Zed --explain flag uses cached results

[winstaan74]

What platforms are affected?

macos, others

What architectures are affected?

others

What SpiceDB version are you using?

v1.35.3

Steps to Reproduce

In Zed, the --explain flag makes use of cached results, even with the --consistency-full flag set. This makes it hard to get a full trace of how a permission decision was calculated. For example, running the same permission check twice gives differed answers each time -

❯ zed --insecure permission check --explain timesheet:1 read_timesheet user:123 --consistency-full                                              
4:00PM INF debugging requested on check
true
✓ timesheet:1 read_timesheet (2.347208ms)
└── ✓ engagement:3 read_timesheet (1.806125ms)
    ├── ⨉ engagement:3 supplier_for_attribute (994.125µs)
    ├── ⨉ engagement:3 manages_attribute (1.536667ms)
    └── ✓ engagement:3 self_attribute (1.711417ms)
        └── ✓ person:2 user (1.196667ms)
            └── user:123 

❯ zed --insecure permission check --explain timesheet:1 read_timesheet user:123 --consistency-full                                              
4:00PM INF debugging requested on check
true
✓ timesheet:1 read_timesheet (cached)
└── user:123 

I see the same behaviour when using the grpc api from a java client with the 'debug' flag set. My desired behaviour is an explanation of the permission checking path that can be displayed to a user.

Expected Result

A full permissions check trace each time.

Actual Result

The explanation for the second permission check is minimal.