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

Review · November 16, 2026

SCIM Events for Merged and Unmerged Users

draft-vivek-scim-user-supersession -00View on Datatracker →

The Problem

A SCIM service provider finds that two User resources are the same person. It keeps one and retires the other.

Today the only signal a downstream receiver gets is a delete event for the retired User. Nothing says which User replaced it. Anything the receiver keyed to the old id is orphaned.

This draft adds an event that says "A was retired in favor of B". It adds a second event that says "that merge was wrong, undo it".

How It Fits In

This is an individual draft at -00. It has a 15 minute slot on the SCIM agenda at IETF 127, in the same session as the rechartering discussion.

It extends RFC 9967, the SCIM profile for Security Event Tokens. It defines two new event URIs in the prov class and nothing else. No new schema, ResourceType, endpoint, or subject identifier format (Section 10). It reuses the RFC 9967 envelope, the scim subject identifier format, and delivery model.

Normative dependencies are small: RFC 7643, RFC 7644, RFC 8417, RFC 9493, and RFC 9967. All are published.

Does it fit the charter? Loosely. The SCIM charter covers coordinating resources between domains using RFC 8417, and it lists deletion use cases. It does not mention merging or duplicate accounts. The adopted use cases draft, draft-ietf-scim-use-cases-reloaded, does not describe a merge either. The charter also talks about extensions "that have found broad adoption". The draft lists no implementations.

The RFC 9967 event registry is Expert Review (RFC 9967 Section 7.3). These two URIs could be registered without a working group document at all.

Summary

There are two parties, the event publisher and the event receiver. Both Users must belong to the same SCIM authority. Cross-provider merges are out of scope (Section 4).

Supersede (Section 6.2): urn:ietf:params:scim:event:prov:supersede. The top-level sub_id is the retired User. The payload has two required fields, a transitionId and a successor subject identifier.

Correction (Section 6.3): urn:ietf:params:scim:event:prov:supersession-correction. The sub_id is the User that now represents the person who was wrongly merged away. That can be the original User, or a new one if the original id is gone. The payload has four required fields: its own transitionId, reverses, superseded, and previousSuccessor.

The events carry the relationship only. Attribute changes on the surviving User go out as ordinary RFC 9967 patch or put events (Section 5). The draft is firm that a supersede event is not an instruction to copy attributes, move credentials, grant entitlements, or transfer consent (Sections 6.2 and 8).

Interoperability

Significant interoperability concerns

The wire format is tight. The receiver behavior is almost all local policy. After a supersede event the receiver "MAY update local references" (Section 7). After a correction it MUST stop treating the merge as valid, then reconcile "according to local policy". Two conformant receivers can do completely different things with the same event.

"Retired" is never defined. The superseded User is "retired according to local policy" (Section 3.1). It might be deleted, deactivated, or left alone. The receiver is told not to treat the event as a delete instruction. A receiver that provisions accounts into an application now has two accounts and no guidance on what to do with the old one.

Ordering against the delete event is only a SHOULD. Section 6.2 says the publisher SHOULD send the supersede "before or with the deletion sequence". If the delete arrives first, the receiver has already dropped its mappings. That is the exact failure this draft exists to prevent.

It is unclear whether these events can share a SET with others. Section 6.1 says a SET MUST contain "exactly one of the two event URIs defined by this document". RFC 9967 Section 2.1 allows several event URIs in one SET for the same resource change. Can a supersede and a delete for the same User ride together? The text does not say.

transitionId overlaps with txn. Section 4.1 argues that txn cannot be used because a correction needs to refer to the decision "independently of a particular token delivery". RFC 9967 Section 2.2 already says txn stays the same across retransmissions and across receivers. The draft asks about this itself in Section 11, issue 5. The type and format of transitionId are also not specified.

Late joiners and lost history. A receiver that never saw the original supersede gets a correction it cannot apply. Section 7 tells it not to guess and to resynchronize. There is no way to read current supersession state from a User resource. Section 11, issue 3 raises this as future work.

Chains are not covered. A is merged into B, then B into C, then the first merge is corrected. The draft says a correction invalidates "exactly one" transition and stops there.

Mechanisms that could be removed

  • Section 12 on pre-authentication and pseudonymous profiles. It is informative, defines nothing, and is already listed as a non-goal.
  • superseded and previousSuccessor in the correction payload. A receiver that knows the reverses transition already has both. A receiver that does not know it is told not to act on them.

Writing Quality

The core idea is clear and the scope is narrow. Goals and non-goals are stated up front. The security and privacy sections make real points, such as a forged supersede redirecting account associations (Section 8).

It is repetitive. The rule that attributes are not copied appears in Sections 1, 3.3, 5, 6.2, 7, and 9.

There are -00 rough edges:

  • Section 3.2 has a bullet list with no sentence introducing it.
  • Section 4.1 has a list of MUSTs with no subject.
  • Section 1 points to "Appendix A". The text is in Section 12.
  • The IANA section says "Section Section 6.2".
  • The author block gives one name and no affiliation.

Authors

From the Datatracker profile:

  • Vivek Vinushanth Christopher. No affiliation listed, based in Sri Lanka per the draft. No published RFCs. This is his only draft, so he is a new IETF author.

Questions for the Working Group

So what should the session decide?

  • Is account merge and unmerge a problem the WG wants to own? It is not named in the current charter.
  • Does this need a WG document, or is an Expert Review registration against RFC 9967 enough?
  • Who has implemented a merge signal today, and what do receivers actually do with it?
  • Should txn replace transitionId?
  • Should the supersede-before-delete ordering be a MUST?
  • Should supersession state be readable on the User resource, so receivers do not depend on event history?