-Condor 7.8.4 contains a bug that causes the negotiator daemon to write malformed entries in its =Accountantnew.log=. When the negotiator reads the =Accountantnew.log= on startup, it will abort upon seeing one of these malformed entries. The master will periodically attempt to restart the negotiator, and each time, the negotiator will abort and exit. As a result, no jobs will run in the local pool, as no matches will be made. This page describes how to get your pool working again once it's in this state.
+Condor 7.8.4 contains a bug that causes the negotiator daemon to write malformed entries in its =Accountantnew.log=. When the negotiator reads the =Accountantnew.log= on startup, it will abort upon seeing one of these malformed entries. The master will periodically attempt to restart the negotiator, and each time, the negotiator will abort and exit. As a result, no jobs will run in the local pool, as no matches will be made.
+
+*Do not upgrade to Condor 7.8.4!* If you have already upgraded to Condor 7.8.4, this page describes how to get your pool working again once it's hit by this bug.
 
 The following sections describe several different solutions.
 
@@ -8,7 +10,7 @@
 
 {subsection: Full Downgrade}
 
-You can downgrade your entire Condor installation to version 7.8.3 or earlier. If you installed the RPM package, you can do this by running =yum downgrade condor=.
+You can downgrade your entire Condor installation to version 7.8.3 or earlier. If you installed the RPM package, you can do this by running =yum downgrade condor=. You only need to downgrade your Central Manager (the machine running the condor_negotiator daemon). The other machines in your Condor pool can be left running Condor 7.8.4.
 
 After the downgrade, you will need to clean your =Accountantnew.log= of any malformed entries. One option is to simply delete the file. If you do this, Condor will lose all knowledge of past resource usage by users and user priorities set by =condor_userprio=, which is used by the negotiator when determining how many machines each user should be allocated to run their jobs.