ROLLIN Labs is the applied-research arm of The Accessibility Company. We build the instruments and the intelligence that turn accessibility into measured evidence. Our first is ROLLINSIGHT, a protocol that makes a commodity phone a calibrated field instrument. Its verification engine runs in production today.
ROLLIN Labs builds instruments that make accessibility data cheaper to collect, harder to fake, and more valuable to deliver. We take on one frontier project at a time and ship it into the product itself. What we research is narrow and deliberate: how to measure the built world with the hardware people already carry, how to make a cheap sensor’s reading worth trusting, and how the resulting evidence compounds into something no survey could afford.
A modern iPhone already carries a time-of-flight scanner, calibrated microphones, and a camera whose every exposure encodes the light around it. ROLLINSIGHT adds what those sensors lack on their own: proof of where a reading was taken, when, by whom, and whether it holds up against a contradicting report. One guided capture produces four kinds of evidence about a place, each tied to proof it happened there and then.
Door widths and clearances are measured against the phone’s depth sensor at the moment of capture, standing in front of the thing. Every measurement is tagged with how it was taken, so measured readings and estimates stay distinguishable down the line.
Ambient loudness is sampled as a single number. No audio is ever recorded or stored. The microphone works as a meter, the way a photographer uses a light meter. At a Cornwall lounge on a July evening, it read 38.9 dB over 25 seconds and published one word: quiet.
Illuminance is read from the camera’s own exposure, with no added hardware. It publishes as a plain-language level chosen to match the decision a person actually makes.
Photographs are taken live inside the capture flow and understood on the device the moment they are taken. A person reviews each one before it publishes, credited to the contributor.
One of the first field captures in the production record, taken at a Cornwall lounge on a July evening and shown exactly as it stands. A record carries only what was actually captured on that visit, each reading tagged with how it was taken and the state it holds. Nothing is inferred to fill the gaps. Absence is information too.
A second capture from the same visit read 37.0 dB, consistent with this one. Even so, a reading is only marked verified once an independent contributor confirms it.
The protocol treats the phone as an untrusted probe and re-derives every guarantee on a server the phone does not control. A reading earns its place in the record on the server’s terms, with its whole history attached.
Every guarantee is enforced on the server. When two captures disagree about a place, the engine hands the conflict to a person to settle, and records that it did.
Who, when, where, on what device, by what method: the origin of a reading is recorded before any value is ever derived from it. Every effect on the record traces back to a cause we can point to.
A reading publishes as reported by the community and is marked verified once an independent capture confirms it. One enthusiastic account cannot promote its own claim.
No audio is recorded. Photos are of places. The capture flow never asks for your photo library; that capability intentionally does not exist in it. The permission footprint is minimal by design.
Each capture a reviewer approves is, at the same moment, a labeled training example: a photograph, on capable devices a depth map, confirmed feature labels, and a real measurement taken standing in front of the thing, with complete provenance. Over time this becomes the only depth-annotated, human-verified record of entrances, restrooms, and ramps of its kind. At scale it trains a model that recognizes them on its own.
We say it plainly: that model does not exist yet. What we have built is the pipeline that will train it, and the discipline that any new capability has to prove itself against the record before the record acts on it.
The engine is agnostic to what holds the sensor. Adding a new probe, a wearable, a fixed venue device, a partner’s hardware, requires no change to how trust is earned; the design is built to take them without a second brain. Accessibility is the first and most consequential application of a general idea: evidence-grade claims about the built world, made by the cheap sensors already in a hundred million pockets.
You should not have to find out at the door.
Technical reports are shared with press and partners under NDA.
ROLLIN Labs is the applied-research arm of JoinRollin Inc, The Accessibility Company. It builds the measurement instruments and the verification intelligence that turn accessibility from hearsay into evidence, then ships them into the ROLLIN platform.
ROLLINSIGHT is the flagship project from ROLLIN Labs: a provenance-first measurement protocol that turns a commodity iPhone into a calibrated field instrument. One guided capture produces four kinds of evidence about a place, each tied to proof of where and when it was taken.
Readings are captured with the sensors already in a modern phone: the depth scanner measures door widths and clearances, the microphone samples ambient loudness as a level, the camera reads available light from its own exposure, and photos are taken live on site. Every reading is tagged with how it was taken so measured values and estimates stay distinguishable.
Yes. The protocol treats the phone as an untrusted probe and re-derives every guarantee on a server the phone does not control. A reading publishes as reported by the community and is marked verified once an independent capture confirms it.
No. Ambient loudness is sampled as a single number, the way a photographer uses a light meter. No audio is ever recorded or stored, and the capture flow never asks for your photo library.
For research, press, or partnership inquiries, reach the team directly.