Review · August 26, 2026
SCIM Use Cases (draft-correia-scim-use-cases-03)
draft-correia-scim-use-cases-03 by Paulo Jorge Correia (Cisco) and Pamela Dingle (Microsoft).
Where this sits in the lineage
This is the October 2024 individual submission. It expired in April 2025 and has since been adopted, so it's the direct ancestor of the working group draft rather than something you'd review on its own.
The chain:
draft-correia-scimusecases-00(July 2023)draft-correia-scim-use-cases-00 through -03 (October 2024), the document under reviewdraft-ietf-scim-use-cases-reloaded, the adopted WG draft, at -02 since January 2026
Here's the thing worth knowing before you read further: the WG draft is -03 with two new boilerplate sections at the front, RO/RA renamed to SRO/SROA, and a round of typo fixes in the section headings. The substance is the same document. If you review -03 you are effectively reviewing the adopted draft, which is unusual and makes this revision worth more attention than most expired individual submissions.
The WG draft expired on 23 July 2026 and hasn't been refreshed.
The problem it's solving
SCIM shipped in 2015 as three documents. RFC 7643 (schema) and RFC 7644 (protocol) get implemented. RFC 7642, the "definitions, overview, concepts, and requirements" document, described what the working group expected people to build in 2015.
Ten years of deployments went differently. Real topologies are multi-hop: an HR system feeds an IDaaS, the IDaaS feeds forty SaaS apps, and some of those apps are authoritative for attributes they write back. RFC 7644 gives you exactly two names for participants, "SCIM client" and "SCIM service provider," and in a three-hop chain the middle box is both at once. There's no vocabulary for who is authoritative for what.
That's the gap. This draft carries RFC 7642's exact title and a lightly edited version of its abstract, so the intent is explicit: replace it.
How it fits in
Squarely in charter. The SCIM charter opens its scope 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 delivers the first three. Deletion is absent, which I'll come back to.
It defines no wire format. No new endpoints, no new attributes, no new headers, no IANA registrations. Six RFCs and one draft are cited, and every SCIM reference is informative. RFC 2119 is the only normative reference in the document.
That combination doesn't hold up. Either this is an Informational document that describes patterns, in which case drop the RFC 2119 boilerplate, or it imposes requirements, in which case RFC 7643 and RFC 7644 belong in the normative section. Right now the 2119 section supports exactly one normative keyword in the whole draft, a "MAY be a subset" in Section 3.1.5.5 that was copied out of the SCIM events text.
What the -03 rewrite actually did
The -00 revision organized Section 4 as eight cumulative use cases with headings up to 44 words long. The -03 threw that out and reorganized around the roles. This is a large improvement and it's the reason the document got adopted.
A role vocabulary. Six roles layered on SCIM's client/server split: Resource Object, Resource Attribute, Resource Creator, Resource Updater, Resource Manager, Resource Subscriber, External Resource Creator. One box holds several at once. The HR system is RC/RU, the IDaaS is RM, the SaaS app is RS. This is the useful part.
A trigger taxonomy. Four ways an interaction starts: scheduled interval, event, admin or end-user action, and SSO. The SSO trigger covers JIT provisioning filling in attributes a federation assertion couldn't carry.
Five action patterns. Client active push, client active pull, active dynamic query, domain replication mode, and co-ordinated provisioning. The last two are lifted from what is now RFC 9967 (SCIM Profile for Security Event Tokens), which is the right thing to reference and the wrong citation to use, since the draft still points at draft-ietf-scim-events.
Two use case tiers. Section 4.1 works through the orchestrator roles in single-tenant and multi-tenant variants. Section 4.2 gives seven concrete implementations with short names: Partner Device Registry, Enterprise "Last Mile" Applications, Reconciliations. Naming the concrete cases instead of describing them in the heading is the single biggest readability win over -00.
Also gone from -03: the claim that mutability functions as a distributed write-authority marker that prevents loops. That was the worst technical problem in -00 and the authors removed it rather than trying to specify it. Good call.
