An operating system for living together

Housing is scarce in every city that matters, and the family structures that used to absorb that scarcity are thinner than they were. The result is a quiet demographic fact: more and more people share territory with people they are not related to — a room in student housing, an apartment in a WG, a floor in asylum accommodation, a building, a subdivision.

Wherever two people share territory, there has to be a common understanding of how living together works. Today that understanding is improvised. The right people living together unlock each other — language practice, shared meals, someone who notices when you're not okay. The wrong people living together are devastating: mental health, physical health, property damage, staff burnout, churn. Conflicts in cohabitation are more frequent than anyone likes to admit, and almost all of them are predictable — a night worker placed with an early riser, a smoker with asthmatics, two people with irreconcilable ideas of "clean."

Why asylum housing first

We built the first vertical for AOZ's asylum housing in Zürich, and that was not a compromise — it was the point. It is the hardest version of the problem: residents under maximum stress, zero choice in roommates, strict privacy obligations, real research literature on refugee housing conflict to build on, and a cost of failure measured in human harm, not just churn. A matching-and-governance system that holds there will hold in student housing, coliving, army barracks or a condo association.

It also forced the right values into the architecture, where marketing can't dilute them:

• Explainability over accuracy theater. Every compatibility score decomposes into named factors with visible weights, in the resident's language. Staff overrule the algorithm freely — and the override is logged, not hidden.
• Data minimalism as product design. No diagnoses, no case details, no religion. Functional needs only ("ground floor", "quiet after 22:00"). Residents feed honest data into matching only if honesty can never be used against them.
• Safety is not majoritarian. A house can vote on kitchen rules; it cannot vote away a minority's safety. The decision-mode config makes that structural, not editorial.

What "operating system" means concretely

Five loops, one product:

1. Placement — who lives where; vacancies minimized subject to compatibility, never the other way around.
2. Matching — config-driven factors (sleep, noise, directional cleanliness, language, risk) scoring person×household, worst-pairing first, blocking conflicts hard-stopped.
3. Governance — layered rules: the organization's floor, the house's own legislation underneath it, per-version acknowledgement so nobody is held to text they never saw. (As of today, the org tier is the literal signed AOZ Hausordnung — see the companion post.)
4. Money & work — shared expenses in integer Rappen, chore fairness, full-history balances.
5. Repair — maintenance to the desk that fixes it, conflicts onto a ladder that starts with talking and ends in agreements with review dates — escalation is a signal, not a punishment.

The architectural bet that makes this a platform rather than a bespoke AOZ tool: verticals are config packs, not forks. The brand is config. The rule catalog is config. The factor weights are config. The contact sheet is config. The same codebase already runs three presentations — AOZ-badged, neutral, and a real Zürich shared flat — and the day another organization arrives, it should bring a seed file, not a fork.

Where this goes — multi-org tenancy, portable resident-owned compatibility profiles, a cross-operator vacancy network — is in the roadmap. But the order matters: first, prove at one or two AOZ locations that placements informed by compatibility measurably cut incidents, relocations and mediation hours. A multibillion-dollar company in this space is built on one boring, verifiable sentence: the conflicts stopped happening.