https://web.archive.org/web/20000706232447/http://www.escher...
"Any fear of complete local failure is alleviated by the fact that each workstation can recover from a central location."
And also, er ...
"Riposte protects against fraud by providing a clear and complete audit trail for every transaction."
The concept looks really nice on paper, also the network communication considerations [2]. But there were bugs in Horizon like when a notebook at a post office was shutdown before a certain time, there were no day-end-markers added to the last transaction, and it seems they had no concept to let the system check this on the next day. There were bugs where the start times for transactions where missing, and these transactions were filtered in certain views however these messages eventually were transmitted.
[1] FUJ00080526 - Fujitsu Report Horizon Data Integrity v1.0, https://www.postofficehorizoninquiry.org.uk/evidence/fuj0008...
[2] WITN00780100 - Richard Roll - Witness Statement.pdf >[...]there were several different types of configuration, depending on the type of PO (mobile, single counter, multi counter or main PO with a separate gateway computer), but from the SPM's perspective these would have all looked the same. >General security protocols were in place at the PO's. Secure passwords were required, which were not supposed to be shared but often were. There was a secure link from the PO counter to Fujitsu's servers —the counters would only respond to requests originating from specific telephone numbers —and all data was encrypted. Additionally, the network was completely (logically) isolated form the internet, it had its own dedicated lines[...]
Computerphile video from 2 years ago about the bugs: https://www.youtube.com/watch?v=hBJm9ZYqL10
Fujitsu casestudy PR document from 2004, still on their website! with a bit of high level info and bumf about the system. Has a few photos of the hardware.
https://www.fujitsu.com/downloads/SVC/fs/casestudies/uk-post...
This system audit manual from Jan 2000 looks like it has a lot of detail but I havent had the chance to read it yet
https://www.postofficehorizoninquiry.org.uk/evidence/pol0002...
I've watched some parts of [1] on youtube, and it seems the riposte system is on the machines in the post office. if it is also used in headquarters, I don't know. they have messages that have incremental sequence numbers to identify the replicates messages.
messages are put into the message store where each message has a checksum. when a message is read and gives a CRC error, someone raises a call, and they rebuild the message store. which recalculates the checksums, I assume.
CRC errors can be caused by data corruption on the counters. so this implies data is stored on the counters. when they rebuild the message store it seems they can use data from the other counter which hopefully has no corrupted message (but the former software engineers can't quite remember if this was true, at least she says so). so it seems they have redundancy between these counters. when CRC erros happen often, they'll replace the counter.
I also assume they describe the Horizon system pre-2010.
[1] "Anne Chambers - Day 67 AM (26 September 2023) - Post Office Horizon IT Inquiry", https://www.youtube.com/watch?v=q72zvd9Iz3Q