Interoperability
For a use cases document, interoperable implementation isn't the bar. Four things in here will still cause problems downstream.
A synchronization mechanism that doesn't exist
Section 3.1.5.3 describes "Active Dynamic Query": the server returns a "Dynamic Query (DQ) token" with a response, the client presents it on the next GET, and gets back only what changed including tombstones for deletions.
Nothing like this exists in SCIM. The phrase "Dynamic Query" appears in no SCIM RFC. RFC 9865 added cursor-based pagination in October 2025, which pages through a result set rather than detecting change since a watermark. The charter lists "Incremental approach to synchronization" as work still to be done.
Describing the use case is fine and chartered. Describing a token-based delta protocol in the present tense, with a numbered figure, as one of five SCIM actions is not. This is the kind of text that shows up in an RFP requirement two years later.
The WG draft fixed this. Section 5.1.5.3 renamed it to "Client Active Delta Query" and appended: "A token-based delta query mechanism is not defined by [RFC7644]; any such mechanism described here is illustrative and non-normative and would require standardization in a future specification." That's exactly right. The figure caption still says "Dynamic Query" and the section title picked up a stray "A" prefix, but the substance is resolved.
Multi-tenancy is a first-class topic with no tenant isolation discussion
Six of the twenty-eight figures are multi-tenant. Figures 11, 12, 15, 16, 21, and 22 all show one provisioning domain fanning out to or pulling from per-tenant endpoints. That's the deployment shape that actually matters commercially, and putting it in the document is the right decision.
The Security Considerations is one paragraph. It covers transport confidentiality, PII crossing jurisdictions, and a pointer to RFC 7643 §9 and RFC 7644 §7. It says nothing about tenant isolation, credential scoping per tenant, or what happens when a SCIM client's credential is over-scoped across tenants.
The only scoping statement in the entire document is buried in Section 4.2.3: "SCIM clients should only be able to access the devices that they manage." Lowercase should, inside a use case, in a document whose Security Considerations never picks up the thread. If the document is going to introduce multi-tenant fan-out as a core pattern, the security section needs a paragraph on isolation.
Credentials as resource attributes
Section 4.2.4 is titled "Provide Credetials to manage Device." The Device Manager pushes credentials to a client application acting as a gateway, which relays them to the device through a commissioning tool over a non-SCIM hop.
Moving device credentials through a SCIM resource attribute, across an intermediary, with a protocol handoff at the end, deserves more than four sentences and a figure. The Security Considerations doesn't mention it. Either cut the case or say what the requirements are on the gateway.
Mechanisms that could be removed
Sections 4.1.3.2.1, 4.1.3.2.2, and 4.1.3.2.3 are one sentence each, all variants of "Same information as Single-Tenant Resource Manager that is the SCIM Server but the Provision domain has multiple Tenants." Avoiding the copy-paste is the right instinct. Three numbered headings with a single cross-reference sentence each is not the way to do it. Fold them into one paragraph under 4.1.3.2.
External Resource Creator is defined in Section 3.1.3.5, listed in the acronyms, and then never used. Figures 17 through 20 have a non-SCIM upstream source, which is exactly what ERC describes, and they label it "Non SCIM." Either use the role in the figures or drop it. IaaS is in the acronym list and appears nowhere else in the document.
Deletion
Still missing. The charter names it twice, once as a use case category and once as "Handling Deletes in SCIM Servers that don't allow Deletes (Soft Deletes)."
Deprovisioning is the half of SCIM that produces security incidents. A document that walks through provisioning topologies across twenty-eight figures and never covers what happens when someone is terminated has a hole in the middle of it. The word "delete" appears in this draft only inside HTTP verb lists.
Writing quality
Better than -00, still not close to ready.
Copy editing first. "Resrouce" appears five times, including inside figure captions. "Costumer," "Manufactor," "Credetials," "Comissioner," "diferent," and "iether" account for eleven more. "tipically" twice, "aplication" twice. Section headings carry some of these, so they propagate into the table of contents: "Provide Credetials to manage Device" and "Client Applications gets directory Services" are both TOC entries. The WG draft fixed the headings and missed "Costumer" and "diferent," which are still in the adopted text.
