I have been a grumpy old Running Mann the last few months. A persistent injury will do that to your mood. Sadly, that means there are no new race reports but the silver lining is that time away from the road has left me with plenty of time to run Comrades statistics.
Most of the stats fall into the interesting and inspiring categories but the odd result jumps out and adds to my grumpiness. Uncovering cheats at the world’s greatest ultra marathon leaves a bad taste in my mouth and a shooting pain in my bad knee.
I work in the software industry. We call problems in production systems ‘defects’. Good software teams will aim to prevent defects from ever getting into production systems and, when a defect is detected, will aim to remove it as quickly as possible. Software teams will also put steps in place to ensure that the defect is never allowed to resurface. I see runners that cheat in Comrades as defective humans. They should be removed from the results and forbidden from re-entering our running systems.
This article discusses the different types of Comrades cheats and cheating, analyses how much cheating occurred during the 2022 event and uncovers some defective humans. At this stage, I have decided not to include the names and faces of potential cheats but do plan to do so after the outcomes of disciplinary hearings.
The seven types of cheats and cheating covered in this article are:
- Tag Teaming (running Comrades as a relay race with a buddy)
- Taxiing (covering a portion of the route in a vehicle)
- Phantoms (one runner, two timing chips)
- Bad Seeds (cheating on your qualifier)
- Bandits (running in someone else’s number)
- Ritchie McCawing (entering from the side)
- Suppositories (pushing from behind into places you don’t belong)
Aside: The article does not cover doping but you can read about that type of cheating here – Drug Running at Comrades (and discrediting American ignorance).
I’ve been doing different stats on every conceivable aspect of Comrades and this sometimes uncovers suspicious data. I was interested in the slowest runners to go through each timing mat by medal earned. I picked up a few anomalies with this analysis, including one definite cheat.
The timing data shows that our first cheat ran the second half of Comrades almost 90 minutes faster than the first whilst the photographic data shows two different people at the start and finish. Running Comrades definitely changes a person but normally it’s an internal change to one’s psyche rather than an external change to one physique before and after Drummond. One party in this relay team is a Green Number holder making the duplicity even more disappointing.
The cheater’s running club also contacted me after seeing my social media posts about Comrades cheats. They let me know that they had also contacted Comrades directly to inform them of the cheat. I am told that he is a senior manager at a well-known mining company and is totally unperturbed about his blatant cheating being outed.
When it comes to cheating, practice makes perfect. In a further twist, photographic evidence shows that this pair ran their qualifier at the Benoni Northerns Marathon as a relay event as well.
Aside: It is incredibly difficult to run a negative split at Comrades (which is where a runner completes the second half of the race faster than the first). The feat is made even more difficult as the “halfway” Drummond timing mat is well before true halfway meaning the second “half” of the 2022 down run was 2.12km longer than the first. Only 83 runners (well under 1% of the field) managed a negative split in 2022. From the analysis I’ve done, 81 are genuine negative splits and two are cheats (the second is outed in the “Taxiing & Missed Mats Analysis” section below).
Taxiing & Missed Mats Analysis
This was the first year that FinishTime were responsible for race day timing and instead of a separate timing chip attached to the shoe, the UHF timing chips were embedded into the race numbers. How accurate are these chips at picking up all 12,000 finishers at every timing point?
The short answer is that the timing mats are very effective at picking up all the runners. There are an incredibly small number of runners, just 229 (less than 2% of the field), who miss a timing mat and “claim” a finish. Most of the missing splits occur at the first two mats after the start line, 71 at the 5km mark and 139 at Lynnfield Park.
I chatted to Craig Eldridge from FinishTime as to the reason for the large number of missed splits over the first two mats. The official timing points have two receivers but the 5km only had one receiver as it was used to get data for the pacing app (rather than being an official split). Lynnfield Park was more concerning as is it appears that there were a couple of brief ‘dead periods’ where data was not uploaded to the cloud (mobile reception in the area is terrible which might be the cause).
All of the other mats range between 6 and 12 missing splits and this number is even lower when you deduct the cheaters from the equation. As an example of just how low the probability is of missing one of these mats, only 8 runners missed Drummond which equates to a minuscule probability of 0.067%.
The probability of missing more than one mat is even smaller. Of the 229 runners that crossed the finish line and missed a split, the majority (207) miss one mat, 18 miss two mats, 3 miss three or four mats and 1 person misses all eight.
