Did Froome climb faster than Armstrong? The answer… at last

The story so far. On Saturday following Chris Froome’s win on stage 8 of the Tour de France the internet came alive with claims that Froome’s time up the Ax-3 Domaines climb was the third fastest time up that hill, just slower than Lance Armstrong in 2001 and quicker than Lance’s time in 2003.

If that’s true it’s an interesting statistic. Either you can say “wow Froome just slower than an admitted doper, how good must he be?” or you can say “wow Froome was faster than Lance Armstrong! Really? Lance was doping”.

However I wanted to know how these times were calculated, where did these “facts” come from? And are we comparing like with like? So I asked (see that here).

And in the 24 hours after asking that question I learned a lot, not about how the times are calculated but how merely asking for further information didn’t go down too well with others (see that here).

And now I have even more, including whether Chris Froome really did climb Ax-3 faster than Lance Armstrong.

To summarise where I’d got to: a Twitter user called @ammattipyoraily was seen by many as the go to person for timing information, indeed it seemed she/he was the source of many of the timings out there as “facts”.

Again to be really clear: this is not about me wanting to prove or disprove the cleanliness of a rider. This is nothing to do with power numbers and this is not about knocking anyone’s theories or scientific work. It’s purely about understanding how these times for climbs are calculated, especially the historical ones. For the sake of total simplicity the rest of this article will only focus on one climb, Ax-3 Domaines and the timings associated with that.

And now I’m starting to get some answers…

Do you want an easy one first? The climb times shown on some TV feeds of the Tour de France are derived from GPS signals from the lead motorbike filming the riders. The same GPS signals are used to determine the gaps between groups and that ‘distance to go sign’ up in the top left corner of the screen. Great as a guide but if the moto is with Quintana on the start of the climb and goes to the top with the attacking Froome who overtakes Quintana it doesn’t give Froome’s actual time.

Now a harder one that I’ve just got my head around…

Our friend @ammattipyoraily kindly got in touch to share the resources they use to calculate a ride’s length, altitude and a rider’s power output. But when asked to clarify the reference points used to compare rides across different years or the source of his historic data they did not reply. That was the second time they have ignored a direct and clear question about the subject. Now that is @ammattipyoraily’s prerogative but when someone so proud of sharing the numbers they have calculated won’t share their methods I get suspicious.

So I trawled the Internet to try and find where these numbers have come from for the Ax-3 climb and guess what I found? The earliest reference I can uncover for them comes from a forum post in June by a user with a name the same as claimed by @ammattipyoraily on Twitter namely Vetooo. Although the forum is in a foreign language the thread title contains a word very similar to ammattipyoraily. So she/he still appears to be the primary source of information.

The next part of my theory is speculation but go with it because it’s not a massive jump in belief. The original forum post with the times for Ax-3 came just days after the publication of the ‘Not Normal’ or ‘Doping or Not’ literature. I was intrigued. Given the substance of those publications was pretty much all about how fast riders went up certain Tour de France climbs was it possible that the timings being bandied about had come from ‘Not Normal’? Seemed like an idea to me.

Now I know this next part is not good investigative practice but I refuse to a) pay ten dollars to find out Lance Armstrong doped and b) give my credit card details to a website so badly designed. So I carried on investigating on the cheap.

To quote the (free!) preview extract ‘Not Normal’: “we present our radars for the 100th edition of the Tour. There are six… You’ll be able to judge interactively, as a simple spectator on your couch, with a stopwatch… You’ll easily be able to calculate possible levels of cheating”. Sure sounds like they have published a list of times for climbs if you can rate how suspect one is with a stopwatch. Oh and Ax-3 Domaines is one of the six “radars”. In fact the six radar climbs are the exact same six climbs that Vetooo/ammattipyoraily has posted times and data for on that forum.

So if I accept it’s highly possible that the source of climbing times is the ‘Not Normal’ team then how do they calculate the times? Luckily, they tell you “How do we get these numbers? Just like you, sitting in front of our television. We start our stopwatch at a pre-determined point… We stop it at the summit”.

In the interests of rigour (are you watching scientists?) the source of these quotes can be found here.

Almost there! If the primary source of these “accepted” climbing times is ‘Not Normal’ then it would seem the technique for getting the climbing times is

1. Find a common point at the bottom of the climb
2. Find a common point at the summit of the climb
3. Run a stopwatch on videos of riders doing the climb

Why couldn’t someone just have said this in the first place?

I fully realise this isn’t a great historic scientific breakthrough but I’d set out on a mission to find out how these “facts” had been calculated and now I was getting somewhere.

Thanks to some kind souls on Twitter I armed myself with Youtube videos of the climb of Ax-3 Domaines from 2001 (to time Armstrong), 2003 (Lance again) and 2013 (Froome). A quick look through these got me some start and summit reference points, I got a stopwatch, I was ready to go.

First up 2001. Oh. The official coverage of that year’s ascent doesn’t show Armstrong on the climb until he’s well on it. I checked and checked again but he’s not shown anywhere starting the climb. In fact the time from Lance’s first appearance on screen to him hitting the finish is shorter than the time bandied about (apparent source Not Normal who sit in front of a TV with a stopwatch). In fact if there’s any correct and factual time for that climb it must have been made on the day and on the ground because its impossible to accurately get his time for the climb from the TV coverage.

So if Not Normal or Vetooo/ammattipyoraily quote a time for Lance Armstrong’s climb up Ax-3 in 2001it’s pretty likely to be an estimate and nothing more. And that was that. If nothing else I’d succeeded in showing these numbers described as facts are the total opposite. Guesswork, hunches and imagination.

And then….

I decided to keep going and compare 2003 and 2013. First up 2003. My reference point for the start of the climb (my reference and mine alone) was the 90 degree left turn at the end of the bridge.

Here’s Lance at that point. It’s easier to see in motion but he’s in yellow.


As he turns left the time on the video is 1:23

I chose the finish line as my end point, it’s not the end of the climb but it’s very easy to identify and appears on the videos to have the same block of appartments to the right so it seems to be a constant.

Here’s Lance crossing the line


As he crosses the line the video time is 25:23, so from the bridge to the finish took Lance 24 minutes exactly.

Now 2013. Chris Froome on the left after the bridge.


The time on the video is 24:15

And here he is as he crosses the line


Time on the video 48:00. Meaning Froome took 23 minutes and 45 seconds. He did do it faster than Armstrong!

I’m extremely confident in my timings. I have no doubts of the time between the two points. I have no idea if the route between them has changed, what the wind was like or any of a thousand variables but I am confident that Froome did climb Ax-3 Domaines 1% faster than Lance Armstrong in 2003.

As an aside: in my research I’ve seen times quoted for riders up the climb in 2013 who didn’t appear on TV. Again these figures must be guesswork.

There are now wider discussions to be had about the context of these climbs and how relevant the comparison is. They’re not for me, I’ve done enough. We are told the sign of a less doped peloton will be slower times in a race. My view is that can’t be applied to one climb and long term trends need to be taken in consideration.

1. Many of the times of climbs are total guesswork, there is no footage available that comprehensively shows every rider passing two fixed points every time a climb has been tackled in recent years.
2. The rankings, such as Chris Froome has the third quickest time on Ax-3, are based on these times so are therefore total bull
3. Very few people who quote these times know how they are sourced and have done little if anything to verify them. Therefore any other “facts” they quote should be treated with suspicion.
4. Chris Froome did climb Ax-3 Domaines faster than Lance Armstrong

UPDATE: @ammattipyoraily has confirmed via Twitter that they are the person who has posted timings on the forum mentioned. They also contest that the timings are their own from their personal video collection.

