Fixes

• Closed a security hole in the AI argument builder and in Reset. Both could be invoked against an argument the caller had no edit access to — including published ones — which would have overwritten its claims, premises and citations. Both now refuse unless you can actually edit that argument.