There are two things that can create this kind of problem:
 
1. If the logs are edited, for  instance to remove junk messages, all the
   database index numbers will change and the numbers in the index mailed
   to the  subscribers may no longer  be valid. This is  unfortunate, but
   for the  time being  the index code  can do nothing  about it,  as the
   numbers come from the database system.
 
2. The index  system knows exactly what messages were  posted in the last
   period, the  order in which  they were  posted, the subject,  date and
   size of  each message. This  information is  kept in a  temporary file
   that human beings are unlikely to tamper with (the list owner does not
   have access to it at any rate).  The database system on the other hand
   works on log files which can be  edited manually by the list owner. To
   detect the separation between messages, it has to rely on the presence
   of a  separator line  followed by a  valid header. If  a user  posts a
   message containing a separator line  and another valid message header,
   the index  system will see  one long  message and the  database system
   will see two short ones. Again,  this is a shortcoming of the database
   system that the index code cannot do anything about.
 
In order  to solve  both problems,  one needs  message numbers  and sizes
inside  the  log  file,  with  some  sort  of  protection  against  human
tampering. This means list owners would  no longer be allowed to grab the
whole file,  edit it with  their regular text  editor, and send  it back;
they  would need  to use  a special  interface, something  like "send  me
message 32812  and I'll  send you  an updated  copy". People  will scream
bloody  number  unless  clients  are developed  that  make  this  process
transparent  to the  user. It  will be  done eventually,  but this  is no
10-minutes fix. Once we have general-purpose PC clients to offer, it will
become a lot easier to start thinking about it.
 
  Eric