Paris-Brest-Paris 2023

Paris-Brest-Paris (PBP) is the premier event in the world of randonneuring, both in terms of the number of participants and the history of the event. The original PBP race dates back to 1891, but the event has run in its current randonneur format since 1931.  PBP attracts entrants from all over the world, and is an ambition for many Audax riders. 

PBP 2023

The 2023  edition of Paris-Brest-Paris began on Sunday 20th August, with a total of 6430 riders from 66 countries taking the start line. Europe was in the middle of the record breaking Cerebrus Heatwave and temperatures exceeding 30C caused problems for a lot of riders. 

Riders started in successive waves, with riders aiming to beat the 80 hour time limit setting off first on Sunday afternoon. The biggest tranche, riding to the 90 hour limit followed on Sunday evening, with a final smaller group starting early on Monday morning with an 84 hour time limit.


The charts on this page have been created using the control arrival times  published on the PBP results page . Please note that the analysis on this page is based on the source data as published on the PBP website, and the numbers and outcomes summarised do not represent the official results of the event. The definitive record of every ride is the brevet card carried by each rider, and stamped by the controllers at each stop. These cards are manually scrutinised and approved by the organisers to determine if a rider successfully completed the ride within the allotted time.

Before creating the charts I cleaned up the data to insert missing records and fix some inconsistencies:

More about data cleaning...

Although the control arrival times are recorded automatically by timing chips, approximately 300 riders had missing results, so some data cleaning was required to identify these and insert estimates for the missing records. There were a number of riders who recorded excessively high speeds on various stages of the event. Some of these  were obviously DNFs who did not reach Brest and turned back early (and therefore appear to have covered huge distances in very short times) and I have adjusted their results to reflect this. There are a few remaining riders who recorded unusually high speeds that cannot be explained by an early turn, and I have left these results unchanged.

Analysis of the results suggested that there was something  strange about the Stage 1 timings, and GPX recordings provided by riders confirmed that the timing mat for the first control appears to have been placed on the exit from the control, rather than  the entrance. I have adjusted the results to reflect this, so Control 1 stopping time is included in Stage 2 (consistent with the other recorded results).

PBP 2023 Charts

Summary Outcomes

OTL: "Outside time limit", DNF: "Did not finish", DNS: "Did not start"

DNF Outcomes

Distance, weather and lack of sleep make PBP a demanding challenge, and as the chart above shows, a significant number of riders did not finish or exceeded the event time limits. The following charts look at the DNFs in more detail:

DNFs are fairly well distributed across the event, but peak on the way out at Loudeac (which will be the end of a long first day for many riders) and the turn at Brest. There is another peak at Villaines on the return when the cumulative distance will have become a challenge. The second chart shows that over 80% of the DNFs have occurred by the end of Tuesday (the second full day of the event).


Unsurprisingly, France outnumbers any of the other nations represented at PBP, but the following charts show the geographical range of the riders taking part:

Taking part is one thing, but what about the success rates of individual nations? We all know how this will go, don't we, with the traditional giants of European cycling dominating the field? In fact, Teutonic efficiency takes a blow, the Low Countries are not as high as you would expect, the UK is predictably nowhere in particular, and France only merits a Gallic shrug. With the highest percentage of successful finishers, step forward...Finland!

Finland's showing looks pretty impressive, until we check out the performance of the countries with fewer than 25 entrants. So take a bow, Armenia, Colombia, Sri Lanka and all the other, often solitary, riders who successfully represented their countries with a 100% success rate!

Stage 1 Speed

Stage 1 is the only result where we can assume that the recorded average speed for a rider is close to the actual riding speed (because the majority of participants are likely to have ridden from the event start to the first control with minimal or no stops). After the first control, riders will spend hugely varying amounts of time at the controls they pass through, and riding speed can no longer be estimated directly from the raw arrival times. In the following chart the central figure shows Stage 1 speed against Finish time as a density plot (darker colours represent greater numbers of riders). The top and right  margin figures plot the number of riders against Stage 1 speed and Finish time.

