If you’re new to Crickles and maybe haven’t signed up yet this is for you.
Crickles is free. Please repay us by completing our extremely short survey.
We never give your data to advertisers or in any other way seek to make money from it.
The main purpose of Crickles is to estimate the cardiac stress that you accrue from endurance sports. We summarise this in our Cardiac Stress Score and it gets rolled up into cumulative Fitness and Fatigue measures.
There is no accepted medical consensus for determining how much exercise is too much. However, with Crickles you can accurately compare how much Cardiac Stress you’re accruing compared to other people of your age and gender. If you’re doing more than everyone else then that’s a lot, right?!
Crickles is consistent. For example, if you exercise without a heart rate monitor or a power meter it won’t rate your exercise load as zero. Also, if you switch between using a heart rate monitor and a power meter you’ll generally get similar numbers. Also, as you get fitter and exercise at a lower heart rate for the same “effort”, Crickles will automatically detect this and recalibrate your Cardiac Stress Score adaptively.
Crickles detects strap errors and some forms of irregularity in heart rate data. To continue to improve this, please complete our short survey if you haven’t done so already.
We get data from your Strava account, once you’ve authorised us to do so. You’ll need your Strava ID, which is a number like 301194 – instructions for finding this are given here.
The sign-up process is a bit more complicated than we’d like – we used to email you as soon as you signed up with some help on how to proceed but we can no longer get your email address from Strava. The procedure is given here.
If you get stuck or confused or have any thoughts about Crickles, please get in touch through the Contact page or by email to email@example.com.
A major advantage of Crickles is the consistency of the Cardiac Stress Score (CSS) across different types of sport or activity. If you don’t wear a heart rate monitor Crickles still gives you a meaningful CSS value – and not zero – that is consistent with the activities for which you did wear a heart rate monitor. For the most common types of activity (running and road cycling) when you don’t have a heart rate monitor Crickles now estimates CSS using a machine learning algorithm that tracks very nicely to the CSS you’d get if you did use a heart rate monitor.
Also, if when cycling you sometimes use a heart rate monitor, sometimes use a power meter and sometimes use both, the CSS values you get do not swing wildly according to how you were instrumented on each day.
This consistency is crucial for aggregating CSS over many activities, as shown on the front page of the Navigator. It is equally essential when calculating Fitness/Fatigue curves.
There is a new report called CSS by Sport that leverages this consistency in a new way. It looks like this:
The left hand chart breaks down your CSS by Sport, or activity type. The period over which this is summed is defined by the Date range in the side panel, which defaults to the past six weeks. You can change this to be any period for which Crickles has your activities.
The right hand chart shows the hourly rate at which you accrued CSS during activities over the same period.
In the example shown, the athlete incurs more CSS per hour when running than cycling but nonetheless generated a lot more CSS over the period while cycling due to the number of hours in the saddle. This is detailed in tooltips that appear when you hover over the bars. For example, hovering over the Run bar in the left hand chart shows this:
We see that there were nine runs in the period, the total CSS from these was 451 and that this comprised 10% of the athlete’s CSS in this period.
Hovering over the Run bar of the right hand chart shows this:
This shows that in the period while running the athlete was generating 83.5 units of CSS per hour and ran for 5.4 hours. Hovering over the Ride bar would show a lower hourly rate (actually 74.9) but a far higher number of hours (47.1).
The CSS by Sport report is (only) interesting as and when you engage in multiple different types of activity.
A distinctive feature of Crickles is that it will estimate your Cardiac Stress even when you’re exercising without monitoring your heart rate with a chest strap or sports watch. This is critical if you want to track your Fitness, Fatigue and Form since these are cumulative measures and if you leave out chunks of your exercise regime they’ll be wrong. Herein lies a weakness of most of the non-Crickles sites that claim to show you this. For example, when you can get it to work, Strava’s intensity metric (Relative Effort) has improved (and thus become a little more like the Crickles CSS) over the last couple of years since it replaced Strava’s Suffer Score. Although there is very little consistency between Relative Effort and Power, you can now at least combine them on Strava’s Fitness & Freshness chart. However, if you go for a swim during which you don’t measure your heart rate, or you do a run or a bike ride without a HR monitor or a power meter, then those activities have absolutely no impact on Strava’s fitness and fatigue curves.
Crickles does not have this flaw. Moreover our Cardiac Stress Score tracks as well to the standard Training Stress Score derived from a power meter as you would expect it to. For example, if you do a triathlon with no heart rate measurement on the swim, a power meter on the bike leg and a heart rate monitor on the run you’ll get a sensible Cardiac Stress Score for each leg of the event and they will feed into your Fit-Fat curves.
