spam_scanners on policy_banks

Hello,

I am trying to have two different policy_banks, each with a different 
set of spam_scanners:
* One policy would use spamassassin
* The other would use SpamdClient (not spamassassin here but another 
scanner with spamassassin protocol).

Relevant settings on amavisd.conf (just one policy_bank for brevity):

----------
$interface_policy{'10024'} = 'POL1';

$policy_bank{'POL1'} = {
   forward_method => 'smtp:[*]:10025',
   notify_method  => $forward_method,
   spam_scanners => [
      ['SpamdClient',  'Amavis::SpamControl::SpamdClient',
        mail_body_size_limit => 65000, score_factor => 1.0,
      ],
   ],
};

@spam_scanners = (
   ['SpamAssassin', 'Amavis::SpamControl::SpamAssassin' ],
   ['SpamdClient',  'Amavis::SpamControl::SpamdClient',
     mail_body_size_limit => 65000, score_factor => 1.0,
   ],
);
----------

I would expect messages with policy POL1 to be scanned only through 
SpamdClient; but what really happens is that both Spamassassin and 
SpamdClient are called, and their scores added (looks like global 
setting is applied).

Is there anything obviously wrong in my config?

Using amavis-2.12.1 from EPEL on Centos 8 Stream.

Thank you,
Javier.