The Scope Creep Automated reviews of new AI-related working group submissions

Review · August 26, 2026

SCIM Use Cases

draft-correia-scimusecases -00View on Datatracker →

draft-correia-scimusecases-00 by Paulo Jorge Correia (Cisco) and Pamela Dingle (Microsoft).

Start here: this revision is three years dead

This is the July 2023 individual submission. It expired in February 2024 and has been replaced twice since.

The lineage:

  • draft-correia-scimusecases-00 (July 2023, the document under review)
  • draft-correia-scim-use-cases through -03 (October 2024), a substantial rewrite
  • draft-ietf-scim-use-cases-reloaded, the adopted working group draft, now at -02 (January 2026)

If you are reviewing this to give the authors feedback, review the WG draft instead. Almost every criticism below was already addressed between -00 and -03. The -00 is 1288 lines; the current WG draft is 2016 lines with a completely different Section 4.

One thing worth flagging: the WG draft's -02 expired on 23 July 2026 and hasn't been refreshed. For a document the charter names as a deliverable, that's the actionable item, not anything in this revision.

The rest of this review covers -00 on its merits, since that's what was asked.

The problem it's solving

SCIM shipped in 2015 as three documents. Two of them, RFC 7643 (schema) and RFC 7644 (protocol), get implemented. The third, RFC 7642, was the "definitions, overview, concepts, and requirements" document. It described what the working group thought people would build.

Ten years of deployments turned out differently. Real SCIM topologies are multi-hop: an HR system feeds an IDaaS, the IDaaS feeds forty SaaS apps, and some of those apps write attributes back. RFC 7642 doesn't describe any of that. It also has no vocabulary for it, because RFC 7644 only gives you two names, "SCIM client" and "SCIM service provider," and in a three-hop chain the middle box is both.

That's the gap. This draft is trying to give the working group a way to talk about who is authoritative for what, in a topology where the same box is a client on one side and a server on the other.

How it fits in

Squarely in charter. The SCIM WG charter opens its scope of work with "Revision of RFC7642 that will: Focus on Use cases and implementation patterns, Pull vs. Push based use cases, Events and signals use cases, Deletion use cases." This draft covers the first three. It skips deletion entirely.

It's Informational and it defines no wire format. No new endpoints, no new attributes, no new headers. It depends on RFC 7643 and RFC 7644 as informative references only, which is unusual for a document whose entire content is about them.

Two of its work-in-progress references have since become RFCs: the SCIM device model is now RFC 9944 and the SCIM profile for Security Event Tokens is now RFC 9967. Anyone updating this text should cite the RFCs.

What it actually defines

Three things.

A role vocabulary. Seven acronyms layered on top of SCIM's client/server split: Resource Object, Resource Attribute, Resource Creator, Resource Updater, Resource Manager, Resource Subscriber, External Resource Creator. One box can hold several roles at once. The HR system is RC/RU, the IDaaS is RM, the SaaS app is RS, and the LDAP-to-proprietary-REST bridge is an ERC. This is the useful part of the draft, and it's the part the working group adopted.

A trigger taxonomy. Four ways a SCIM interaction starts: scheduled interval, asynchronous event, admin or end-user action in a console, and SSO. The SSO trigger is the interesting one: a SAML assertion does JIT provisioning but can't carry every attribute, so SCIM fills in the rest afterward. The event trigger points at Security Event Tokens and correctly notes that SETs state facts about changes that already happened rather than commanding the receiver to do something.

Topology diagrams. Section 3.1.4 walks through push and pull patterns with ASCII figures, then Section 4 stacks eight use cases from /Me up to multi-IdM consolidation after a merger.

Interoperability

For an Informational use cases document, interoperable implementation isn't the bar. But two things in here will cause interop problems if they survive into the WG draft.

The mutability claim

Section 3.1.2.1 says authority over an attribute is expressed through mutability: the authoritative source holds readWrite, everyone else gets readOnly, only one readWrite per attribute is allowed, and this is passed RM-to-RM. The draft says "with this mechanism we can prevent loops."

That's not what mutability does. In RFC 7643 it's a server-side declaration in a schema definition telling clients what they're allowed to write to that server. It is not a token that travels with the data, and it says nothing about who is authoritative in some other domain. Repurposing it as a distributed write-authority marker is a new mechanism, not a restatement of an existing one, and the draft doesn't specify how the value propagates or how a receiver would even know a hop occurred.

Then the same section says "This document is not going to give implementation recommendation how the different orchestrators roles should be developed to avoid such loops." So it claims loop prevention and declines to specify it, in the same paragraph. Pick one. Either describe the loop problem and say it's out of scope, or specify the mechanism.

The draft also attributes mutability to RFC 7644. It's in RFC 7643.

"The client will always do a GET first"

All four push diagrams say step 1 is "Before creating/updating a RO/RA the SCIM client will always do a HTTP GET to get current information from the SCIM Server."

That's not in RFC 7644. It's one implementation pattern, and a fairly expensive one. Stating it as a property of the protocol in four consecutive figures is the kind of thing that ends up in an RFP checklist two years later. If the point is that clients need current server state before a PUT, say that and explain why. If the point is that PATCH exists specifically so you don't have to do this, say that instead.

Mechanisms that could be removed

Section 4's eight use cases are cumulative, and by 4.5 the additions are mostly organizational commentary rather than new mechanics. "Typically we will see this use case in mid to large organization where resources were organized in a non standardize platform" appears, nearly verbatim, in four sections. Cases 4.5 through 4.8 add one genuinely new thing each: multi-source consolidation with precedence rules, the subscriber becoming authoritative, write-back to the creator, and multiple resource managers. Those four ideas don't need four cumulative case descriptions carrying forward all the prior text.

Section 3.1.4 has the same redundancy. The four-step push sequence is printed identically under 3.1.4.1.1, 3.1.4.1.2, 3.1.4.1.3, and 3.1.4.1.4. State it once, then show what changes.

The /Me case in 4.1 is already fully specified in RFC 7644 §3.11. It's here as the simplest case in a progression, which is fair, but it doesn't need its own authorization discussion.

Deletion is missing. The charter asks for it by name, including soft deletes. Deprovisioning is the half of SCIM that actually causes security incidents, and a use cases document that covers provisioning topologies in eight escalating cases and never covers "what happens when someone is terminated" has a hole in the middle of it.

Writing quality

Rough. This reads like a first draft that went out to meet a submission deadline.

Grammar and agreement problems are on most pages. "Is and object that is going to be manipulated." "Is one element of the Resource Object." "The client in this case has no responsibilities on the management of Resources." "there can only be on readWrite for a specific RA." "will do al the Resource management." "list list of contacts." A cron job is spelled "chron."

The introduction has a real technical error, not just a typo. It says SCIM "provides provisioning and de-provisioning of resources in a separate context from authentication (aka just-in-time provisioning)." JIT provisioning is provisioning that happens during authentication. The parenthetical contradicts the sentence it's attached to. The draft's own Section 3.1.3.4 gets this right.

Reference hygiene is inconsistent. Cross-references are written [RFC 7643] with a space, which doesn't resolve. RFC 8417, RFC 9110, and RFC 9112 are cited inline but missing from the reference section. ABFAB is named in the introduction with no citation. The two draft references are formatted as bracketed English titles rather than draft names. RFC 2119 is the only normative reference, but the Security Considerations use MUST in an Informational document.

Section 6 (IANA) and Section 7.3 (Acknowledgements) are headings with no body.

The Section 4 headings are the most visible symptom. Section 4.8 is titled "Multiple IdM doing CRUD operations on SaaS applications, and Objects coming from external SCIM and non SCIM sources including the IdM itself, where some Resource Attributes come from SaaS application, and are updated in the SCIM object creator." That's 44 words, and it consumes 14 lines of the table of contents. The -03 revision fixed this by reorganizing Section 4 around the orchestrator roles and giving the concrete cases short names like "Partner Device Registry" and "Reconciliations." That reorganization is a clear improvement and it's the model to keep.

On length: the concepts here would fit comfortably in half the text. Section 3 is stronger than Section 4 and does most of the work.

The authors

Paulo Jorge Correia (Cisco, [email protected]) has zero published RFCs and no other active Internet-Drafts. Datatracker shows this use cases document as essentially his only IETF work. He's a new IETF author, and the writing quality of -00 is consistent with that. Worth noting he stuck with it: he took the document through three individual revisions and it got adopted, which is more follow-through than most first submissions get.

Pamela Dingle (Microsoft, [email protected]) also shows zero RFCs on Datatracker, but that number badly undersells her. She runs identity standards at Microsoft and chairs work at the OpenID Foundation, where most of her published specs live. She has three active IETF drafts right now: draft-ietf-spice-glue-id, draft-wzdk-scim-agent-resource, and draft-mora-oauth-entity-profiles. She's writing a lot at the moment.

The pairing explains the document's shape: the architectural framing and the role model reflect someone who has watched a lot of real deployments, while the prose reflects a first-time draft author. That's a normal and productive combination in the IETF, and the -03 revision shows it working.

What to do with this

Don't review -00. It's expired, replaced, and the criticisms have mostly been addressed upstream.

The two things worth raising against the current WG draft:

  • The mutability-as-authority mechanism needs to be either specified properly or dropped. Right now it claims to prevent loops without saying how, and it assigns a meaning to mutability that RFC 7643 doesn't support. This one matters because other SCIM drafts are starting to reference the orchestrator role model, and if they inherit the authority semantics they'll inherit the ambiguity.
  • Deletion is a chartered deliverable and it isn't there.

And the -02 needs a refresh. It expired last month.