Colour by effort window

To assess an activity, Crickles looks at your best sustained effort over four durations - 6, 20, 40 and 60 minutes - and takes whichever of those produces the highest figure once each is weighted appropriately. Until now the page didn’t tell you which of the four had won. Now it does, through the colour of the point: the lighter the shade the shorter the window, the darker the longer. Reds on the heart rate chart, greens on the power chart.

This is the same figure that appears as Best_HR_Bucket and Best_Power_Bucket in the activities table on the Dashboard, so if you want to check a particular ride you can find it there too. The window now also appears when you hover over a point, alongside the value, the date and the activity name.

It’s worth a look over a year or two. If your points are mostly light you’re a rider whose best efforts come in short bursts; if they’re mostly dark you hold your efforts for longer.

Showing and hiding windows

Under each chart there’s a key showing the four windows. These are clickable. Click 60 min and those points disappear; click it again and they come back. If you want to see only your hour-long efforts, click the other three off. It’s a quick way to pick a pattern out of a crowded chart.

Older activities

Some activities show up in grey, marked unknown. These are ones from before Crickles began recording which window produced the best effort, so there’s nothing to colour them by. They’re mostly at the left-hand end of a long look-back period and there’s nothing wrong with them.

A calculation improvement

Alongside the visual changes, sustainable heart rate and power are now calculated slightly differently. Some implausible figures arising from faulty recordings - a power meter that wasn’t reading, for instance - are now discarded rather than treated as measurements. The effect is small and in one direction: your sustainable levels may come out a little higher than before.

This applies automatically to new activities. If you’d like your earlier activities re-analysed on the same basis, refresh your data as described here - just hit Confirm Refresh, there’s no need to select the remove cached stream files option.

As ever, do get in touch if anything doesn’t look right.

Metric: Day of Week

When chosen, this sets the y-axis of the Dashboard to show the day of the week.

Colour by: Activity Type

This enables you to colour each point on the Dashboard by the activity or sports type as an alternative to heart rate intensity (HRI). When this coloration is selected you’ll see that the points become diamonds rather than circles as a further cue that a different Colour by option is in play. On a large screen you’ll see a key at the bottom of the chart showing which colours correspond to which types of activity.

Using these two new options together, generates a display like this:

The TOTAL line indicates the effort for the week - to see the detail just hover over each point.

You can pick your preferred choice of Metric and Colour by option independently - for example, you can show Average Heart Rate on the y-axis and colour by Activity type to see the average heart rate for each activity and the pattern over time selected in the Look back period.

If you find a configuration that you prefer to the default settings, just use the Save settings button and that will become your new default for the Dashboard.

Pro tip

You may have noticed that on a small screen - typically a phone - the table of activities under the chart and the Summary section that you see on a larger computer display is not visible. To see it on a phone simply rotate the phone sideways and the extra display width will bring the table into view.

This indicates that Crickles has detected a short discontinuity under effort in the heart rate graph - you will usually be able to see this by eye on the Charts page for that activity.

While the occasional Irregular reading is not uncommon, Crickles analysis shows that a high frequency of such readings is strongly associated with heart rhythm issues. This analysis relies on the heart health surveys that many users completed in the past, and more recently the three questions about heart health that we ask you to complete on the Profile pop-up. We’re very grateful to all of you who have supplied this information.

The analysis can be read on this website under Research > Detecting Arrhythmia.

Mark, the Crickles cardiologist, postulated at the time of the research that it was probably explained by Irregularity being an indicator of ectopic beats, which, while themselves not uncommon, can be a precursor to heart rhythm issues. By analysing ECG data taken from sports devices to measure HRV, we have now been able to show that:

1. It is possible to detect ectopic beats from such device data;
2. The presence of ectopic beats on an activity is indeed strongly associated with the Crickles Irregular flag.

This new analysis is also available on this website at Research > Ectopic beats and Crickles’ Irregularity signal.

It is important to stress that Crickles is not a medical tool. A high incidence of Irregular activities does not mean that you have a heart problem and an absence of any such readings certainly does not mean that you don’t.

Cardiologist Mark gives this advice:

“I would suggest that if you do see a higher incidence of irregular activities, the first thing to do is to correlate that with symptoms – if you are noticing something too – ectopics – a strange sensation in the chest, often described as a thumping or irregularity, or if you are getting other symptoms such as chest pain, increasing shortness of breath or dizziness – then that makes it more likely you are having problems and things should be checked out. Ectopic beats can be triggered by an infection or overtraining, amongst many other causes. Sometimes, time and a temporary decrease in exercise intensity are all that is needed, but they can also be a sign of a more serious issue. Remember that ectopic beats are almost universal over the age of 50, and a low percentage (typically <1%) is not concerning if there are no other issues. Please remember that this is not a medical-grade tool. We also need to point out that not all cardiac conditions cause heart rhythm issues. If you are not feeling right, even if Crickles isn’t flagging an issue, it doesn’t mean you don’t have a problem.”

1. It works much better on mobile devices.
2. It loads instantly.
3. Your activity updates are usually immediately available.
4. You won’t get any server disconnects.
5. It won’t slow down at times of high user volume.

Most functionality has already been migrated to the new version and features that have not been migrated yet will be there very soon, along with a general tidy-up. The current version of the navigator will still be available here until this is complete; you are advised to use one version or the other and not to flit between them.

For current Crickles users if you use the same email address to log in that you’ve used before then all you need to do to get going is request a new password using the Forgot password feature. A complete guide to getting started on the new version, no matter whether you are a current, lapsed or new user, is available here.

If there are issues they are most likely to be in your first visit to the new Navigator. If you run into difficulties (or if you love the new version!) please let me know.

During this time, the Navigator should be unaffected but you may find some glitches in activity updates while both the current and the new system are processing the same real time events. As ever, if your data looks wrong you can refresh it from the Profile page.

If you’d like to join the (hopefully short) beta programme, just drop me a note.

The new algo will apply to any new activities from now on. It you want to retrospectively re-analyse all of your prior activities you need to refresh your data, as described here (there is no need to remove cached stream files so don’t check the box).

Crickles is entirely programmed in R and Python with the vast majority done in R. Python is used for AWS lambdas. The http server that manages interaction with Intervals perhaps ought to be in Python too - it has been in Python in the past (before Intervals displaced Strava as our source of data) and will probably be re-written in Python again in future. For now, it happens to be in R.

All offline analysis as well as the Navigator web app are written in R. Since there are now many more programmers using Python than R you might legitimately wonder why this is so.

To illustrate why R is preferred for this particular kind of work I asked Claude to generate a fair comparison of how the two languages do or could handle the sustainable performance analysis that is fully described, with code snippets, on this website here. Since the Polars library has substantially enhanced Python’s ability to handle dataframes, I asked Claude to give specific attention to where this could be beneficial.

Claude generated this comparison. As you can see if you read it, Claude concludes that:

1. Both R and Python are fully capable of handling the analysis.

2. With Polars, Python could complete the computation more quickly.

3. The code in R is more readable, concise and elegant than in Python.

Since this analysis is conducted occasionally and is offline, the third factor is more important than the second. In fact, little to no heavy-duty processing in Crickles occurs at run-time and so R usually wins for the same reason.

As it happens, the kinds of machine learning (ML) model used by Crickles are better supported in R whereas different types of ML model that could (but aren’t) used have much better support in Python; but this isn’t a factor in the analysis of sustainable performance analysed by Claude.

In Claude’s paper the R code is my own whereas the Python code is generated by Claude.

One frequent issue is that a sports watch can take several minutes to establish an adequate skin contact and/or for perfusion to stabilise so that there is a good optical signal that the watch can process. Then, a typical heart rate chart might look like this:

Here, after an initial quick rise at the start, it is most probable that the athlete’s heart rose steadily to a plateau above 150 bpm. It is very unlikely that the heart rate suddenly spiked at around 16 minutes, as recorded by the watch (and the athlete whose run this was does not believe that it did).

Crickles now detects such patterns and assigns them a Regularity value of OHM_jump. It also locates the time when the jump occurred. This is most clearly visible as the dotted vertical line on the heart rate chart, as shown in the image above, but it can also be seen on the Activities page, along with the time at which the OHM becomes reliable, which is called Sensor_Warmup (16M 7S in this example). It can also be seen on the Timeline page, where OHM_jumps are indicated with a deep blue dot in the middle of affected activities:

To see this, where is is applicable, for activities that have already been analysed you will probably need to refresh your data.

If you see this pattern frequently you can try one of the commonly cited recommended fixes such as wearing your watch a notch tighter and above the bony bump on your wrist and warming up before you start your activity. In any case aggregate heart rate metrics for such activities will be wrong insofar as the heart rate before the jump is understated. Crickles will adjust Cardiac Stress Score (CSS) modelling to account for this, when detected, in the next revision.

Power data can, with suitable equipment, be obtained for a range of types of activity including cycling, running and indoor rowing. However, power numbers for, say, running and cycling are not commensurable and Crickles thus only shows power data for Rides and Virtual Rides on the Effort Spots and Sustainable Levels pages.

Until now, the Sustainable Levels page did not show a sustainable heart rate chart when a sustainable power curve could not also be derived. This meant that the page did not appear for non-cyclists or for cyclists without power meters even if they had rich heart rate data. This has now been changed so that the Sustainable Levels page shows a heart rate chart and/or a power chart whenever either or both can be derived.

