Every stream carries an owner record, so why does one naming an individual engineer rather than a rota fail within a year?

## What an owner record is for The **owner record** is the team identity every stream carries, as a required field, for as long as the stream exists. It exists to answer one question, asked years later by an operator who was not there when the stream was created: *who is answerable for this?* Every other thing the estate might want to do with a stream that looks unused — challenge it, freeze writes to it, hand it to another team, take it out of service — needs a party on the other end of that question. Without one, the honest options collapse to two: leave the stream running forever, or remove it and discover who cared afterwards. That makes the owner record **operational data, not documentation**. It is read by the periodic idle sweep, by whoever is holding a pager while a cluster runs short of room, and by whoever is accounting for what the estate costs. A field nobody reads can be wrong indefinitely at no cost. This one is read precisely when being wrong is expensive. ## Why a personal name decays A personal name does not record accountability. It records a **link between one engineer and one stream** at one moment, and in most organisations that link has a shorter life than the stream: people change teams, go on leave, take a different role, leave the company. None of those events touches the field. The damaging part is that the failure is silent. The record still contains a plausible human name, so nothing alerts, nothing reports it, and a sweep that lists it looks like it has found an owner. A name in that field is an orphan waiting for a resignation. A second, smaller problem is that one person is one point of contact: even while they are still employed, they may not remember a stream they created three years ago, and nobody else is obliged to answer for it. ## Resolving to something that survives people The useful test of an owner value is not *is it accurate today* but *what does it survive*. | Owner value | Survives | Fails when | |---|---|---| | A named engineer | Nothing beyond that person's tenure in that role | They move team, go on extended leave, or leave | | A free-text team label | Membership churn, if the label is still recognised | The team is renamed, split, or two teams share a similar label | | A rota or shared mailbox tied to an on-call obligation | People joining and leaving; most role changes | The team itself is dissolved, which must then be detectable | | The service that produces the stream | Reorganisations that move teams but keep services | The service is retired, which is the right moment to revisit the stream anyway | None of these is permanent. The point of the bottom two rows is that when they do stop resolving, they stop resolving **loudly** — a dissolved rota address bounces, a retired service is known to be retired — while a stale personal name keeps looking valid. ## Enforced at creation, not curated afterwards There are two ways to have ownership. One makes it a required field on the path that brings a stream into existence, so a stream that exists has an owner by construction. The other keeps a separate list, maintained afterwards, of who owns what. The second is a second list, and two lists drift. The maintained list is accurate about the streams somebody remembered to add to it, which is not the same population as the streams that exist. Enforcement at creation makes existence and ownership the same event, so the record cannot fall behind the estate. Enforcement does not, by itself, make the values good: a required field that every team fills in with the name of the platform team is enforcement working and governance failing. That is why the value is **re-resolved** rather than trusted — an owner record that no longer reaches anybody is a finding in its own right, ranked alongside the streams that show no use. ## What the record has to do when it is read - Resolve to a destination that can actually be reached today, not a string that merely reads like a team. - Fail loudly, not silently, when it stops resolving. - Be repointable in place, so a handover is an edit rather than a reason to create a replacement stream. - Carry exactly one authoritative owner. A courtesy escalation contact alongside it is fine; two authoritative owners means neither answers. - Use the same field, with the same meaning, for every stream in the estate, so the sweep produces one list rather than several. ## Where this stops Ownership here is the obligation and the record that carries it. Which other settings are stamped on a stream at the same moment it is created is a separate subject, and so is everything that happens after the evidence of disuse is in — moving readers to a replacement, writing to both through the change, and the removal itself.