06-reference/research

federated seam ownership frameworks port set spec

2026-09-16·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
organizational-intelligencedata-meshteam-topologiesseam-ownershipdata-contracts

Nobody publishes a framework for owning an unowned seam, but three traditions each own a third of the answer, and the durable form of the claim is "editor of record plus conformance harness," not "owner"

The question

"What established frameworks exist for formalizing 'UNOWNED' seam ownership at federated hub-and-spoke analytics org boundaries, and how do enterprises implement a port-set spec in practice?"

Context: the Organizational Intelligence (OI) roadmap is built on claiming the unowned boundary between a hub platform capability and the federated spoke teams around it. "Seams not spokes." Nothing in the vault grounded that move in named, citable governance literature until now. (Naming note for continuity with older vault docs: this work was formerly called CAF; that name was retired 2026-08-10. Current framing is OI: Org Platform to Org Map, Intelligence Platform to IMA, plus OIP and Pulse.)

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

Answer the load-bearing sub-question first, because it has a sharper answer than the hypothesis it was filed with. The filed hypothesis was that durable claims attach to an artifact others depend on plus a service others want, while territorial claims attach to approval rights. The evidence supports that, and it also refines it in a way that changes what to build. The refinement: the durable/territorial line is not artifact-versus-approval, it is computational-versus-human enforcement of the same rule. The GDR's binding mechanism is a pre-published rule plus an automated validator that any team can run at deployment without the rule's author in the room. That is an approval right in substance and a service in practice, and it survives the author's absence, reorganization, and personal relationships. The identical rule enforced by a human review gate is territorial, because it converts every consumer into a supplicant and its cost is paid in queue time by people who did not choose it. Team Topologies supplies the corroborating measurement: an X-as-a-Service claim is scored by "service adoption and user satisfaction," not by how many interfaces you defined. A seam claim that cannot be measured by voluntary adoption is a toll booth. Boundary object theory supplies the mechanism for why: an object that requires its author to adjudicate every local case is not plastic enough to be a boundary object, it is a bottleneck, and organizations route around bottlenecks.

The second finding is a live correction to the roadmap framing, not a footnote. The vault currently carries "whoever writes the spec owns the seam" as if it were received wisdom. It is not in any of the literature checked. Every named framework that addresses a cross-boundary object assigns it to a plural body: federated computational governance in data mesh, a two-name Implementation Steward pair in the GDR, an Architecture Board in the TOGAF tradition (unverified this pass). This is not bad news. It is a better position than sole ownership, because sole ownership of a boundary between five funded spokes is the exact posture that reads as territorial and invites a coalition against it. The role that is both durable and defensible is convener and editor of record of a federated spec, plus sole author of the conformance harness. The convener sets agenda and sequencing. The editor of record holds the pen and resolves comments into versioned supersessions, which [[fabric-spec-v0]] already commits to ("written comments resolve into v0.1 as superseding sections, never silent edits"). The harness author holds the thing everyone actually runs. Nobody has to surrender a spoke for any of that, which was the original design intent.

What to produce, concretely. Not one large spec document. A port decision record series modeled directly on the public GDR structure, one short record per port, each carrying five named sections: Context · Decision · Consequences and accepted trade-offs · Implementation steward (two names, the consuming spoke owner and the implementing team) · Where policy becomes computational (the exact lint or gate script, by path, that enforces this record). Version each record with a stated breaking-change rule borrowed from data-product practice: breaking change bumps major, everything else minor or patch. The existing v0 draft is the raw material; splitting it into per-port records is what makes it adoptable a piece at a time rather than ratifiable only as a whole. Ship alongside it the one asset published practice consistently under-delivers and the existing draft already half-specifies: a runnable conformance harness that any spoke can point at its own adapter to get a pass/fail against §7. The document is the territorial half of the claim. The harness is the durable half. If only one of the two ships, ship the harness.

The failure mode that makes spoke teams resent it, named in order of likelihood. First, the spec becomes a gate the author personally staffs — every conformance question routes to one inbox, and the seam owner becomes the slowest step in five roadmaps. The mitigation is mechanical: any rule that cannot be checked by a script does not go in a record, it goes in a non-normative appendix. Second, neutral-referee positioning, owning nobody's product but everybody's rules. No framework surveyed supports it; Team Topologies in particular says the only stable shape is that one side becomes a platform with a product interface. The hub must be shipped as a product with consumers and adoption numbers, not administered as a rulebook. Third, conformance defined as similarity to the reference implementation — spokes will read "be conformant" as "be like the incumbent's stack," and they will be right unless proven otherwise. The v0 vocabulary-mapping table is the right instinct; the honest completion is to land a second conformant adapter authored by a different team before declaring v1, and to cite that team's work as the proof. Fourth, SLO asymmetry: publishing what spokes owe the hub without publishing what the hub owes them. Data mesh puts SLOs on the output port, the publisher's obligation, deliberately. A port set that specifies only consumer obligations is the tell that a seam claim is territorial, and experienced platform consumers read that tell fast.

Why this is in the vault

It grounds the OI "seams not spokes" roadmap thesis in named, citable literature for the first time, and it makes two changes to the position: retire "whoever writes the spec owns the seam" as an unsupported coinage in favor of "convener, editor of record, and harness author," and restructure the [[fabric-spec-v0]] draft into GDR-shaped per-port records with a computational-enforcement section so each port can be adopted independently. It also gives the founder the exact four-item resentment checklist to run against the spec before circulating it to spoke owners.

Open follow-ups

Related

Sources

Vault

Web (fetched this pass)

Web (search-result summaries only, not fetched — lower confidence)

Named but NOT verified this pass