Analysis of Runners who Missed One Timing Mat
The good news is that I did not find too many cheats. The even better news is that the suspected cheats are on the race referee’s radar and disciplinary proceedings are in progress. Pending the outcome of disciplinary hearings, I have given code names to runners based on which split they missed and the medal they earned. The finish line mat was live for 10 minutes after the final cut-off and I have included all runners that crossed the finish line in the analysis.
The Math: I calculated the runner’s pace over the missed split and then compared it to six other pacing calibrations:
- The runner’s overall average pace on race day
- The runner’s average pace before the missed split
- The runner’s average pace after the missed split
- The runner’s fastest timed single split on race day (excluding the first 5km)
- Other Comrades 2022 runners who earned the same medal
- Other Comrades 2022 runners who started from the same batch.
In the tables, a green font indicates that the missed mat runners were slower (and unlikely to be a cheat) and red indicates that they were faster over the missing split. A runner that ran much faster over the missed split than over the timed portion of his or her race raises suspicions.
The most significant pace deviations in detecting potential cheats are the comparisons to the runner’s fastest split and to other runners who earned the same medal (4 and 5 above). A large negative deviation means that the runner ran their fastest pace over that missed split and / or was an anomaly when compared to other runners who finished in a similar time. For example, if a runner’s missed split was 50% faster than their next fastest split or other runners who earned the same medal, it’s highly probable that they cheated.
Another tool I used was this table showing the Average Pace per Double Split. This calculated the average pace per batch and medal over two consecutive splits so that a like for like comparison can be done to runners who missed splits. This shows that the segment between the 5km mark and Cato Ridge is the fastest for almost all runners, which means that it is much more likely for runners that missed early splits (like Lynnfield Park) to have a wider deviation but runners that missed later splits are highly unlikely to outperform their missed split. For example, if someone missed the Sherwood timing mat and ran their fastest split over this segment, they are either a complete anomaly (as this is the slowest average pace for everyone) or they cheated.
What I Found
Runners who Missed One Timing Mat
Start Line: There were just six runners that missed the start line timing mat and had no other missed splits. Three (Start Mat Vic 1,2 and 3) were wheelchair runners photographed at the start so they are 100% legitimate.
There is also nothing overly suspicious about the other three runners, so this is a “Clean Audit”.
Early Splits: Due to the volume of runners that missed the 5km and Lynnfield Park mats, I have filtered the table below only on runners where there might be some suspicion. Lynnfield Vic 59 is almost definitely a cheat, “running” at sub-3 minute pace over the missing 25km but he can only manage an average of 7:53/km thereafter. Additionally, he would have been running in the top 10 to Cato Ridge but is mysteriously missing from all television footage.
Bearing in mind that this is the fastest double split segment and most runners go out too fast, I think the probability that any of the “Slightly Suspicious” runners cheated is low.
Late Splits: I’ve included all 22 runners who missed a mat from Cato Ridge onwards. Cato Bronze 1 is Darius Bezuidenhoudt, he was happy for me to use his 100% legitimate finisher data as a point of reference for more suspicious times.
Winston Vic 3 is a Green Number runner who very narrowly makes the Drummond cut-off and then runs almost 90 seconds a kilometre faster than any of her other splits over the next 25km. Whilst the pace is possible, it is highly improbable based on her other data and that of the field – there was only one other person that went through Drummond in over 6 hours and finished. Further to this, the runner has finish line photos with what appears to be her son. My working assumption is that this runner got into a car after Drummond, waited for her son at the side of the road and then rejoined the race somewhere before Pinetown. In further intrigue, the son also missed the Lynnfield Park mat and he managed to pull off his best performing segment over his missing split.
Aside: It’s incredibly rare for a runner to miss a split cut-off by less than ten minutes and still finish the race. Only 9 people managed to legitimately do so at this year’s event.
Whilst a missing split pace of 4:50/km is definitely possible for A seeded runners, Drummond Bill 1’s time is out of sync with the rest of his data so I flagged him as suspicious. However, a possible explanation could be that this runner held back until Cato Ridge and then decided to see if he could blast through for a silver and failed. The good news is that additional validations exonerate this runner, so his missing split time was one of the few legitimate anomalies.
Sherwood Vic 1 is also suspicious because it’s so late in the race and very few runners are able to radically outperform the rest of their race data at this point. I wondered whether this was a runner who had to move it to make the final cut-off but he was very safely under 12-hours. In his defence, he was able to run earlier splits faster than this pace (and the pace is potentially within the reach of a D seeded runner).
Runners who Missed Two or More Timing Mats
Missed Two Consecutive Splits: There are 14 people in this category, the majority of whom (12) missed the 5km and Lynnfield Park splits. For me they all check out as legitimate finishers and the pace is consistent with other runners who did not miss the splits. The early splits are the fastest so the “pink flags” in the comparison to overall pace and pace after the missing split can largely be ignored.
Missed Two or More Non-Consecutive Splits: I am comfortable with all four runners who missed two non-consecutive splits. The numbers are all within normal ranges of their other data and comparisons with the rest of the field. The same cannot be said for the only runner to miss three splits. This runner has already been outed on social media for been a “shapeshifter” at Two Oceans and Cape Town Marathons where someone else ran in his number. His pace over the first two missed mats (the start line and Lynnfield Park) are very unlikely and, whilst his Drummond time is within range, my feeling is that he did a little bit of running and lot of driving around the route. I suspect he carefully timed his appearance at the mats he was recorded at to look realistic for a silver. If this is the case it shows an alarming amount of premeditation. It will be interesting to see what his disciplinary hearing brings up.
Missed Four Consecutive Splits: Two runners are recorded at Cato Ridge (30km into the race) and are not seen again until the finish line. MIA from Drummond 1 runs at an average pace of 8:22/km to Cato Ridge before running twice as fast (4:18/km) from there to the finish. A clear cheat or an enthusiastic spectator? There is no clear finish line photo of him so he may have bailed, made his way to Moses Mabhida Stadium and then had his race number picked up at the finish line as an enthusiastic spectator.
The second falls into the highly suspicious category as he would have had to run a massive negative split over the last two-thirds of the race to achieve a Bronze medal after running an average pace of 8:07/km to Cato Ridge. His time could potentially be possible for an A seeded runner who has a really rough start and then comes right in the second half of the race – but since our potential villain ran from H batch the suspicions are increased.
Missed All Splits: The final investigation is a runner that missed all eight splits before appearing on the finish line. This is a sad rather than malicious story. The number belongs to Bloemfontein runner, Moeketsi Selemela, who was murdered in July (his estranged wife is one of three suspects on trial for his murder). Selemela’s brother, Thabang, and another Centre City Titanium Athletics Club member, Thami Binda, carried a large banner of Selemela with his Comrades race number attached over the finish line in commemoration of their fallen Comrade.
I collaborated with Andrew Collier, a data scientist and fellow Green Number holder, on this one. Andrew coined the term “phantom runner” in this blog post identifying cheats from the 2014 event. Essentially this is where one participant runs with two chips resulting in a ‘phantom runner’ whose time is recorded at each split but there is no photographic evidence of the runner being there.
Aside: A number of people have been caught cheating via this method over the years. Wits statistician Mark Dowdeswell has also diligently detected and reported phantoms over the years.
Andrew has written code that compare runner’s split times with each other and then does a cumulative calculation of the difference in timing mat splits. There are many people who will end with exactly the same finish time but very few who would go over all eight timing mats at exactly the same of day.
We looked for two specific cheating methods:
- Where a runner runs the entire race with two numbers
- Where a runner takes over a bailing runner’s number during the race.
Andrew and I could find no evidence of cheating via this method – all the runners who had crossed all (or many) timing mats at the same time of day were enjoying each other’s company on the road to Durban. In past years there were always a few phantoms but I guess it is a lot more obvious running with two large race numbers than with two small chips, so hopefully this method of cheating is now extinct.
Bad Seeds (Qualification Cheats)
This is the petty theft of Comrades cheating but, like having your mobile phone stolen, this type of cheat is a major irritation to the honest runner diligently running their qualifiers and earning their seeding. I have not deliberately gone out to try and find qualification cheats but several people contacted me with details of cheating, so I have passed these on to the race referee.
A couple of examples are:
- An elite female athlete that posted a 2:45 qualification time at a small marathon where there is no evidence that she ran in the race. It is possible (but unlikely) that this was a data capture error and she ran 2:45 at another marathon.
- One special idiot who drove a substantial portion of the Arthur Creswell 52k ultra, posted the splits on Strava for the world to see and used the race as his qualifier.
Bandits are people who run in someone else’s number. At Comrades this would normally be because a runner cannot run due to injury or some other reason and gives (or sells) their number to someone else to run in. There have been reports from other races of spectacular banditry backfires, most notably when the bandit removes their own mask by winning the race and outing themselves.
