Legal robbery explained - the logchecking process
A "golden log" is a big achievement in a contest - a log without own mistakes. Such logs exist even with more than a hundred QSOs made. But for most of our participations we have to live with the fact, that we are robbed of QSOs and multipliers by a gang of humans, computers and software called the logcheckers. Here is how they do what they do - giving you perhaps some hints how you can minimize your losses to them.
First step - log delivery
When you submit a log it will be checked by our upload tool if it meets requirements like the file format. It will show you either an info that your log was accepted and uploaded - or inform you whether and what problems occured with the log. It is i.e. very good in finding QSOs out of the contest period (a hint that you may have logged in local time or with the wrong time zone set in your PC) or detect wrong file formats. Many try to upload logs with a wide variety of formats from jpg over adi to cryptic output of exotic logsoftware - other than the needed cabrillo. But for the sake of the checking software which needs standardized logs the upload tool and we have to be somewhat picky.
Second step - automatic logcheck
Now the software reads all loglines and compares them against each other (and of course whether a claimed contatct is in the other station's log, otherwise deleted as a so called NIL). It verifies QSOs when calls, band, time and exchange of both stations match. It deletes QSOs if it detects a mistake by one station, i.e. you log DOK X09 from DL8MBS while his log shows that he has sent DOK X06. In most cases the software can detect who made the mistake, especially with miscopied DOKs and serial numbers.
It is a bit more difficult with callsigns
If you log as your QSO number 321 DL8MBH who sent you 599 X06 but you are not in the log of a DL8MBH at that time, the software would basically decide a NIL for you and delete the contact. When there is a log of DL8MBS who claims at this time a QSO with you, having logged number 321 and sending X06 - then the logsoftware decides that you miscopied the call, deletes the "DL8MBH"-qso in your log as a "badcall", AND at the same time verifies for DL8MBS the QSO with you in his log as ok - even if his call is basically not in your log. But the case is clear. Somewhat tricky but fair - if he copied your call and exchange correctly and obviously was the station worked.
Generally: In case of a mistake we only delete the QSO for the station who made the mistake, we don't delete the QSO for both as long as only one station has a mistake.
In this way the software can check by itself about 70 percent of all the claimed QSOs. Unfortunately there are the cases the software can not reliably decide on its own, leading us to stage three.
Third step - detective work of referees
Let's start with band-errors when both stations in a QSO claim the contact on different bands for whatever reason from defective CAT to forgotten notice of a bandchange. The human referee has to look at the QSOs done by both stations and their counterparts before and after the contact to decide who will lose the qso.
The next big problem arises from QSOs with stations who did not submit a log. The software creates so called "tracelogs" for them based on all claimed QSOs with them in the logs of submitting stations. So we can see whether that station was really active at this time, what his DOK or range of sent serial numbers was. It often allows to delete questionable QSOs.
The "conventional wisdom" that it is not too necessary to care for precise logging of stations who supposedly do not submit - will turn out as defeating yourself. What we do to check such calls and how we use our SDRs (tnx to Heiko, DL1RTL, for running them) is described here.
But the major animals to hunt are uniques - calls that appear in only one single log. From normal operating experience it is very unlikely that someone tuning the bands in a contest works only one single station. It may be the case with rarer DX-stations that operate in the contest and are worked by a lot of DX-hunters that have no interest in further contest QSOs. So uniques can happen but in most cases they are simply the result of miscopying a callsign.
Is an unique an "unique"?
It takes creativity of the human and a database of all active callsigns to find similar stations who were qrv at this time and with fitting exchanges (easy with DL8MBH vs. DL8MBS, but it can be tricky with B7P vs. SZ6P). That often allows to declare a "unique" in reality as a miscopied callsign ("badcall").
We may also check databases whether a callsign is issued at all. If not the QSO gets lost, too. We may also send out mails to stations whose callsigns were logged and we doubt that they were really active in the contest (someone claiming on his qrz.com site to be only active on VHF is somewhat unlikely to have worked someone in a HF contest). Cluster- and RBN-spots also give hints who may really be behind an "unique".
On a funny sidenote: We had cases when someone worked a spotted station and logged the spotting station ;-)
During this process a few thousand uniques morph into "uniques" subsequently deleted as "badcalls". But if we cannot prove that a unique is another specific station being active the unique will be counted as a valid qso. "In case of doubt..."
Fourth step - "The bill, please!"
Once software and judges have done what they could it is time to recalculate the score of the now a bit smaller logs to get the final scores. In the score lists we show the point deduction resulting from the logchecking process for the TopTen stations.
But every station gets a detailed account how its log was judged showing the number of dupes, own logging errors with exchanges or calls, NILs (QSO in your log but not in that of the other station), band-errors as well as the number of remaining uniques and the errors caused only in other stations' logs.
For your eyes only
These so called UBN-reports will be mailed to each participant after the logcheck - but before the publication of results. This allows questions and eventual corrections. Each participant receives only the individual UBN report. We do not publish all UBN-reports.
But If you want to compare your log accuracy: The average deduction in 2015 was 8,7 percent of the claimed score, but only about 3,1 percent amongst TopTen operators. And rankings can become very, very close: In 2015 two category wins were decided by a single digit number of QSOs. In one case one single contact with a multiplier decided the first place.