If this problem affects you the quick fix is to refresh your data as described here. You just need to hit the Confirm Refresh button; there is no need to select the remove cached stream files option.

If you have a persistent problem please get in touch.

As you will have seen on the Get Started page, Strava does not allow Intervals to pass activities onto other apps such as Crickles so it’s essential that you connect your Intervals account directly to your Garmin, Wahoo, Zwift and other devices. However, many of us use Strava as a social network for sports and other apps for personal analysis. In this common use pattern it is natural to enter meaningful activity names only on Strava. This means that you either see the default/wrong names in Crickles or you have to update the names once on Strava and again on Intervals or your device(s).

Fortunately, Intervals offers a hack. If you connect directly to your device(s) and to Strava, Intervals will spot when activities loaded directly and from Strava are the same. In this case, it will pass to Crickles most of the details read directly from your device, which is typically a richer dataset than Strava collects, and pass through just the name from Strava in place of the activity name from your device. For example, if you take a bike ride you might find that your Garmin assigns it a default name of, say, Islington Road Cycling while Strava calls it Morning Ride. If you connect Intervals only to your devices, it is the name from there - in this case Islington Road Cycling - that comes through to Crickles. If you are in the habit of giving descriptive names to the activities on your device so that you can, for example, make sense of them in Garmin Connect, then as you add these descriptive names Crickles will get them too. However, if your preference is to leave the device names alone and update only the Strava names these will be passed to Crickles - in preference to the device names - if you also set up a Strava connection in Intervals. The settings you need in Intervals to make this happen are these:

If you have not connected to Strava before and do so now you will find that the names you in Crickles of activities that you’ve done already will update to the Strava names, and any new activity names will come from Strava going forward. It is not possible to mix and match, with some activity names sourced from Garmin and others from Strava.

Note that in no case is Intervals permitted to pass on full activity details from Strava to Crickles. If in Intervals you connect your Garmin and also Strava and you have some activities on Strava that are not on Garmin or on any other device connected directly to Intervals then Crickles will not be able to see them. Likewise, if you download all of your old data from Strava to Intervals but do not also collect the same data directly from your devices then these old activities will not feed through to Crickles either.

Many thanks to David from Intervals for passing on this tip.

The research

Since its first release Crickles has had a measure called Regularity that is calculated on all activities with heart rate data. This measure serves two purposes:

1. In some cases the heart rate data indicates an obvious recording error. In such cases Regularity is reported as Check_Strap. Sometimes this can be fixed by simply changing the battery in the heart rate strap, if it’s a chest strap, or sometimes the strap itself may need replacing.
2. In other cases there are jumps in the heart rate readings that are not obviously just errors and that are classified as Irregular or Unclear.

Some years ago we found from analysis of data of users who also completed a short heart health survey that a high occurrence of irregular readings in cyclists was correlated to a very high degree of statistical significance with heart rhythym issues such as atrial fibrillation. Mark, the Crickles cardiologist, has observed that this is consistent with the fact that ectopic beats, though not intrinsically worrisome, can sometimes presage the onset of arrhythmia.

A full description of this methodology and the findings is now available on the Research menu of this website here.

The technique for flagging irregularity described in this paper, which is used in Crickles, is a method that is extremely cheap to compute. Subsequent to this paper, the post-graduate mathematicians Jack Buckingham, Andrew Nugent and Yi Ting Loo at Warwick University analysed more sophisticated approaches to irregularity measurement. Their research corroborated the association between irregularity and arrhythmia. While they didn’t improve upon the predictability of the Crickles measure, they did lay out the matter extremely elegantly for future researchers. This paper is now also available on the Research menu here.

If it is indeed feasible for cyclists to get a pre-symptomatic warning of a potential propensity to arrhythmia obtained entirely from rides that they are already uploading to Crickles this could be a valuable pormpt to see a GP or cardiologist.

It goes without saying that the regularity indicator is not a medical diagnosis and an absence of irregular readings is certainly not a clean bill of cardiac health nor a guarantee of reduced risk.

Improvement to the performance of various aspects of the Navigator is continual and I hope that you will find a trend towards ever snappier experience over the coming weeks and months. If you have feedback on any aspect of system performance, especially if you are fairly sure that it’s not an issue with your own device, please get in touch.

