FEP-baf5: Administrator Collection

[julian@activitypub.space]

This is the discussion thread for the draft FEP-baf5: Administrator Collection

> This FEP introduces a mechanism for discovering the administrators of an ActivityPub instance. It extends the "Group Moderator" pattern from [FEP 1b12][1b12] and the "Application Actor" concept from [FEP 2677][2677] by defining an OrderedCollection of administrators referenced from the instance's application actor.

The full draft can be read here.

[silverpill@mitra.social]

1. I find the use of attributedTo here confusing, because normally this property is used to indicate who owns an object or a collection, and its value is an actor. I am aware that attributedTo is used in a similar way in FEP-1b12, but it would be better to introduce a new property (administrators) instead of continuing to abuse attributedTo.
2. You say that your FEP supersedes the same-origin assumption described in FEP-fe34, but I think it describes a reciprocal claim, also described in FEP-fe34: https://codeberg.org/fediverse/fep/src/branch/main/fep/fe34/fep-fe34.md#reciprocal-claims. I suggest clarifying which aspect of FEP-fe34 is being superseded.
3. In the section "Security and Authorization" you say "authenticity" but what actually is being verified is a permission.
4. The entire problem of non-same-actor updates and deletes can be avoided by using different activities. For example, Update can be replaced with an annotation activity. Delete can be replaced with Remove (from thread).
5. FEP links lead to w3id.org site, not directly to FEPs.

[julian@activitypub.space]

> @silverpill@mitra.social said:
>
> 5. FEP links lead to w3id.org site, not directly to FEPs.

Wasn't aware this was a problem? Figured the redirects would be okay.

[julian@activitypub.space]

> @silverpill@mitra.social said:
>
> You say that your FEP supersedes the same-origin assumption described in FEP-fe34, but I think it describes a reciprocal claim, also described in FEP-fe34: https://codeberg.org/fediverse/fep/src/branch/main/fep/fe34/fep-fe34.md#reciprocal-claims. I suggest clarifying which aspect of FEP-fe34 is being superseded.

#2 I suppose supercedes is the incorrect term. It extends fe34, in a way. Would that be acceptable? Definitely not meaning to imply that fe34 is insufficient in any way.

#3 <img class="not-responsive emoji" src="https://activitypub.space/assets/plugins/nodebb-plugin-emoji/emoji/android/2714.png?v=f187f9278b7" title=":heavy_check_mark:" /> okay

[silverpill@mitra.social]

Wasn't aware this was a problem? Figured the redirects would be okay.

The canonical location of a FEP is on Codeberg, but no, it is not a problem.

#2 I suppose supercedes is the incorrect term. It extends fe34, in a way. Would that be acceptable? Definitely not meaning to imply that fe34 is insufficient in any way.

"Extends" is fine, I just think you're describing a reciprocal claim from FEP-fe34, so you could use that term (or maybe FEP-fe34 needs to be updated if "reciprocal claim" is not a good name for this mechanism?)

[julian@activitypub.space]

The reason why attributedTo was chosen is because there is prior art to using that property to represent a collection of moderators. You could make the same argument against 1b12 (that moderators should be the key instead of attributedTo), too.

The argument as to whether a custom property fits better is certainly valid, and worth debating. However, I would want to point out that keeping with prior art has the benefit of making this FEP much easier to adopt by threadiverse implementors.

[julian@activitypub.space]

> @silverpill@mitra.social said:
>
> Extends" is fine, I just think you're describing a reciprocal claim from FEP-fe34, so you could use that term (or maybe FEP-fe34 needs to be updated if "reciprocal claim" is not a good name for this mechanism?)

I don't know if this is truly a reciprocal claim. My understanding from a reading of the relevant section from fe34 suggests a claim of A → B is reciprocal if there is an inverse claim B → A.

fe34 solidifies the concept of same-origin trust (which iirc is only something like a line or two in AP spec?). baf5 builds upon that with an additional opt-in specificity. So baf5 assumes fe34 compatibility, but not the inverse. But we're splitting hairs I think <img class="not-responsive emoji" src="https://activitypub.space/assets/plugins/nodebb-plugin-emoji/emoji/android/1f604.png?v=f187f9278b7" title=":smile:" />

[julian@activitypub.space]

> @silverpill@mitra.social said:
>
> 4. The entire problem of non-same-actor updates and deletes can be avoided by using different activities. For example, Update can be replaced with an annotation activity. Delete can be replaced with Remove (from thread).

While true, this is outside the scope of the FEP. Update/Delete were mentioned in the FEP as they are recognizable, but this sort of explicit authorization* is relevant to any activity type. Offer, Undo, Bite, Zooboomafoo, etc.

* I have to be careful when I use the term "authorization" because if I say it three times @thisismissem will show up and start talking about OAuth2/OIDC again.

[thisismissem@activitypub.space]

> @julian said:
>
> * I have to be careful when I use the term "authorization" because if I say it three times @thisismissem will show up and start talking about OAuth2/OIDC again.

Correction: it's just a single mention that makes me appear, people tend to confuse me with a genie but we're quite different.

(Also I saw other replies here but had an MCAS attack hangover today so didn't have energy to reply. I'll try to reply soon)

[brettm@swarm.coiloptic.org]

@silverpill@mitra.social @julian@activitypub.space @technical-discussion@activitypub.space

I agree with Silverpill 'attributedTo' should not be overloaded by this extra meaning

[silverpill@mitra.social]

My understanding from a reading of the relevant section from fe34 suggests a claim of A → B is reciprocal if there is an inverse claim B → A.

Yes, and in my understanding these claims are:

- This actor is authorized to delete/update this object.
- This object is hosted on the server where this actor is an administrator.

But I don't insist on importing this concept.

I would want to point out that keeping with prior art has the benefit of making this FEP much easier to adopt by threadiverse implementors.

I consider myself a threadiverse implementer too, and I don't really like the idea of dealing with ambiguous properties :)

At the very least, could you add inbox and outbox properties to the Application actor example? https://codeberg.org/devnull/feps/src/branch/instance-admins/fep/baf5/fep-baf5.md#instance-actor-and-application-actor

[thisismissem@activitypub.space]

I agree with Silverpill that attributedTo is a little misplaced here, and a more explicit term would probably be better. e.g., administrators or moderatedBy (this is coming from the T&S taskforce and references a Moderation Group actor, not a collection).

[silverpill@mitra.social]

I am working a FEP which in some ways is related to your proposal:

FEP-5219: Groups and permissions

It introduces a new collection called affiliations, which is intended as a single source of information on who is allowed to do what (within a specific domain).

The FEP is mainly concerned with FEP-1b12 groups, but the affiliations collection could also be attached to Application actors, and be used in a way you described in FEP-baf5: Administrator Collection

[julian@activitypub.space]

@silverpill@mitra.social no objections at first pass, will need to do a deeper read through.

Most importantly though it won't matter unless you get @nutomic@lemmy.ml and @rimu@piefed.social on board.

[silverpill@mitra.social]

That's the plan. I am trying to introduce new features (such as fine-grained permissions) while keeping it as close to existing implementations as possible.

@nutomic @rimu

[thisismissem@activitypub.space]

I really strongly encourage you all to come to the next ActivityPub Trust & Safety task force meeting: https://www.w3.org/events/meetings/29a5bd4f-bb6e-4830-bbf7-7ba290e89afa/20260701T100000/

Let's figure things out together instead of just in isolation.