I have asked for a screenshot of Lance Armstrong starting the climb in 2001. This has yet to be provided. Further, @ammattipyoraily tweeted in June that they had ordered the Not Normal book

UPDATE 2: @ammattipyoraily has admitted time he’s been quoting for Armstrong on the climb in 2001 is an estimate

41,166 total views, 52 views today


  1. dimspace

    Good work on that.
    So now we can conclude from your figures which are backed up by screengrabs that Froome was up to a minute faster in 2013 than Armstrong was in 2003.
    I have established using video and have put my references on twitter that Armstrong was around 1 minute quicker in 2001 than Froome did in 2013 based purely on the climb of around 8km.
    So we have something like
    2001 Armstrong –
    2013 Froome +40 seconds
    2003 Armstrong +1m40 seconds
    But the 2001 figures and 2003 figures are based on different points.

    Of course that contridicts the figures we had for armstrong in 01 v 03.

    In the end what we know, is what youve pointed at, to establish a time you must personally get both times yourself so you can say for sure you have used the same points.
    You cannot use someone elses figures for 2001, and without knowing the start and end points come up with a 2013 figure. Unless you guess, and come up with a time that fits the story you want to tell.

  2. wiggles

    Your methodology should be sound as the video is a live stream and the only error would be estimating the time at which the rider crosses your reference points. Maybe just a couple seconds max.
    Too bad the UCI or ASO don’t publish the times as I bet they either know them or would have easy access to estimate them.

  3. Paul Jakma

    Good article. Though, it would have been nice to have been mentioned, given I did the same work and tweeted my results and methodology, the replication of which seems to form the basis for this article. E.g.:

    I’ve just timed Froome, Ulrich & Armstrong for the 2013 and 2003 Ax-3 Domanes summit finishes…

    Time taken as starting from going through the right-hand turn after coming over the bridge – final common ref before climb..

    I have Froome @ 23m31s, Ulrich @ 23m38s, Armstrong @ 23m45s. Froome finishes 7s faster than Ulrich.

    Armstrong finishes 54s up on Valverde, who did it 24m39s.

    The ’13 riders would have kept momentum & gained a couple of secs, as in ’03 they were taking off helmets before the start of climb signs.

    Vids used were https://www.youtube.com/watch?v=EO1RlYxVg9k … and https://www.youtube.com/watch?v=-TjX73Tf1hk … . I have Armstrong start @ 1:38, fin 25:23, Froome start 24:29 fin 48:00

    I think the times are accurate to ± 2s at worst.

    Not sure what the wind was like in ’03 v ’13. Some suggestion ’03 was windless, and ’13 slight variable headwind.

    There’s other factors of course, more fatigued in ’03. Though, fatigued dopers still seem able to easily beat all but one of the ’13 times…

    @JournalVelo Yes, it’s same place or pretty close to. a) there’s a series of easy to identify bends immediately before and

    @JournalVelo b) there’s a distinctive apartment building to the right. There can’t be more metres difference, if there’s any.

    At least one of the videos I used was quite a random pick – certainly not the first. And there are *lots* of videos of these climbs on Yt in various different languages.

    Also, it should be noted Vettoo has Armstrong at a faster time in ’01. I was just interested in replicating at least one of his climb times and Ulrich’s in ’03 (3rd fastest) was easiest.

    • JournalVelo

      Thank you for your comment.

      As I pointed out last night I had already undertaken the majority of my work when you sent those tweets.

      My work is different to yours hence why my figures are different to yours and I have focused on fewer riders.

      At the very most I used screenshots from the video links you provided from my Youtube history.

      I understand you would like an acknowledgement but it’s simply not due as your work had no bearing on my results

      However I do appreciate you being part of the debate.

  4. Paul Jakma

    Also, it’s worth noting that, even if the TV does not show the precise time a rider passed a reference point, you can still determine a “can’t be any later than” time, and often make a very good estimate, to within a few seconds, with a known upper-bound on worst-case error.

    Even if those estimates ended up out by 10s (a huge amount), over 10 minutes that’s just 1.6% error, and 0.8% over 20 minutes. That’s comparable and even better than the accuracy of good power meters (2 to 1%).

    You shouldn’t be railing about using TV to estimate times – there’s no problem there. Rather, the problem is people not stating their methodology and/or not stating their accuracy. Hence why in my tweets I gave ±2s estimate for accuracy!

  5. Paul Jakma

    Dimspace: Of course the figures contradict – there are measured from different places. These figures include the time taken through part of the town, before the climb proper. The riders in ’03 would have lost a couple of seconds here as they were nearly all quickly fiddling with helmets and handing them to soigneurs. The roads and 2 turns before the climb were more cluttered with those helpers as well as spectators in ’03, so possibly another second lost there. Hell, even with the same video, 2 different people might read a time 1s different past the same ref, never mind if they use different ref points.

    Finding that the times don’t match exactly is no surprise. What matters is the gaps and the error. Even a couple of secs difference over a 20 minute climb is simply insignificant.

    There is no reason at this time to think the Vettoo data is inaccurate. Is it going to be *exact*? No. No physical measurement ever is. Is it sufficiently accurate for the purposes it’s being used. Sure.

    The biggest source of error in the climb power estimations has very likely little to do with the TV derived times.

  6. Paul Jakma

    Sorry, a couple of secs difference in the absolute time. A couple of secs in the /gap/ would be far more significant of course. However, generally, the gap times aren’t that critical – certainly not to the power estimates.

  7. Col

    Are we sure that each rider is doing the climb ‘flat-out- or (‘full gas’ in current parlance?). Froome was clearly in TT mode toward the end of the climb, but as the individuals cited weren’t racing each other, how can straight time comparisons be of any scientific value? Froome may have eased for a while between points x&y; Pharmstrong may have gone 100% for just points b&c. Who knows? Who cares?

  8. ijsbrand

    Nice work. But how sure are you that the finish was at the exact same spot? After all finishes are just lines, often arbitrarily drawn somewhere.

    • JournalVelo

      Not sure no. As you say they are just lines but in some longer shots both lines appear to be next to the same building of apartments.

      Again that uncertainty could make times longer or shorter.

  9. kcr

    Should point 2 not read “find a common point at the summit of the climb”?

    Comparing times to identify doping does not seem very credible. What about headwind/tailwind, how hard did they ride the rest of the stage?, etc, etc Too many variables to make any sensible conclusions.

    • JournalVelo

      It should and it has been changed. Thank you for pointing that out.

      Regarding the context it’s a fair point. For me there are too many variables involved for timings to be anything other than a guide.

  10. Gilbert

    Not sure if this has any effect at all but …

    You seem to be using the youtube playback clock for the timings.

    So you’re not timing the climb you’re timing the how long the video playback takes.

    Given the videos are shot on different equipment using different resolutions possibly different frame rates, and are then transcoded by the youtube servers to prduce the video output you are actually timing, is it not possible that the times are affected?

      • Paul Jakma

        I’ve just checked the 2 videos which I and JournalVelo have used, and the times visible in the video match up with the time I can measure locally, to within a 1s – measured over at least 2 minutes. So whatever timing error, if any, was induced by re-encoding is below 1%.

    • Paul Jakma

      Good point. The resolution wouldn’t matter, but transcoding to different frame-rates might. There’s a trivial way to measure any error introduced though: find 2 points with times encoded into the video (e.g. on screen clocks, or finish line clock), that are at least a few minutes apart, time them with a local watch, compare the difference.

      It’s good to think about all the sources of error. However, we can find ways to deal with them. We *ALWAYS* must in physical measurements. Uncertainty is inherent in any discussion of physical systems, but it’s not a reason to throw up our hands and give up and say “we can never know”.

      • JournalVelo

        Again I’ve never said this has been about the reliability of using TV pictures although there’s plenty come to light to say it isn’t wise.

        This has always been finding out how exact and factual these timings banded about are.