The following charts show the same data, but split by the different time limit groups. Looking at the distributions for the 84 hour and 90 hour groups  it is clear that the median Stage 1 speed is higher for the "In time" group compared to the "Out of time/DNF" group. That initial speed is an important indicator of the final outcome.

As well as the main peaks in rider numbers near the time limits, the right hand marginal plots also show earlier mini peaks, which I assume are the result of riders targeting  other "milestone" times; 60 hours for the 80 hour group, 70 hours for the 84 hour group and 80 hours for the 90 hour group.

Stopping Time

The biggest limitation of the PBP source data is that it only records arrival times at each Control, and it is necessary to know how long riders spent at each control to unlock some more interesting analysis. I created a simple model to estimate riding time on each stage for each rider and used this to estimate the stopped time at each control. A few participants kindly provided me with GPX data recorded during the event,  and I used this to confirm that although the model was not particularly accurate at estimating individual, shorter stops, it was reasonably effective at identifying and estimating the longer Control stops made by riders.

The estimated stopping times form the basis of the subsequent analysis.


With both arrival times and estimated departure times for every rider, it is possible to estimate how many riders are stopped at a control at any particular time (this analysis assumes all stopped time takes place at the Controls, which will obviously not be true in real life). The following charts show the numbers of outbound and returning riders at every control over the duration of the event. The grey bars represent the hours of darkness.

The control charts show the twin peaks of outbound and returning riders at each control, and how the gap between the peak loads narrows as we get further from Rambouillet. Loudeac appears to be the most heavily used control, presumably because its location places  it at the end  of the first two long days of riding for many participants. The return loads tend to stretch and reduce in peak volume for the later Controls, reflecting the number of DNFs and the stretching field.

Here is the same information, displayed as an animated bar graph:

Finally here are the Control volumes plotted against each other as a 3D chart:

First Stop

The following charts illustrate where riders made their first (estimated) long stop and how that relates to the ride outcome. The charts are plotted as percentages of the riders in each group, rather than absolute rider numbers, so the relative shape of each group can be compared. Note that the estimated data suggests that approximately 20% of all riders make no stops of 3 or more hours, so the percentage of riders with no long stops within each group is plotted as the starting value on the left hand of each chart.

The first noticeable feature is that the left hand ends of the charts show that a much higher percentage of DNF/OTL riders do not record any long stops (43%, vs 14% for riders who finished successfully). I assume this is because the DNF cohort includes riders who abandoned before recording a long stop, and more slow riders who are chasing the control cut-offs, with little time in hand.

Looking at riders that do make a long stop, and the distance they travel before making that first stop, the percentages are very similar in both groups up until Loudeac. However, a smaller percentage of OTL/DNF riders travel further than Loudeac before making a stop, compared to riders who finished successfully. The second chart suggests the median time elapsed time before making a long stop is slightly longer for successful finishers, compared with the OTL/DNF group.

In summary, more successful finishers achieve a long stop, and more of them travel further before making that long stop (as a percentage of their cohort) compared to the OTL/DNF group.

Stop Patterns

The following chart summarises the patterns followed by riders for long stops (at least 3 hours) and shows there is a lot of diversity in the sleep strategies followed by riders. As noted above, the most common single pattern is "No stops of more than 3 hours", so around 20% of riders appear to be getting by on minimal stops and catnaps. However, the biggest group of DNFs were in this category.

The other riders, who did take at least one stop of 3 or more hours are grouped by patterns. It turns out that riders adopted a very wide range of different sleep patterns, so to make the chart readable I have bundled any patterns followed by fewer than 50 riders into the single "Other patterns..." category, forming the second largest grouping. In the remaining categories, Loudeac inevitably features heavily, and Loudeac out and return  is the most common specific pattern.

The next charts bring together some of the previous analysis as animated event timelines:

Relative Rider Progress

This chart shows distance covered against elapsed time for every rider (including the "high-speed" riders mentioned in data cleaning!)

Event Animation

In the final chart the riding speeds calculated to estimate stopping time are used to estimate progress between controls and present an animation of rider volumes over the duration of the event: