Build it. Watch what reality does to your assumptions. Change it.

The idea was a volcano made of ice. Centerpiece of a toddler chemistry experiment and a flamingo-rescue storyline. Simple idea. The build was not.

The first molds were castle-shaped, dyed purple, frozen in whatever container was on hand. They looked good. They didn't erupt the way a volcano needed to: the reaction channel was in the wrong place, the walls were too thick to melt fast enough for a toddler's attention span, and the color bled everywhere before the baking soda even hit the vinegar.

So the shape changed to a cone, the channel moved to the center, and the mix ratio got rebuilt from scratch through repeated freezer tests. None of that was visible in a sketch. It only showed up once real ice, real dye, and a real toddler's patience were all in the room together.

Safety mattered just as much as chemistry. Real goggles, sized for small faces, non-negotiable before anyone got near the reaction. The eruption itself is loud, fizzy, and a little unpredictable, exactly the kind of moment where a kid needs to feel both safe and fully in it.

Ice Volcano was designed around a flamingo-rescue storyline: a flamingo and duck trapped in blue-tinted ice, kids working the "rescue" with salt, warm water, and patience. One kid took the leftover ice bin, decided the flamingo needed to be preserved in a glacier rather than melted out of one, and spent the rest of the session building his own ice structure around it instead of following the rescue script.

That's the moment that shaped how I think about designing for unpredictable users. He wasn't using the material wrong. He had another hypothesis, and the experience was open enough to let him test it. That's the standard now: not "did the kid follow the plan," but "did the kid have enough room to have their own idea."

The stomp rocket bubble launcher started as a straightforward air-pressure demo. A kid decided the launcher would work better as a net, taping his own modification onto it mid-session to catch bubbles instead of launching them. Rebuilding the loop, watch, notice, change, doesn't require a big project. It shows up in five minutes with tape and a kid's own idea, just as much as it shows up over months of ice-mold iteration.

Splat Lab moves the canvas outdoors, propped against a tree trunk under the redwoods, and tests bubble foam as a painting material instead of just a sensory one. Same underlying question as Bubble Lab: what happens when soap film meets a surface. Different context, different answer, another version to build from.

A multi-level water ramp, built from PVC gutters mounted on stacked milk crates, let kids pour water and toys from the top and watch them travel down to a bin at the bottom. The gutters held their angle and the water ran clean, no pooling, no redesign needed.

What kids actually did with it was more interesting than what it was built for. They kept gravitating to the bottom level, lifting the gutter off its support to lay it flat, then setting it back at an angle on purpose, watching how a steeper drop pushed their treasures down faster and further into the bin. Nobody asked them to test the relationship between slope and speed. They just wanted to see their stuff move faster, and ended up running the same experiment a physicist would.