From this weekend, the estimation of cardiac stress when you have neither a heart rate monitor nor a power meter is significantly improved in Crickles. This has been achieved in two ways:
For (non-virtual) bike rides and runs, we have a sophisticated new statistical model for estimating Intensity in the absence of heart rate data. You can now see this explicitly for each ride or run on the Activities tab of the Navigator in the Intensity column.
For all other activities, the current model has been improved and re-calibrated using the very large amount of data that Crickles athletes have made available to us over time. Rather than giving an Intensity reading for each activity, this generates a Cardiac Stress Score for each activity. CSS for these activities is now shown on the Activities tab. It can also be inferred on the Timeline, where the y-axis position indicates CSS. You can also see the daily sum of CSS values in the hover tips on the Fit-Fat curves where it is called Stress Load.
Historical activities have all been re-calculated according to the improved model. If you always use a heart rate monitor you won’t notice any change. However, to the extent that non-measured activities make up a meaningful portion of your overall exercise load, you may notice. None of the metrics should be hugely different as the prior model wasn’t terrible! In my personal case, I’ve been doing swims and pilates regularly over the past several weeks alongside cycling and running with a heart rate monitor. I can see that the new model assigns a lower CSS to the swims and the pilates than the old one and so my Fitness and Fatigue levels are now marginally lower than they were before the change.
The Crickles community is essentially defined by the athletes who can see analysis of their activities on the Crickles Navigator. The sign-up process is relatively simple, although a little less simple than it used to be before Strava stopped making user email addresses available to Strava Apps. Here’s what you need to do:
For cyclists who use a power meter, the FTP chart on the Navigator now explicitly states your estimated Functional Threshold Power (FTP) at the start and end of the chosen Date Range. By default, the end of the Date Range, and hence the date of the second FTP level shown, is today. This enables you to see at a glance your current FTP and how it has changed over the period. You can also reset the Date Range to see your FTP at the start and end of different periods.
We are occasionally asked about differences between the FTP estimates on Crickles and on Strava or other sites. Usually they’re all close but occasionally estimates can differ materially. There are two known reasons for this:
Short protocol FTP tests – for example, some sites form an FTP estimate from an 8 minute burst. In contrast, Crickles infers FTP from efforts lasting 20 minutes or more since the definition of FTP is the power level that can be sustained indefinitely (or for an hour, depending upon what you read). Strava appears to give more precedence than Crickles to short duration power and you can see a difference in FTP estimates if you’re going full gas over that kind of timeframe.
There can also be temporary differences in the time it takes for different sites to reflect an outstanding performance or set of performances in their FTP estimates. The Crickles FTP estimate shown on the chart incorporates information from all rides except the latest one.
It’s been a little while since the last noticeable change to the Crickles Navigator. From the feedback we get it seems that people find the Fitness and Fatigue functions to be particularly useful, and truer to actual feelings of fitness and fatigue than the alternatives available elsewhere. This supports our belief that the Cardiac Stress Score (CSS) is a decent functional measure of the cardiac stress incurred during exercise.
There are a number of changes that I’d like to make in 2019. First, I plan to update the analysis of data from the Crickles survey. Last time I checked there were suggestive relationships between reported cardiac health and CSS-based measures on Crickles. Now that we have materially more responses, I’m curious to see whether these relationships attain a level of robust statistical significance.
As well as new features such as user-configured alerts, I’d also like to strengthen the Crickles technical platform. This requires a little more investment. To fund it, I’ve explored a number of collaborations with other health and fitness products to see whether the Crickles analytics could add value to them and generate enough income for our own improvements. In every virtually case I have so far found that these other products do not respect user data in anything like the way that Mark and I require and so collaboration is impossible.
One consequence of what seems to me to be widescale abuse of user data is that Strava themselves, whether on principle or in the light of the EU’s General Data Protection Regulation, have tightened up app access to Strava data. Last year they removed the ability for an app to identify a user’s Strava followers/friends. This meant that thereafter Crickles was unable to offer group selection based on your current Strava friends. This is sad but we can’t blame Strava: other apps were, to my knowledge, chaining through users through their follower relationships to trawl data without appropriate consents. At the start of this year, Strava removed the ability for apps to access users’ email addresses. This makes the Crickles sign-up process more difficult, but, again, Strava really had to do that to avoid inappropriate data harvesting by unscrupulous apps. (It also meant more work for me that I didn’t want to do!)
This is all frustrating but it’s the reality of the current social network landscape. Most applications base their business on non-transparent forms of “social listening”: offering functionality as a lure to gain personal user data that can be re-sold to advertisers. Crickles will never do this. What we may instead do is offer a premium tier and levy a modest charge for it. I have mixed feelings about this but it would help me cover my costs and fund the improvements to Crickles that we’d like to make. A premium service may include, for example, analysis of more/all of your activity history; email and/or text alerts; personalised reports; and new/advanced analytics.