  11. lllludo

    Funny to see how you’re trying to discredit @ammattipyoraily’s measures.
    If you’ve ever been to Ax les Thermes where the bridge is, it’s flat so you’re not measuring the climb. So when you conclude “Chris Froome did climb Ax-3 Domaines faster than Lance Armstrong” it’s not valid from a logical standpoint. Exactly for the same reasons you criticize the finnish guy.
    Nevertheless @ammattipyoraily measured 23’14” for Froome on the climb and 23’24” for Armstrong 2003. This is consistent with your bridge-finish measure. So both you & @ammattipyoraily agree on this one. Froome was faster than Lance and all of the 2003 doped riders.
    For the climb time measures you should refer to Frederic Portoleau. He’s very accurate. Nobody has ever contested his measures. He’s even sometimes doing on site measures. 2 problems though : 1°) he’s French and French don’t like British 2°) he’s the guy doing the measures for Not Normal.
    Another problem I see is regarding his threshold calculation because they don’t know the exact weight of the riders and their bikes. So they estimate … For example he estimates Froome is at 67kg at the moment. It might be more. Or it might be less …

    • JournalVelo

      Thank you for your comment. I’m not trying to discredit anyone’s work. My issue with the timings is their accuracy and that they are published as facts rather than being open and admitting which ones are estimates. It’s dishonest and can have consequences on reputations.

      • Paul Jakma

        The weight is an obvious source of error. Just 1 kg over on 68 kg is a 1.5% error. Wind and drafting are other big ones I guess.

        Errors of even seconds in climb-time estimates are nothing when they’re diluted over 600, 1200 or more of those seconds. 😉

        • JournalVelo

          Once again this wasn’t about power or calculations. This was about how factual are the climb times that were being accepted by many as facts.

  12. Ricky

    Great article. I know your intent wasn’t to debunk the theories of some, but in my mind it did. Given the element out there desperate to wrangle numbers and data together to provide some kind of accusation against Froome of doping I wasn’t shocked to see how quickly they jumped on the times up this climb, but to take the times alone is completely unfair when building some kind of a case. (Not talking about your work here which was to point out that some of the stuff we’re being fed is inaccurate).

    What I want to know from those trying to insuate something against Froome (or any modern rider, though they tend to come from Sky) when comparing Froome’s time and Armstrong’s is, how many climbs there was before, what intensity where they raced at, what was the wind direction, wind speed, the humidity, the temperature, was the pace intense on the lower slopes in one case compared to another where the lower slopes may have been ridden at tempo?

    I know from watching this years effort that they had one major climb before hand. In 2001 they had several big climbs. Also the 2001 stage was later in the Tour and not the first mountain stage. On top of that the pace was intense right from the start of the climb in 2013 though I haven’t watched the 2001 stage to see how they rode into the climb.

    So many variables that lead me to shrug my shoulders when I see the times up an individual climb — at least when it comes to drawing some kind of conclusion. Sure it’s cool to see how fast they can do it, especially if I ever go to ride one of these climbs myself, but it also comes with a big, so what? This whole Watts/kg stuff is nonsense and a wild estimation that unless you’re seeing the reading off of the power meter on Froome’s cranks, can’t be guaranteed as accurate.

    But good work to you on this article.

    • Paul Jakma

      The power meter on Froome’s cranks is also inaccurate!

      Indeed, if you really wanted to measure power *very* accurately over a climb, a crank-based system may even be one of the less accurate ways to do it. Simply because the strain-gauges used to measure torque are noisy and prone to being biased by things like temperature. Also, trying to accurately integrate instantaneous strain measurements is hard – any persistently biased errors can accumulate. This is why *good* power meters are still only accurate to 2% (SRM claim 2% for theirs) or worse.

      A system based on measuring the time across a decent distance could actually be *less* prone to accumulated error. This means such a system could even be *more* accurate than the integration of strain gauge measurements, beyond a certain point. Particularly once you use models which can take aerodynamics into account (air density – calculable from pressure & temp), drive train efficiency (measurable), etc.. into account, which certain models do. If you measured the CdA of the rider in typical poses, and put a pitot tube on their bike to measure air pressure as they cycle, you might be able to work out *quite* accurate power figures.