The terminology section is broken. Section 2 renders as running prose: "Here is a list of acronyms and abbreviations used in this document: * CRUD: Create, Read, Update, Delete * ERC: External Resource Creator * IaaS: Infrastructure as a Service * ..." The markdown bullets didn't survive conversion to xml2rfc. Twenty acronyms in one unbroken paragraph. This bug is still in the WG draft.
Four of the ASCII figures have a rendering glitch. Figures 17, 18, 19, and 20 each contain a line reading |! !| where box borders should be.
There's a technical error in Section 3.1.4.4. On the SSO trigger: "This is particularly useful with protocols like SAML, which are limited by the number of characters in the URL." SAML isn't limited by URL length. The HTTP Redirect binding is, and nobody uses Redirect binding for assertions carrying user attributes; POST binding has no such limit. The real reason SCIM completes JIT-provisioned records is that IdPs don't put every attribute in an assertion, which the same section already says correctly one sentence earlier. Cut the URL clause. It survives into the WG draft at line 478.
Diagram conventions shift between sections with no note. In Section 3.1.5 the figures number both the request and the response, so an arrow is an HTTP message: Figure 5 shows Server <--- Client for the GET and Server ---> Client for the response. In Section 4 the figures number one arrow and it points the way the data goes, so Figure 10 shows Server ---> Client labeled "SCIM Client performs Active/Delta Pull." Both conventions are defensible. Using both in one document without saying so means a reader has to reverse-engineer which one applies to each figure.
Section 7, Acknowledgements, is a heading with no body.
On length: 2016 lines, and Section 3 does most of the work in the first 500. Section 4.1's four-way expansion across single-tenant and multi-tenant times client and server produces eight subsections where the differences between several of them are one sentence. The concrete cases in 4.2 are the strongest part of Section 4 and they're the shortest.
One inherited oddity: the title promises "Requirements" and there is no requirements content. RFC 7642 has the same title and the same gap, so this isn't new, but a revision is a good moment to either write the section or change the title.
The authors
Paulo Jorge Correia (Cisco, [email protected]) has zero published RFCs and no active Internet-Drafts. Datatracker shows three expired drafts, all of them this document under its various names. He's a new IETF author and the copy quality is consistent with that.
Worth crediting the follow-through. He took this from a rough July 2023 submission through four individual revisions to working group adoption. That's more persistence than most first drafts get, and the -00 to -03 reorganization is a real piece of editorial work.
Pamela Dingle (Microsoft, [email protected]) also shows zero RFCs on Datatracker, which undersells her badly. She runs identity standards at Microsoft and most of her published specs live at the OpenID Foundation. She has three active IETF drafts: draft-ietf-spice-glue-id, draft-wzdk-scim-agent-resource, and draft-mora-oauth-entity-profiles. She's writing a lot right now.
The pairing shows in the document. The role model and the multi-tenant framing come from someone who has watched a lot of real deployments. The prose comes from a first-time draft author. That's a productive combination and -03 is evidence of it working.
What to do with this
Don't file comments against -03. It's expired and adopted. Everything below applies to draft-ietf-scim-use-cases-reloaded, which inherits all of it.
Highest value first:
- Deletion. Chartered twice by name, absent from the document. This is the substantive gap, not a nit.
- Tenant isolation in Security Considerations. The document made multi-tenancy a core pattern. The security section is a paragraph inherited from 2015 that predates that decision.
- Stale references.
draft-ietf-scim-eventsis RFC 9967. The device schema this draft leans on across seven use cases is RFC 9944 and isn't cited at all, despite "device" appearing 44 times. - Normative status. Pick one: drop RFC 2119 and lowercase the single MAY, or move RFC 7643 and RFC 7644 to normative.
- The broken terminology paragraph and the four glitched figures. Mechanical, and they're in the adopted draft.
- The SAML URL-length claim. One clause, factually wrong, still live.
And the -02 needs a refresh. It expired last month, and for a document the charter names as a deliverable that's the thing blocking everything else.