Meanwhile, we’ll continue to support Crickles as it is now and occasionally add new features, as we have been doing. Also, I’ll report back on the findings from the survey when they’re in.
Some of you may have noticed this, or something like it, in your Strava feed:
I had a similar “discussion” in my Strava feed from another club – Regents Park Cyclists – a few days ago. The more revealing photo in that instance apparently depicted someone called “Brian”. It goes without saying that we must assume that these come from hackers and/or pimps, and that the photo is no more likely to bear a true likeness to the git who posted it than the message is to reflect an honest yearning for a connection of human warmth.
As far as I can see, there’s nothing to stop anyone from joining any club on Strava and “starting a discussion”: the only remedy is to make clubs invitation only, which would be a shame. If we get more Anastasiia’s, that’s what I’ll do. For now, I’ve booted “her” sorry ass out of the club, even though it’s a futile gesture.
The serious issues behind this, beyond the huge ones of cyber/identity theft/fraud and human trafficking, do impact us directly in smaller ways. At the end of this year apps like Crickles will no longer be able to obtain the email addresses of their members from Strava – presumably because the abuse of this facility is already a problem. This is certainly an inconvenience.
I’ll write more about this, and about data policy on Crickles, in due course.
For now, be aware that the Strava Crickles club has no meaningful relationship to Crickles as you know it from this website and the Navigator. It’s simply a bulletin board with, as it happens, quite a different membership from “true” Crickles. Your data as seen on the Navigator is held securely on the Amazon Cloud (like Strava’s data) and the appearance of Anastasiia in your feed does not imply that it has been hacked.
Fit-Fat charts on the Navigator have had a subtle but important upgrade. Initially, the tab looks unchanged:
The main difference at first is that by default all of your Crickles history now appears.
There is also a new drop-down in the side panel that enables you to chart just one of the fit-fat lines:
For example, if you choose Fatigue as here you’ll see this:
This makes it easier to see changes in each of the three charts.
Furthermore, if you hover over any of the charts you can see the Stress Load (i.e. the CSS) alongside the numerical value of the Fitness, Fatigue or Form on each day. Moreover, you can also draw a rectangle around any part of the charts to zoom in on that time slice – that’s why the Date Range on the side panel is no longer relevant for Fit-Fat. Here’s an example of a zoom in on the above chart to see the time around the end of 2016/start of 2017 in more detail:
Yesterday I was cycling up one of the local hills that is often on my route home when I realised that it’s increasingly unlikely that I’ll ever beat my personal record (PB) for the climb. Of course, I’m getting older and presumably less strong and that in itself makes it less likely, but what struck me was that even without a decline in athletic capability the chances of setting a new PB decrease after a certain time. Let me explain…
Imagine that you do the same event most weeks – say a park run or a TT circuit – over a number of years. Further, let’s suppose that your “true average time” for this is, say, 25 minutes but that your actual time on any particular week is affected by a number of varying factors. These could be, for example, the wind and weather, the volume of traffic you encounter, how well you slept, the state of your kit (bike, wheels, shoes…) and so forth. Perhaps we can summarise these into three separate variables for (1) environment (wind direction, traffic…), (2) personal condition (restedness, current form…) and (3) state of equipment (which bike and wheels you used…). It doesn’t matter what the variables actually are so long as we suppose, for the purpose of this thought experiment, that they are normal, trend-less and independent. Then, we can add all the random factors together into a single “net random factor” and this will also be normal.
Now in reality you may immediately object that in real life the “personal condition” variable is not trend-less because we initially get fitter as we start a new form of exercise and then over time we age and get less fit. However, my key point is that it will feel like this even if it’s not the case! To see this, let’s suppose that our fitness/form varies from week to week according to a variety of factors but, on average, doesn’t change over time.
Given these assumptions, our weekly time will always be 25 minutes plus or minus a varying amount. In the first few weeks it’s in the nature of statistical fluctuation that we will probably quite often set a PB. However, as time passes we’ll have those occasional “magic days” when the wind and air pressure are favourable, we have only green lights, we’re on our fast bike or in new shoes, and we’re rested and well trained. When all of those factors align we’ll set an exceptional time. In terms of our model, the net random factor will be at a rare and favourable extreme value. Because such values are rare, the time between PB’s will increase.
To illustrate this, I modelled it for an athlete who begins doing the regular weekly exercise at age 20 and continues each week for 50 years. This chart shows five sample random paths, which we can think of as corresponding to five athletes all starting at age 20 and performing the same weekly exercise until age 70:
As you can see, in each of the five cases, before the age of 50 (30 years on from the start at age 20) the athlete has attained a PB that isn’t seen again before age 70. For example, the red line represents an athlete who sets a couple of PB’s within the first five years, then a new one at around age 40 that isn’t matched again in the remaining 30 years. The green athlete has the most encouraging career and the orange athlete the least, while the violet athlete manages the latest PB at age 49.
