Est. 2031 · Bench 04 · Somewhere with bad weather and good bandwidth
Creative technology research
Tools for people who make things on screens.
Enter the bench Six experiments running. None of them are products yet.
Bench 04 · manifest · 2031—
We are four engineers, two illustrators and one very patient machinist. We take a question from someone who makes things, and we carry it all the way to a prototype that sits on their desk and does the thing.
Instruments. Colour readers for people who mix their own paint. Type engines that breathe. A haptic brush that lets you feel a texture before you commit to it. Small, sharp, opinionated tools.
We don't do growth loops, we don't take equity, and we don't call a prototype finished because a launch deck said so. If it isn't in someone's hands and being complained about, it isn't done.
Six benches, a rotating cast of residents, one rule: every experiment ships with its lab notes public. If we got something wrong you should be able to read exactly how.
Sweep running · six channels
Every card has a lab note folded in behind the tab. It is where the failures live, and they are the more useful half.
OC-01 · running
A contact colour reader that samples pigment off a real sheet of paper and reports it as a value you can mix again tomorrow.
Read accuracy 78% on uncoated stock
The first sensor loved gloss paper and hated everything a painter actually uses. We swapped the light source twice and moved the aperture down to 4 mm. The miss now reads as neutral instead of wrong, which is the whole trick.
Still broken on newsprint. We know.
OC-02 · running
A soft-light rig that measures the falloff of a desk lamp so you can screen print the shadow it actually throws, not the one in your head.
Falloff match 91% across 400–900 nm
Nine hours of this was us learning that a lamp tilted 3° changes the penumbra by a fifth. Now the rig refuses to calibrate unless the arm is level, which has annoyed exactly one person and that person was us.
OC-03 · running
A spatial grid that snaps to the beat, for animators who keep re-timing the same twelve frames by hand and pretending it was intentional.
Snap tolerance ±4 ms at 120 fps
Six revisions were all the same bug: our clock assumed a steady display. It never is. We now sample the refresh and give up gracefully rather than lying about where the beat is.
Resident Hannah called this "snapping like a good editor". We took it as a specification.
OC-04 · running
A haptic texture brush. Drag it across the page and you feel the surface you are drawing — grit, weave, the tooth of the paper — before the ink goes down.
Texture bands 18, continuous
Everything felt like sand for the first month. Turns out the driver had a 40 ms rise time and we were asking it to be a fingertip. Fixed the rise, fixed the entire product.
Field test with six illustrators. Four asked to keep it. Two said it was "too honest". Both useful.
OC-05 · hold
A variable type engine that breathes a headline to fit its measure instead of truncating it. Currently on hold while we argue about descenders.
Optics range 0.92–1.18, unsettled
Every setting that made the caps look right made the descenders look seasick. We spent a fortnight proving that this is not a bug you can letter-space away, and then we stopped.
Parked, not abandoned. The notes stay public either way.
OC-06 · prototyping
A small mechanical arm that takes its light cues from the scene you are animating, so a turntable renders with the same shadows you saw on the desk.
Joint backlash 0.4° after re-cut
Nine revisions is mostly one bush we machined wrong. The replacement cost more than the original arm. Worth it: the shadow is finally the shadow.
Drawing above is to scale. The arm is 340 mm.
Visual mode
Continuous motion. Satellites accelerate as your cursor comes near them.
Telemetry
The orbital field is not decoration. It is a readout. Change how it draws and you change what it is telling you.
Log · 05 Oct → 02 Nov 2031
Short, dated, unedited by marketing. If it went badly, it goes in here.
02 Nov 2031
The interesting result was a comment in the code we left behind, about how the first approach was "obviously right" and obviously wrong. Both things at once. It now heads every repo we start.
— Ravi, bench 4
19 Oct 2031
Three residents asked for one this quarter. We built it in a day. It has become the most argued-about object in the building, which we take as a good sign about most objects.
— Ilse, director
05 Oct 2031
Tidepool took five months and most of it was calibration, not engineering. Every illustrator we tested with described the same missing thing: the paper. We had been ignoring the paper.
— Noor, bench 3
Channel open · four a year
We take four external collaborations a year. Makers, illustrators, typographers, people with a workshop and an argument. We ask that you can come to the bench at least once.