Another run across America underway - is it legit? Boots currently on the ground

You are making the mistake of using Strava time, which is only moving time and doesn't include any rest periods. My time is total cumulative time, which includes everything. 6.16 hours total cumulative time is when 30 miles starts for PJ

Use distance on x-axis not time. No cadence + distance equals OOPSIE. Night all.

I won't be back with more data until Sunday, I have a life too.

In the meantime, all I'll say is it's obvious Sneakers you are technically incompetent or operating in bad faith. In post #3548 you said that I need to filter out the data points where PJ isn't moving and is resting. Now your excuse is that your data includes rest periods so mine must be wrong. Every time I, or anyone else, says something, you just flip the script or distract with some weird incoherent explanation. You've made multiple technical errors.

I'll leave this group by saying, I will happily look at another run on Sunday. But I caution making any conclusions from any of the data. Nothing in the data will explain WHY there is no cadence for such long parts of the run. I strongly caution y'all to continue to not listen to anything Sneakers says. I am actually on the side of PJ, I don't think we can draw any conclusions against him for now, but Sneakers doesn't know what he's doing.

@weatherman, yes exactly like I did, distance on the x-axis makes more sense than whatever Sneakers is attempting.

I took the the last 15 minutes of my work day yesterday to fiddle with two runs' data too. One was PJ's March 18 and the other was his Western States 100.

I calculated instantaneous speed (delta distance/delta time) and counted up the percentage of run featuring zero cadence AND instantaneous speed > 0. I'm away from my computer now, but Western States featured approximately 11% of the total run having zero cadence and a non-zero instantaneous moving velocity. March 18th was about 25%.

I'd like to compare this to other long runs of PJs as well as other transconners' runs. Also, probably worth looking into a longer rolling average like 10 or 20s vs instantaneous velocity. And also bring HR into the mix.

I don't believe Strava's cadence and position values are different from your .fit files. They are indeed different on the UI in the app that everyone keeps talking about, we see the plot is clearly smoothed. But I believe that's all front end. I am still very interested in the .fit.

I work in scientific research software, so I know what analysis is, although I do not do analysis myself.

Very correct.

Literally just spent an entire afternoon with a PhD going through HDBScan so I can make an interface for it. I may have misused some terms, but they were more than happy with my understanding of it. HBDScan does not statistically prove or disprove anything, but, like you said later, it tells a visual story, which is my job. Please tell me what I got wrong regarding my understanding of clustering.

I'm not here to prove my data analysis superiority, because I know I don't have it. I do software. But I'm here to point out that Sneakers is borderline illiterate with actual software and data.

I believe this to be correct, but everyone automatically disagrees with Sneakers. Probably because he's said a million nonsensical things.

I can't flex my data analysis skills, but I can flex my software skills. Why would Strava "simplify" or "smooth" this data the API hands to the front end, only for it to have 50,000+ points? That would be the least-efficient simplification I've ever seen. If you look at the data, it's also definitely not very smooth.

What I believe to be happening is this:

* Your watch records "raw" data, all this is a bunch of black box functions. We have to take them at face value unless you want to go speak to a bunch of electrical engineers and embedded device software folks to understand the inner workings of an accelerometer and the different modes and "learning" methods it has.
* Your watch packages this as a .fit which includes key-value pairings for things like time-cadence, time-distance, and time-HR.

* Your Garmin app hands this data to Strava through some API that you authenticated
* Strava takes some of this, not all. We know some of these watches derive values like vertical oscillation or ground contact time from accelerometer data. Strava just ignores some of these values

* Strava passes these values as they have them to the web and mobile applications when asked. I'm guessing the mobile application has a different endpoint and might actually receive a simplified dataset in an effort to be bandwidth-conscious.

* The web application takes this data, runs it through some front end black box. Actually not a black box by definition since it's on the front end. Might be obfuscated though! I haven't checked. It uses this output to show you the plots you see.

Not sure why all on this board keep arguing about a run that is no longer significant from an athletic achievement standpoint. This is "LetsRun" after all not "LetsPlotData" or "LetsHateononRunfluencers". This is just another one of the 1000 + runs across the US that have taken place (despite PJ using the same 400 number on his website that WG used). If you aren't claiming a record riding in a car at some point is not a big deal even. Crossers do it for various reasons, especially those that are self supported. Escaping dangerous weather is usually the main reason. The $ raised so far is impressive. 1800 miles is still a long way. Nothing more to see on this one really. The 2025 cycle of those aiming for records or to raise awareness of things people already know about has already started! If you are still on PJ you are already behind.

So you looked at your binning and cross correlations and didn't think about any physiologically weird relations? I wish I knew the bank at which you head data analysis so I know not to invest with it.

Here is my attempt at a summary of the portion of this thread dealing with the current transcon attempt.

The cast:
PJ = Paul Johnson
SOTG = Sneakers on the Ground, a LR poster
People = Every LR poster except SOTG

PJ announced that he would be attempting to break the WR for a transcontinental run.
People looked at PJ's previous results and found no runs that would predict that he'd be able to (even come close to) breaking the record. Some wished him well, some insulted him for his hubris, and some reached out to PJ and his crew to suggest that he might want to set a more modest goal.

PJ started his journey and found 70-mile days to be very difficult. He then changed his stated goal from 40 days (a WR) to 50 days (still very impressive).

People looked at PJ's Strava activities and saw stretches that had no cadence data. They concluded that this could be evidence of cheating (riding in the van for portions of the mileage).

SOTG responded that the use of trekking poles can cause gaps in cadence data, and provided proof in the form of Garmin forums about that issue.

People responded that those others' cadence-data gaps are small (seconds at a time) whereas PJ's gaps are large (miles at a time).

SOTG responded that the gaps only appear large because of how people are analyzing the data on Strava.

People responded that the Garmin data might be more robust, and that releasing it to be analyzed could either vindicate PJ or prove him guilty.

PJ's crew did not release the Garmin data; they have claimed that there's nothing in the Garmin data that isn't in the Strava data.

People asked why, then, they won't just release the Garmin data, and have said that the only reason not to do so is if PJ is cheating.

SOTG defended PJ's crew, saying that they should not release the Garmin data because:
1) those asking for it are not doing so in good faith, and
2) those asking for it would not know how to analyze it correctly.

Since then, SOTG has gone back and forth with a rotating cast of others, and no progress has been made. SOTG tells them to prove their ability to analyze data by doing so with the Strava data; if they don't comply, he says his point has been proven; if they do, he says that they haven't proven anything. They respond that they could maybe prove something if they were able to analyze the Garmin data instead of the Strava data, and the cycle repeats.

I am genuinely impressed with your summary. I think you captured the gist of everything succinctly

You forgot the regular schooling of Sneakers by a succession of people.

I wish that were the case, then I wouldn't need to pay for tuition!

Ha, bravo 👏

Who is planning to chase records in 2025?

I'm not sure, but it looks like this thread has years left to go. I'm looking forward to the 30th year anniversary when we all sit down together. Will C presents trophies for best cheat, Goodge sings karaoke but has to mime it, PJ ironically arrives on his mobility scooter, Sneakers tells us we can't read the dinner menu properly and orders the beers for us as he's the only person suitable and we all live happily ever after!

Update: Paul with a probably brutal 50 miles today. All downhill, most 16min aka walking. Is it over ??

Complete with a lengthy cadence drop out... going downhill. Wonder if he was using poles on the gentle, paved descent. That would be a bad sign.