Why Fitness, Fatigue and Form usually miss the mark

The classic version of Crickles had a FitFat page, showing Fitness, Fatigue and Form curves, as well as a related Fitness Trend page that showed smoothed annual fitness curves overlaid for each year of available history. These were amongst Crickles’ most visited pages and addressed one of the main problems with implementations in other apps: consistency of measurement. For example, for any activity where the user is not wearing a heart rate monitor or using a power meter, apps such as Strava score the Fitness and Fatigue impact of the session as zero. Crickles, by contrast, uses machine learning models to estimate cardiac stress (CSS) in the absence of heart rate data and CSS is the consistent metric from which Fitness, Fatigue and Form are calculated.

However, this still does not address all of the problems with Fitness, Fatigue and Form metrics. These are:

1. It is assumed that all people respond to exercise in the same way over the same timeframe, which is very questionable and not convincingly shown in any robust studies.
2. There is no way to test the predictive value of the fitness, fatigue and form values shown in most apps.
3. If an app user believes that their response to exercise is different from the model assumptions there is no way for them to personalise the calculations to reflect this belief.

Without addressing these issues, while a smoothed Fitness curve, such as the old Crickles Fitness Trend, might indicate the general shape of fitness over a long period, the accuracy of fitness, fatigue and form metrics and their utility for predicting peaks of form is absent as they are not trained on the user’s data.

Given that all of the other methodologies in Crickles are solidly data-driven, we have held off re-introducing FitFat into the new version of Crickles until these problems were addressed. Now they are.

Ian

For Navigator users this service has the added benefit that if there is ever a technical issue, such as the downtime this week, you can instantly be alerted with status updates - for example, if the Navigator was unavailable and is now back online.

I don’t know if many people still use RSS readers such as NetNewsWire but if you do a consequence of this change is that you’ll be able to get your Crickles content through your RSS app. To do this you’ll need to supply our feed URL, which is https://crickles.org/index.xml.

As ever, if you experience issues with the subscription service and follow.it can’t help you, do get in touch.

Ian

As we observe in the paper, this illustrates that sporadic mis-reporting of very high values by the sports devices used to record heart rates is not especially uncommon. As the same paper shows, Crickles identifies some forms of suspicious value in the Regularity field for each activity with heart rate data. As the paper also explains, we surmise that even where these high heart rate values are erroneous they may nonetheless be informational. Furthermore, given the frequency of such dubious readings, Crickles analysis does not rely on maximum heart rates for setting zones or any other purpose, preferring instead to use a measure of maximum sustainable heart rate that is much more robust; this is explained here. So for Crickles, the weird high values are of value and not a problem.

Intervals.icu faces the same issue and, presumably for this reason, filters out by default all heart rate values that exceed the maximum heart rate level calculated or entered for each user.

Until recently, the detailed heart rate data that Crickles obtained from intervals.icu had been subject to the same filtering. For most activities this makes little or no difference. However, for unusually high heart rates the filtering can have two unwanted effects. First, if the high rates are legitimate and reflect a break-out effort then removing them will suppress the Crickles estimate of maximum sustainable heart rate, and this could distort the analysis of subsequent efforts. Secondly, filtering normalises some activities that Crickles would otherwise identify as Irregular. Although Crickles does not currently show an analysis of user irregularity patterns, we intend to re-introduce it. This leverages the analysis from the detecting arrhythmia paper to give users an indication of how they compare to other users on this measure, which is strongly correlated with atrial fibrillation. If the heart rates are filtered at the high end this will be wrong.

Should you do so this will delete Crickles’ record of data obtained from intervals.icu. You can temporarily carry on using the Navigator but the next time you refresh the page or prompt a data reload, Crickles will re-fetch your activities from intervals.icu. This may take a couple of minutes.

You also have the option to delete Crickles’ cache of your detailed activity data. This is not required if you’re just checking that Crickles has your full, correct set of activities. If you do choose this option the next refresh will take about as long as your first visit to the new version of Crickles.

Any data that Crickles was able to retain from a Crickles classic account will not remain in place and will not need to be reloaded.

As ever, let me know if you run into any problems with this feature or have any questions.

A second purpose is to provide a site where methodology descriptions, occasional news and the Navigator can be seen together. Methodology descriptions and underlying research are grouped under the Research heading on the menu bar. At the time of writing, there are two papers there:

1. A detailed explanation of how Crickles calculates sustainable heart rate.
2. Our analysis of the relationship between the frequency or irregular activities and a diagnosis of a heart rhythm issue such as atrial fibrillation.

It is our intention to build out much more such content. To guide us, we are still keen to get feedback on how comprehensible, interesting and possibly even useful this is. The derivation of sustainable heart rate is relatively straightforward in comparison to the models used to determine heart rate intensity and cardiac stress. We want to ensure that the exposition of the material matches the nature of interest from the Crickles community and we can expect to have to learn how we go to get this right.

Meanwhile, for the time being other content remains on the old home page, which you can still find here.