I saw your comment in zanzibar paper > While not given a great detail, Leopard is known to be specialized only to nested group membership, which is often deep and also easy to compute, having no computation besides membership. So I assume, theoretically, as long as I have relation

member

(or any other name with the same sematics) it is possible to use Leopard even when different resource types are linked.

the problem I am trying to solve is the following. https://play.authzed.com/s/bWNmGRqUHTVr/schema Let's say, I have users. Each user may have one or more different keys to a global organization. Permission to access organization is computed based on the keys.

check organization:myOrg#full_access@user:1

leads to all

level_two

keys being fetched from db and then dispatched for key-user relation check. That is a known issue for wide graphs. I am trying to understand whether Leopard would possibly help here, given that no `group`/`member` are involved here

also I am a bit confused with tiger cache proposal since 1) at https://zanzibar.tech/ it is mentioned that

In SpiceDB we are planning to generalize this index as the Tiger cache, for any configured relation(s) or permission(s). Read the proposal.

2) and then in propsal you have

NOTE: These APIs, if traversing deeply nested object types, would also benefit from a Leopard cache, when we implement that.

does that mean tiger is not a leopard replacement?

Wide relations aren't super problematic; its the deeply recursive ones where Leopard helps