordScored that returns -1 by default, meaning nothing happened. mVowelFirst overrides it to return the score times 1.5 when the first letter is a vowel.

Scores are never negative in Word Play, so -1 is free as a sentinel. Because every script is guaranteed to have the function, one list can hold all perks.

Resolve perks in the player's order and ping the one that fired. The score script loops over owned modifiers in the player's arranged order, and any result other than -1 updates the score and triggers a ping animation and sound.

One loop replaces 100 if statements, and feedback costs one line in the loop rather than code in each perk.

Add new hooks to the base class to reach every game event. A bool OnUpgradeUsed returns false by default, so perks that ignore it are unaffected. Hooks cover selling a modifier, scoring each letter, losing a life, game over, opening the perk screen, rerolls, new rounds, refreshes and adding tiles.

One modifier raises its bonus points whenever an upgrade is used.

Keep each perk's data in its script for inspector tuning. Each modifier holds its few lines of logic plus icon, description, rarity and numbers exposed in the Unity inspector, which let Mark add, debug and balance more than 150 perks.

Reuse the same inheritance pattern for other item families. Upgrades inherit a base class with OnTileDropped, overridden to destroy a tile, turn it emerald or raise its score.

Mark suggests the same pattern for enemies sharing movement and health, or guns sharing aiming and reloading.

A two-day jam prototype proved demand before real production began. Mark Brown had carried a Balatro-plus-spelling idea since early 2024 and used a Patreon-only game jam in December to try it. The prototype, called Wordy, got a great response, so he started a fresh Unity project right after the jam.

The finished game sold more than twice the copies of his previous game. He says the prototype's enthusiasm kept him going through the rest of development.

Explore the core loop with physical stand-ins before writing code. Mark tested Word Play's loop with Scrabble tiles and a deck of handmade modifier cards.

The broader advice is to prototype in whatever medium reaches the idea fastest, whether paper, a spreadsheet or a non-final engine.

A two-day jam prototype proved demand before real production began. Mark Brown had carried a Balatro-plus-spelling idea since early 2024 and used a Patreon-only game jam in December to try it. The prototype, called Wordy, got a great response, so he started a fresh Unity project right after the jam.

The finished game sold more than twice the copies of his previous game. He says the prototype's enthusiasm kept him going through the rest of development.

Build a satisfying spelling game first, then add systems in layers. Mark set out to make a really good, juicy spelling game before any perks or rounds. The order was spelling game, then roguelike loop, then perks, then bug fixing and balancing.

A shuffle button and letter bag came first, then the shop, perks and special tile architecture. Each layer sat on a solid one, which he says kept development faster.

Test perk functions with a few perks before scaling content. The first version had 11 perks, built mainly to test whether a boost hits one tile or the whole word. Once the architecture worked, perks grew to 52, 74, 90 and finally 160.

Cut scope by picking whichever option takes the least time. Mark judged choices by development time, even when the game was arguably worse for it. Using simple icons for all 160 perks instead of unique art like Balatro's jokers saved weeks.

Word Play shipped in seven months, against more than three years for his first game, Mind Over Magnet. He credits planning the process and is proudest of finishing without losing his mind.

Design input for mouse, controller, keyboard and touch from the start. Spelling had to work seamlessly on every input, which supported playtesting on iPads and a release on PC, Mac and Steam Deck.

Once the MVP was done he added accessibility features for a word game: alternative controls, dyslexia-friendly fonts and spelling suggestions, plus difficulty modes.

Get a full loop playable early to collect real playtest feedback. With the loop complete, Mark shared builds on Discord, took an iPad to schools and played it with developers at GDC. Feedback led to rearranging tiles on the board and bonus points for long words.

Without the length bonus, a single mega word barely scored more than a short one. GDC feedback came from developers behind Patrick's Parabox, Spell Tower and Alto's Adventure.

Similar games arriving signals a genre, not a reason to cancel. In February Mark found Wordatro and Birdigo, both mixing Balatro with words, and briefly considered cancelling. He kept going, focused on his own game's quality, and chatted with both developers.

More word roguelikes followed, so it felt like joining a genre rather than copying one game. He regretted not showing the game earlier. "I can't control what anyone else makes. The only thing I can control is how good my game is."

A Steam Next Fest demo brought 20,000 players and concrete feature requests. The June demo drew 20,000 players in a week, with Yahtzee and Northernlion playing it. Feedback led to submitting words over 10 letters and a way to petition for new dictionary words.

Before the demo, Apple Arcade took ten days to pass on a TestFlight build; after it, a different Apple team offered App Store help.

A final QA group of patrons and sharp demo players tuned balance. Testers found bugs, typos, and perks that were overpowered or pointlessly weak, and helped shape the difficulty curve and Steam achievements.

They also pushed Mark to build features he had put off.

Launch revenue tails off but Steam sales and festivals keep it going. Word Play launched in summer 2025 on PC, Mac and Steam Deck with a 10% first-week discount and a free demo. It made about $44,000 in 2026, roughly $400 a month.

Most money came at launch, then small spikes every time there was a Steam sale or festival.

Name the resource after the action, not the loss. Losing a life for spelling a word correctly felt wrong, so a Discord user's suggestion of "plays" replaced lives. It also gave the game its name.

Every other word game name was taken, and Word Play did not exist on Steam.

Perks that react to the word spelled create real strategy. The first word-specific modifier boosted the score when the first and last letters match. Mark says these mechanics proved very important to strategy and balance.

Another example multiplies the final score by 1.5 if the first letter is a vowel.

Special rounds and special tiles vary how players use the same tools. Special rounds demand six-tile words, a mandatory tile or lose two plays, or a word-length cap that starts at four and grows. Tiles include golden, diamond, emerald, potion, dot, glass, mirror, exclamation mark, locked and compound tiles like ING.

The glass tile is a one-time copy of any other letter.