In short, write model doesn't mean that you cannot read the data from the store. You just cannot read it from the Presentation (UI, client,...). Within write model you can start a transaction, read policies, read anything from the write store that helps you to validate the command, then execute the command and write the state to the store.

So, write store should be the primary store, i.e. all policies should be available in write store. In general, all data should be in primary store first, then replicate to the read stores. All transactions in the write model are strongly consistent. But for the whole application, it's eventually consistent because the client reads from the read stores.

It's similar to database clustering. All transactions must be executed on the primary shard, while queries can be sent to replica shards. DB cluster doesn't prevent you to send the queries to the primary shard, but obviously it would be better for scaling if you route the queries to replicas.