Trovvy Research

It is not a department. There is no lab, no grant, no board, and nobody to report to. Trovvy is a pet app that has taken five years of nights and weekends and has no investors, and Trovvy Research is the same person on the nights when the question is a research question rather than a product one.

The name exists for one honest reason: so that a dataset released in 2029 still has something to be attributed to, and so a researcher who finds it in ten years can work out where it came from. It is a citation handle, not a masthead.

It is also not only this. Concentric is the first thing published under the banner and it will not be the last — there is animal-behaviour work sitting unpublished that belongs here too, and given what Trovvy spends its days on, that is the more predictable half of the output. A banner that holds several projects needs to be one name rather than one project’s name, which is the whole argument for having it.

It sounds like a non sequitur and it is not. Both projects are the same question wearing different clothes: how does connection actually work, and who gets extended what.

Trovvy is built around a specific observation about pet care — that it is almost never one person. It is a household, a sitter, a trainer, a vet, a neighbour with a spare key, and the animal at the centre of it who cannot tell any of them what happened yesterday. The app exists to hold that circle together: shared records, roles and permissions, a passport you can hand to somebody who is about to be trusted with something you love.

Concentric asks the same thing about people, and asks it of you. Who is inside your circle, and what changes at the edge of it? Standard personality tests assume the answer is “nothing much” — that you are roughly the same person toward everybody. For a great many people that is plainly false, and no existing instrument has a way to notice.

One project builds the circle. The other tries to measure its shape. It turns out to be the thing I find most interesting about living creatures, in both directions.

The older thread underneath both is narrower and more stubborn: making behaviour countable without flattening it. An ethogram does that for an animal in a field — a fixed set of defined behaviours, coded as they happen, so that what gets analysed later is what actually occurred rather than what the observer remembers. Concentric is trying to do the same job for something much harder to observe, using self-report, which is a considerably worse instrument and the only one available. Both fail in the same direction when they are done badly: the record ends up tidier than the thing it describes.

Animals and people live better when the people around them are actually connected — when the sitter knows about the medication, when the trainer and the household agree on what they are working on, when somebody five years from now can still see the whole story. Trovvy exists to make those connections easy enough that they actually happen.

No investors, no growth team, and every feature so far has come from listening to pet parents and real trainers. That is a slower way to build something and it is the only way I have found that produces a thing worth using.

trovvy.com — free for pet parents, on iOS and Android.

Concentric — trait expression as a function of closeness

A free Big Five instrument that measures something no existing one does: whether your expression of a trait changes with who you are with, and by how much. Human behaviour. In the field now; the dataset is public domain and released weekly.

An ethogram recorder for field observation — built in the mid-2010s, unpublished

An ethogram is a catalogue of precisely defined behaviours that an observer codes against, and it is the backbone of most quantitative animal-behaviour work. The hard part has never been the catalogue; it is capturing against it in the field, in real time, without looking down for so long that you miss the thing you came to record. This is a tool for doing that and getting the log out in a shape something can actually analyse. It has existed for over a decade and has never been released, which is a slightly ridiculous thing to admit in public and is precisely why it is listed here.

A long tail of things that never left the bench

Computer vision, neural-network prediction, sensor electronics. None of it released, most of it built to answer a question rather than to ship, and it is mentioned only so that the pattern is visible rather than as a claim: the reason a pet app ended up running a personality study is that the interesting problem has always been the same one, and the tool is just whatever it takes to get at it.

It means the method is written down before the results, the data is published whether or not it says what I hoped, and every number carries the interval that says how much to trust it. Those are the parts of research that make it worth the name, and they do not require an institution.

• Not IRB-approved. There is no institution to approve it. The consent model, the privacy rules and the k-anonymity policy are published in full so you can judge them yourself rather than trust a stamp.
• Not funded. Nobody is paying for a particular answer, which removes one bias and adds another: this moves at the speed of one person’s spare evenings.
• Not commercial. Concentric sells nothing, gates nothing behind an email, and does not feed Trovvy. The data is CC0. If it were a funnel it would be a remarkably bad one.
• Not certain. The Big Five core here is solid and externally validated. The closeness measure is a genuine bet that may not survive contact with data, and it is labelled provisional everywhere it appears.