Saturday, July 30, 2011

Strava urban KOM matching

On July 12 I rode up Kansas Street on Potrero Hill in San Francisco, as someone else asked me to generate data for a segment there since my HTC Incredible phone has relatively good accuracy here. So I did that, and after defining the segment, I had the KOM. Not too surprising, since typically if any other Strava users had made a hard effort between the same two end-points they probably would have created the segment earlier. Here's that ride. As you can see, I was lucky to have beaten out Steve Smith's effort from May. He'd done a San Francisco climb-fest and had done Kansas only one second slower.

The issue on that ride was in addition to the Kansas climb Strava also gave me credit for Rhode Island, one block over. I had two KOMs, one block apart, from the same ride.

So yesterday since I ended up at the 4th and King Caltrain with my small backpack and no laptop, I decided to try and validate my Rhode Island "KOM". I wasn't quite sure where the end-points were (the embedded maps don't render on my phone), but I got up to speed where the climb started mid-way between 17th and 18th. This was my first mistake: the climb was defined to start right at the corner of 17th. Then at the intersection of 18th I stopped to wave a waiting car past. This cost me a bit more time. But the rest of the way was smooth going. I went all the way to Southern Heights, beyond which Rhode Island steeply descends.

I felt it was a good effort, and had a good chance of taking some time off that faux-KOM I'd set earlier. So when I got home I uploaded the data from my phone, went to the laptop, and checked the result.

I was 3 seconds short. But what I had done is "broken" my standing KOM up Kansas, and in fact set a KOM on De Haro, one block in the other direction. So instead of one bogus KOM, on Rhode Island, and a legitimate KOM on Kansas I now had three bogus KOM's: Kansas, Rhode Island, and De Haro.

Strava obviously needs to tighten its segment matching criteria, especially for urban segments where streets are densely located. Some areas, like the top of Old La Honda Road, have signal reflection problems where relatively large errors can result. But you can't set the matching threshold everywhere based on these worst-case locations.

Anyway, now I need to go do a hard ride up De Haro, as well. Fortunately Wisconsin, one block further, doesn't go through so I know there isn't a segment there I can contaminate.

Of course, the only reason I have KOM's on these climbs, valid or otherwise, is they're sufficiently obscure that none of the significant number of faster riders around here have bothered. But that's part of the beauty of Strava: it taps into the animal instinct to mark territory. There's more territory than the top dog can pee on, leaving space for even lesser dogs like me.

appendix: after posting this I went out and rode hard up De Haro. This fixed my bogus De Haro KOM, but also registered a new KOM up Rhode Island. So I think for these segments to be distinctly meaningful Strava simply needs to fix its matching algorithm.

No comments: