Quinn Labs · Closed experiments

The Graveyard

Most laboratories publish the wins. The results that shaped this place were mostly failures.

Each entry is an experiment we stopped, with the reason we stopped it and what it taught us.

A dead end is still a result. Writing it down saves someone else the same month.

Bin-mounted camera rig

Not technically viable

Can a bin know what it's holding?

What we tried

A weatherproofed camera and classifier mounted inside commercial waste bins to identify material as it was thrown in.

Why we stopped

The lens fogged within two days outdoors. Condensation and dust beat the classifier long before accuracy mattered.

What it taught us

Environmental sealing was the experiment. The model was the easy part.

Would we revisit itWorth reopening with a sealed optical window and a heater, or no optics at all.

Self-repairing coating trial

Too expensive today

Can products repair themselves?

What we tried

A self-healing polymer layer applied to high-wear plastic components so surface damage closed itself.

Why we stopped

The chemistry works. Per-unit cost was roughly nine times the part it protected, and the healing cycle needed heat the product never sees.

What it taught us

Materials that work in a lab still have to survive a bill of materials.

Would we revisit itRecheck when the base polymer becomes a commodity rather than a speciality order.

Roadside energy harvesting

Right science, wrong timing

Can roads become energy storage?

What we tried

Piezoelectric strips under a short section of yard surface, harvesting energy from repeated heavy vehicle movements.

Why we stopped

The physics held. The yield was in watts, the installation cost was in thousands, and the surface needed relaying to install it.

What it taught us

Anything requiring civil works has to beat the cost of digging, not the cost of electricity.

Would we revisit itOnly sensible where a surface is being replaced anyway.

Universal parts recogniser

Not technically viable

Can furniture explain itself from a photo alone?

What we tried

Identifying every flat-pack part from a single phone photo, with no tags, codes or manufacturer data.

Why we stopped

Similar panels are genuinely indistinguishable at a glance — the difference is often a hole position on the hidden face.

What it taught us

Recognition needed the manufacturer's own data. That reframed Furniture DNA from vision-only to a shared data layer.

Would we revisit itAlive again as Furniture DNA, with tags and manufacturer collaboration instead of guessing.

Driver behaviour scoring

Problem wasn't big enough

Can scoring make drivers more cooperative?

What we tried

A cooperation score shown to drivers after each shift, ranking courtesy events across the fleet.

Why we stopped

Drivers found it patronising, and the behaviour we wanted was already happening. Scoring added stress rather than removing it.

What it taught us

The goal was less friction, and a leaderboard is friction. KindRoad now shows a single acknowledgement and nothing else.

Would we revisit itNo. This one is closed on principle, not on technology.

None of these were wasted. Every one of them narrowed the question that came after it.

See the questions still open →