Skyline
Salvagers
A game for 8- and 9-year-olds where what you collect is what you build with.
Role
Game design, AI prototyping, research
Timeline
4 weeks · Fall 2026
Team
3 designers · INFO 697
Pepper the mail pigeon needs to get a heavy crate across a gap between floating islands. Kids salvage parts from junk, build an invention, test it, and change one thing when it misses.
The challenge
Eight- and nine-year-olds are natural sorters, comparers and testers. But the games they play hide value, nudge them toward spending, and rarely let them do anything with what they collect.
How might we make collecting something kids use to build or solve things, and let them see how their finds can be sorted?
Salvage: a spring from a squeaky couch
Build: one setting per part
Test: the dotted line keeps the last try
My approach
What we found
They think in things, not ideas. Kids this age sort, order and compare confidently, but only with visible, concrete things.
They experiment, but not like scientists yet. Changing one thing at a time is still hard, so the game has to make it easy.
They can’t reliably spot a trick. Most kids miss manipulative design, so we left it out entirely: no timers, currencies, ads or purchases.
In my teardown of Prodigy Math, a right answer turns straight into magic, so the feedback feels tied to the child’s own action. But the math sits in front of the fun instead of inside it. Across all three games, collecting mostly marked progress. That left an open space: a game where what you collect is the tool you play with.
My Prodigy Math teardown
My Crazy Eights: eight collecting worlds in eight minutes
What we designed
One loop a child can understand by looking, tapping and watching. A miss names what happened and keeps the build, so trying again is cheap.
Salvage
Build
Test
Change one thing
New blueprint
The core loop. Three ways across: launch & catch, push across, ramp jump.
What we learned
What we built
Kids reason best about things they can see
Each part is named by its job: Spring · Launches
Testing one variable at a time is hard
One setting per part, three named steps; the last try stays as a dotted line
Dead ends frustrate
A question after two misses, a tip after three, never the answer
Reading varies a lot
Prompts read aloud, icons beside words
Key decisions
My first build let kids drag parts, stretch the net and squish the spring by hand. Too fiddly for this age. It became: tap a part, then pick one of three big buttons. The challenge moved from the hand to the decision.
The build screen, Nov 1 → Nov 4
A critique found three names for the same things and a weak payoff when a build first worked. Now the world bursts first, then NEW BLUEPRINT UNLOCKED, then two clear choices. And one vocabulary everywhere: Parts, Blueprints, Treasures.
The blueprint moment, before and after
Testers couldn’t tell how to move the spring. I built a bigger MOVE SPRING control with a glowing spot, an arrow and a demo hop, and it tested more confusing. I reverted it. The fix that stayed: the button names where the spring goes, “Move ↓” or “Move →”.
Testing
I built the game on my own with AI coding tools, over 53 versions in a week. I wrote every prompt; the AI wrote the code. Each round, I watched where players hesitated and made one targeted fix. Classmates finally played the full loop alone, without instructions.
First drafts with my notes (“looks too dismal… too blocky”) → 5th iteration
Players didn’t know how to start. The nearest junk now wiggles: “Tap the object to salvage it!”
Players got lost after collecting. Pepper points back to the workshop, which pulses.
Blueprints felt unearned. A bigger celebration, and the blueprint pinned to the wall.
Outcome
My role
I built the whole game on my own: I took a teammate’s concept, refined the gameplay and wrote every AI prompt across 53 versions. On the research side, I did the Prodigy Math teardown and explored themed-world concepts.
What I learned
The best fixes came from watching exactly where players hesitated, not from adding features. And designing for kids often means deciding what the game will never do.
