ALTER TABLE races in replication on slave side

Setup:

Master: Percona 5.5.24-rel26.0.256
Middle: Percona 5.5.20-rel24.1.217
Slave: MariaDB-10.0.0

On Master:
Say you have the following table:

CREATE TABLE `audit_publication_publication_rtb` (
`audit_publication_publication_rtb_id` int(10) unsigned NOT NULL AUTO_INCREMENT,
`publication_id` int(10) unsigned NOT NULL,
`old_publisher_id` int(10) unsigned DEFAULT NULL,
`old_rtb_id` varchar(255) DEFAULT NULL,
`new_publisher_id` int(10) unsigned DEFAULT NULL,
`new_rtb_id` varchar(255) DEFAULT NULL,
`record_action` int(1) DEFAULT NULL,
`created_at` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (`audit_publication_publication_rtb_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

This table is being inserted into constantly.

I then run the following statement:
ALTER TABLE adfonic.audit_publication_publication_rtb
ADD old_status VARCHAR(32) AFTER old_rtb_id,
ADD new_status VARCHAR(32) AFTER new_rtb_id;

This propagates fine to the Middle server.

However, on the Slave server running MariaDB 10.0.0, this seems to cause replication to break:

Slave_SQL_Running is No
Last_Error is Column 4 of table 'db.audit_publication_publication_rtb' cannot be converted from type 'varchar(96)' to type 'int(10) unsigned', Last_SQL_Error is Column 4 of table 'db.audit_publication_publication_rtb' cannot be converted from type 'varchar(96)' to type 'int(10) unsigned'

At a glance, it looks like the inserts in new format are somehow overtaking the alter table statement in the replication stream.