I build things, ship them, then read the data honestly.

A shooting incremental: every hit permanently raises your damage.

• Concept to published in one day.
• Architecture: every player has their own health bar on the same shared wall. I simulated the balance before building and confirmed it live: 75 hits took the wall from 2,800 to 0, exactly as modelled.
• Epic gave it a test of 25,235 impressions. Click-through on the Homebar was 23× higher than in the General Rows, which meant the concept sold but the thumbnail failed at small size.
• 48% of players left inside the first minute, so I diagnosed the early game as the problem.
• Release 3 added V-Bucks purchases, two shops, fast travel (my cousins suggested it after playtesting) and visible pet effects.

Test impressions from Epic

25,235

Homebar vs General Rows click-through

23×

Left in the first minute

48%

Hits to break the wall, as modelled

75

Football where your gun is your boot: shoot the ball to pass, shoot it to score.

• I chose a physics prop over Epic’s Ball Spawner after reading their open bug reports on it.
• V-Bucks store with zero randomness, by design, to stay clear of Epic’s gambling rules.
• The honest result: 157 players on day one, then almost none. The cause was an 8-player mode that can’t fill a lobby at 37 concurrent players, plus the wrong genre tag. That’s a cold-start problem, not a gameplay one.

Players on day one

157

Players one match needs

8

Concurrent players at the time

37

A construction-equipment catalogue, built solo in 30 days.

• The team edits it through its own admin panel.
• A public MCP API lets AI agents query the catalogue.

Equipment models

335

Categories

77

Spec rows

1,394

Languages

4

A Telegram bot qualifies leads, reads payment receipts with AI vision, and enrols buyers after one human tap.

• Crisis messages never reach the AI; a human is alerted instead.
• 321 automated tests.