Over 1,000 such random paths the average age of the last PB is 45, halfway through the athlete’s career. Remember, this is just an artefact of statistical randomness and happens even on the assumption of constant athletic ability!
The psychological impact of this is obvious. Most of us judge our “true” capability to be that recorded by our best times. As those PB’s recede into the past, even if our average times stay the same, there is a perhaps a tendency that it will induce a feeling of nostalgia and a worry of ageing. In truth, a new PB could (in this fantasy world of constant physiology) occur just as easily tomorrow as it did 20 years ago. The problem with ageing, viewed statistically and irrespective of physiology, is that we may not have enough tomorrows left in which to replicate the most extreme favourable conditions of our plentiful yesterdays.
Mark and I quite often get asked about suspicious heart rate readings by people using Crickles. Often these are probably just Garmin/strap errors: the majority of our population occasionally see heart rate values that look wrong…
The chart shows the distribution of maximum recorded heart rate by athlete. 62% of athletes show a maximum HR over 200 bpm, for 36% it’s over 220 bpm and the maximum to date stands at 365 bpm. These values are dubious. Data cleaning is therefore an important part of Crickles algorithms.
While it is not an aim of Crickles to train algorithms to give a medical diagnosis of heart problems, we do flag when activity data looks unreliable for use in quantifying the cardiac stress score (CSS). The Activities page on the Navigator now shows a new column called Diagnostic. This is only populated for activities where a heart rate monitor was used – if not, it appears blank. (It may also very occasionally appear blank for other reasons.) Where a Diagnostic value appears it will be one of the following:
Check_Strap – it looks probable that there was a recording error and the heart rate data for this activity is wrong;
Irregular – the heart rate data stream looks questionable but Crickles cannot reliably ascribe this to a strap error;
Regular – the heart rate data is good for use in the measurement of CSS.
This algorithm that produces this diagnostic does about as good a job as I can do by eye at identifying odd-looking data streams, and (unlike me) it can do this consistently on the hundreds of thousands of activity records in Crickles. However, it is not in any sense a medical diagnosis and the appearance of only Regular values is no guarantee of good health.
When Crickles athletes email us with concerns about their cardiac health I do sometimes opine on how relatively un/usual the data may look but the medical aspects of such questions are always addressed by Mark, who is a cardiologist. Mark can look at the data in the context of symptoms, such as chest pain or fainting, and the athlete’s medical history.
To explore any Check_Strap or Irregular activities you may have, on the Activities tab you can:
Change the Date Range in the side panel to select the time horizon you want to explore;
Use the Search box on the top right of the screen to pick out Check_Strap or Irregular values;
Use the small triangle next to Diagnostic to sort your activities by Diagnostic.
The Regularity page has had a make-over to show the frequency with which Irregular values occur. Previously only available as a beta feature by request, this page now has two charts. The one on the right is the chart that was present previously:
This shows whether your recent aggregate heart rate pattern is different from its historical pattern. Significant changes such as that shown can be due to an intentional change in your exercise regime – for example, reducing the intensity of exercise. If the gloss at the top indicates a significant change with sufficient data for a valid comparison (as here) but you haven’t knowingly modified your exercise habits it may be worth digging in further.
This chart responds to the three checkboxes in the side panel, as before.
The left-hand chart on the Regularity tab is new:
This shows quarter-by-quarter how often you’re getting Irregular as the Diagnostic for your activities on Crickles. As with CSS, there is no firm science on what constitutes a good value but what we can do is show how you compare to the (Crickles) crowd. Values above the two orange lines, and especially the solid orange line, are unusually high.
The size of each quarterly point indicates how many activities contributed to it. A high Irregularity Ratio is less meaningful when it is derived from only a few points. As a guide, 30 points can be taken to constitute a good sample. The gloss above the chart tells you exactly how many Irregular diagnostics you’ve had in the current quarter, and, for good measure, the number of Check_Strap diagnostics (which is not shown on the chart).
If you consistently see an Irregularity Ratio above the orange lines based on a meaningful number of activities, it’s worth changing your heart rate strap. If you continue to see a high ratio, we’d be interested in hearing from you.
While Irregularity Ratio is a useful measure for data verification, there is no science that establishes an association with cardiac health. Intriguingly, a number of our active athletes have filled in the Crickles survey and, amongst these, the average Irregularity Ratio happens to be 56% higher in athletes who report a diagnosis of Atrial Fibrillation than amongst those who don’t. However, to attain significance in a statistical test – or to find that it’s a coincidence – we’d need many more people to fill in the survey. If you haven’t done so yet, please do so here. The survey is super-quick to complete and all responses are equally useful, even if you have only good health to report.