      There are many ways to measure things. For certain things, the most accurate ways to measure depend greatly on how /long/ you want to measure them over. Over shorter-intervals, an accumulating measure (wheel sensor for speed, strain gauges for stress, etc) may be best, while over longer intervals measuring from an relative to an external reference and calculating may be more accurate (GPS, etc).

      As a simple example, imagine a wheel-sensor calibrated to within a 1mm accuracy, that’s around 0.05% accuracy with a 25c wheel. Highly accurate, right? Except, whichever way it’s off, it’s going to be off in the same way (particularly if it’s under, and wear just increases it further – but lets ignore wear for now). But after 5km, the error is 2.5m, after 10km it’s 5m and that absolute error *keeps* increasing. At 10 km, your GPS may well be give more accurate reading for distance, even though it’s per-measurement accuracy is several metres – rather than the millimetre accuracy of the wheel sensor! But the accuracy of the GPS isn’t affected by the distance travelled!

      I’m not saying the TV-time estimations here are more accurate than the riders’ power-meters, necessarily. However, automatically poo-pooing estimates because they’re based on time-of-climb calculations is not a good idea.

      I’ve not tried to work out how high a climb needs to be before the time-based estimates become more accurate than the power-meters though. 😉

      • JournalVelo

        Again I’ve not dismissed anything based on the fact that the timings are estimated other than the timings themselves or more accurately I’ve dismissed the rankings and comparisons of those timings.

        However I’ll indulge you. You all the work any of your suggestions would entail, all that work, equipment, man hours and you have a nearly accurate measurement of power output.

        Then what?

          • JournalVelo


            Then you get everyone to agree on a level classed as exceptional which won’t be easy.

            Then a rider puts in a ride that’s exceptional.

            So after all that work, negotiation, investment and time…

            Then what?

          • Paul Jakma

            Well, you can at least quantify, statistically, just how exceptional that person’s performance is. Amazingly exceptional performances deserve an added degree of scepticism, the history of the sport tells us. Froome himself acknowledged that in a rest-day interview.

            As Science of Sport has pointed out, Sky have options. They can choose to counter the heightened scepticism with increased transparency. Transparency which they said they would bring when they started the road project. However, Sky have chosen not to be transparent, because they believe transparency would be misinterpreted and be valuable to competitors.

            That was Skys’ choice.

          • Paul Jakma

            I’d rather the sport worked on making it so that time and effort wasn’t needed, by being more transparent with data. That would shift the debate from discussing the errors and inaccuracies of TV/time power estimations, on to being able to discuss the significance of exceptional performances. That’d be a step forward I think.

            I don’t think the debate on doping is ever going away. Secrecy around data only fuels speculation, which surely is worse for the sport.

            The fewer people have the data, the fewer who can credibly argue that it really is *clean* – if that is the case!

          • JournalVelo

            Releasing the data raises two questions.

            1) if you release it publicly what’s to stop some crank declaring a rider must be doping simply because they have decided a number doesn’t look right?

            2) release it to the UCI like the blood passport. They spot a power number they think is exceptional. Then what?

  13. lllludo

    No problem. Every adult knows that twitter is all about uncertainty and I do not think @ammattipyoraily’s posts on climbing times, that will always include an error margins, have consequences on Froome’s reputation.

  14. Pingback: Doping in de wielersport - Pagina 13
  15. Chris

    It would seem that just as many if not more folks here are intent on proving Froome’s innocence rather than guilt. Some other questions are: is his bike lighter than Armstrong’s was? Couldn’t that give him a better time? What the heck was he given in the feed bag? How can he accelerate so quickly sitting in the saddle when seemingly everyone else has to stand? I’d love to believe its just sheer training superiority but if so, tell us what it is?