Comrades pre-race registration is rigorous meaning that any banditry would have to be deliberate and pre-meditated. I have mixed views on banditry (perhaps the topic of a future blog post) but with Comrades, where each finish counts towards Green Numbers and there is a meticulous record kept of every finisher, it is very clear that banditry is both against the rules and ethically wrong.
I am sure that there were some bandits at Comrades 2022 but unless Big Brother level facial recognition cameras are installed, they are very difficult to detect. The best way to unmask bandits is for fellow runners to raise the red flag when they see someone in the results whom they know did not run.
The one stat that I used to detect potential banditry was comparing predicted Comrades finish time (based on submitted qualifying time) with the runner’s actual finish time. The traditional equation to predict a recreational runner’s Comrades time is marathon time * 2.5. The chart above lists the people who beat the prediction by the largest amount (for runners that submitted an ultra marathon qualifier a distance adjusted marathon qualifier time was calculated based on the table below).
I have done some cursory checks like comparing the race photos to the gender, running club and age of the largest “beat the prediction” runners and there are a handful of runners that either use remarkably powerful anti-aging night cream serum or banditted.
For those that don’t follow rugby, this means that the runner “entered from the side” by joining the race along the route and not running the full distance. I did receive a troubled comment on one of my social media posts from a former Comrades runner that someone he knows started two kilometres into the race based on Strava data. Some other commentators pointed out that she might just have forgotten to start her watch when the gun went off.
Fortunately, after a series of direct messages and checking the data, I was able to clear things up and confirm that she definitely started the race. This analysis may also have saved a friendship based on the feedback I received, “Thank you very much, I appreciate the info. I am a very close friend of her husband – very glad now I haven’t said anything to him.”
These are runners from slower batches that manage to start in the front with the elites, so named because they leave a bad smell as they push through from behind and cause major discomfort squeezing past patient runners who are waiting for the race to start. Although this is the mildest form of Comrades cheating, people that push themselves into places they aren’t supposed to go are a major pain in the arse.
This one is quite easy to check by comparing the actual start time with the runners’ seeding batch. For example, without sneaking into a higher seeding pen or aggressively pushing past patient runners, there is no way a D seeded runner could cross the start mat in the first 10 seconds or an H seeded runner in the first minute. The table provides a heat map of the time taken to cross the start line by batch.
The graph below shows the number of runners by batch that crossed the start line in 30 second intervals. It’s interesting that by far the most (over 1,700) cross in the first 30 seconds but after that there is a rapid drop-off down to a stable rate of just over 800 per 30 seconds from three minutes onwards. It’s unclear whether the drop off is primarily due to congestion or whether slow runners pushing further ahead than they should have a big impact on decreasing the throughput. However, what is clear is that for every additional 800 starters those at the pack of the pack can expect to add 30 seconds to their start time.
As for suppositories who take things to the next level, I had a look at the race photos and it does appear that one H seeded runner converted the H into a B. The G on the race numbers also looks very similar to the C so I expect several Gs sneaked into the C pen using this design flaw.
Looking at the detailed data, the most forceful start line suppositories are international runners and members of SAPS running clubs (I guess they think they can still use their blue lights at the Comrades start). There also seems to be a fairly high DNF rate amongst suppositories as well as a low correlation to fast medals for their batch (e.g. very few of the B suppositories earn a silver medal). This suggests that it’s better to save your energy for the race rather than rucking and mauling through the field before the gun fires.
Why do people cheat?
I suppose this leaves one final question? Why would anyone, especially recreational runners, cheat at the Comrades Marathon? Surely it is just cheating yourself? It is really hard to understand what goes through the minds of these defective humans.
For those that cheat on race day by tag teaming or taxiing, my only thought is that the cheat worries about the shame of facing friends and colleagues without a medal. I would think that the ignominy of being outed as a cheat would far outweigh any embarrassment of missing out on a medal. No one at your running club will remember that you didn’t finish Comrades 2022 but you will be eternally remembered as the person who cheated at Comrades and dishonoured your running club.
The only other thing I can think of is that people presume that they will get away with cheating. This article is my contribution to ensure that they don’t. There seems to be an appetite within the new KwaZulu Natal administration to eradicate all forms of cheating at Comrades. Here’s hoping that the defective humans are removed from our roads and that, if I do have to write a follow-up article after the 2023 event, it is a lot shorter than this one.Follow Running Mann: