Of course Brian is a religious zealot but the 100 option should have been resolved as YES quite a while ago. It just seems @TimothyJohnson5c16 is not active on the platform and we'll need the mods to assist.

@MaxA Poisson distribution is a religion ? Actual recorded observation limitations of the crowdsourcing is a religion ?

489 registered in Texas. Also, note that 248 distinct robotaxi have been observed in Austin not the 200 observed in 90 days. Based on the number of observations some of the vehicles are likely driving and then not observed based on the low observation rate.

@brianwang @MaxA Max is defending you here, and just having some fun poking at the perceived Tesla Evangelism.

To the later point, it's got to be pretty tough on the mods at the moment. There are quite a few unresolved markets due to inactive traders(or alternatively overly busy traders who forget about one of their 100 markets) I wonder if there is some community based solution where traders might be able to resolve a market passed a certain point without mod intervention. Though the exact mechanism to do that is it's own major challenge

@ShaneBo have Ai or The system deterkine if abandonment has occurred. Send out the notice to the old mod and the community that a list of markets will get into the conversion process. Only say a week without active action by original mod. It goes to new resolution. Which is AI does a first pass at assessing the. Resolution state. People involved can make their cases for and against. The Ai gets a second prompt including the added cases. And rthen it goes to final resolution. Human mods can also volunteer to take over in some kind of process

If for the next 90 days there were 1000 vehicles operating. By sampling 1256 rides, we expect to spot 715 unique plates. Without increasing sampling, there could be 10,000 cars and sampling 1256, we could not find more than 1256 unique plates. We have to use poisson statistical modeling to try to recover the data when we are using a sampling source.

@brianwang I love the use of "we" where any serious person would've said "my clanker".

But if some cars are only half as active or one quarter as active then the number of active vehicles is higher. 420 based on the assumption of equal usage.

Last 30 days, tracker recorded 576 Tesla rides tracked. There are likely 20-40 rides per day being given per vehicle. 20 times 3 miles for 60 miles per day. 576 rides out of 314 vehicles is 1.7 rides per vehicle. BUT it is 0 to 5 rides per vehicle from the samples. I claim the 188 other vehicles are there and operating but are not being sampled.

@brianwang You want the 500 resolved when you only get to 502 as a maximum???

Surely Tesla register some vehicles for usage without actually "actively operating" them, just as a reserve pool in case demand increases more than expected?

I suggest waiting for more data before resolving the 100 market which I think will probably need to resolve yes but I want to see more data to be sure.

@ChristopherRandles I think there are 520. 479 registered in Texas and 40-50 in Florida. The low sampling rate is adding 1-2 per day in Tampa. That indicates to me that a batch of 20 or so got added to 20 already there in Tampa. I am not resolving now. BUT poisson distribution and the samplng interpretation is what the data is telling us. It does mean we CANNOT resolve NO to 500 anymore. We should have resolved YES to 100 long ago. Poisson modeling of Florida and other states without registered count when summed with registered vehicles is a valid estimate. The 314 raw count is now precisely proven to be low equal usage approximation gives 420. we know not equal usage because about 50 added in the last week which means only 5 might show up at the sampling rate. the maybe some are not being used theory means we have upper and lower bounds. Lower bound is now 440 420 plus guaranteed missing of many new vehicles against some new vehicles not used yet. You are saying 0 use does not count but 1-10 per day is tough to see. Statistical range likely is 440-502. Actual practical upper bound is 520-530 .Worst case lower bound is 420. One month after registration it should be assumed the vehicle went into service. 5% of registered vehicles sitting as bench warmers is the most that I would accept as a logical approximation. IF we were counting the safety driver cars in SF Bay Area the tracker count is 700 BUT 9 months ago they registered 1000 vehicles with plates and their drivers. They had to hire the drivers and pay for the cars. $30-50 million worth of vehicles and drivers are not sitting unused. I am heading off stupid resolutions at the end of the year. Getting the math evidence entered into evidence and thinking. Pursuading people now.

@brianwang say the poisson conversion of the raw count in Florida shows the 300 raw count is 550 actual population. The registered vehicles in Texas is 499. Then we could no longer resolve N=1000 to NO. I would then say N=1000 would need to be YES. Definitely N=500 would have to be YES. If registered goes to 5000 in Texas and raw count is 2000 in December, conversion could be indicating 5500 in florida. The N=10000 would be in practical dispute. Despite registered and raw Texas at 3000 and raw Florida at 2000. We would need more data and trying to use log rides as sampling counts and conversion to population count estimates.

@brianwang I also want to head off the standard for the count being we must see unique license plates driving around over N=???? like say 10000for resolution. But we are only sampling 1256 and could not possibly count that number of unique license plates. I got burned on this kind of impossible standard when mods only used named dead in Russia and Ukraine as the casualty count. The method and standard did not match reality and being "conservative" on the method guaranteed bias against one side of the prediction. IF there is a - we will use a method where it will be impossible for this standard to resolve N > 2000 then all of the funds for N=10000 or greater must be returned and those question nullified. We should specify the standards now while things are still some days from full resolution.

@brianwang I accept it shouldn't be an impossible standard to reach and glad to hear you are not wanting to resolve 500 yet.

10% registered but not in use seems quite possible to me but I would think this is probably on the high side. You say you don't want to accept more than 5%. That seems a little low to me but maybe we could reach a compromise at say 6%?

If there were 5000 vehicles registered in Texas but on your poisson distribution method with the number of vehicles and rides logged suggested 7000 then this would seem to me to be evidence that your poisson distribution method was overestimating the number and the equivalent number for Florida should be curtailed.

2000 vehicles identified from 2500 logged trips certainly indicates there is quite a lot more than 2000 vehicles. I support a best estimate calculation rather than using something known to be underestimating. However, Poisson distribution is very widely distributed and I think I would be concerned there is probably too much risk that it overestimates the number of vehicles in active use. Show it is not overestimating in Texas where we have an upper limit of registered vehicles and I might be more convinced.

The trouble with specifying a method now is that we don't know what information will be available and what will provide the best least uncertainty data. It may come from ride information rather than estimating up from logged rides. If we can get good estimates of a few metrics like
Rides per day per vehicle,
Revenue per vehicle
Mileage per vehicle per day

That might be helpful.

At the start of the 90 days about 120 vehicles in the tracker. 1265 rides. $10-11 per ride at $3.27 per mile. 3 miles per ride. Less than 4000 miles of rides recorded over 90 days out of 1+ million miles.

502 tesla unsupervised. 479 Texas registered unsupervised robotaxi plus 23 crowdsourced identified florida robotaxi. Per city crowdsource count. Austin 200, Dallas 52, Houston 32, Tampa 16, Miami 4, Orlando 3. The process for crowdsource counting is incomplete. Most people who take a ride do not use the robotaxitracker app to record their ride. Sampling is less than 5% percent. Here is my evidence on sampling level by robotaxitracker. Tesla reported when there were 3 times fewer vehicles there was cumulative 1 million unsupervised miles. the weekly miles needed to be 150,000 miles three weeks ago. the per ride number are 3 miles per ride. 1265 rides in the last 90 days on the tracker. 5000 miles of rides in 90 days on tracker for Tesla. 90 days where it is well over 1 million miles more of the rides in the last 3 weeks post cybercab day. This is a small sample statistics. Definitely cannot identify all of the cars if you are only identifying 5% of the rides. 5 weeks before reporting 1 million cumulative unsupervised miles they reported 380,000 total at the earnings call. In between they reported 17% gains week over week. But since then the number of vehicles has surged. The miles per day per vehicle can also be solved within ranges.

@brianwang
I agree the robotaxitracker is only sampling a small percentage of the rides.

But what numbers to use?
Registered in Texas - could be registered but not actively operating.

7 day number by robotaxitracker is highly likely to be undersampling.

30 day usage number by robotaxitracker is what they seem to prefer to use. This is dropping rapidly from 256 [8 days ago] to 204 [yesterday] to (173 Texas +17 Florida=~190 [today]). Is this logged trips dropping? Likely, but does this also reflect less usage after some excitement after launch event?

If there is just a couple of weeks of high activity after which vehicles drop out of use, is that "actively operating"?

1 million miles on its own. Yes that was reported, but does this count test mileage as well as while vehicles 'in operation'?

@ChristopherRandles It was reported as paid operating unsupervised miles.

@ChristopherRandles Waymo claim 4000 cars. No ONE is saying they are not driving less than 90% of those vehicles. Tesla has 1000 named plates and registered vehicles in SF Bay ARea. The fact that sampling can only see 700 after 1000 were all registered at the start of year is absurd. But it does show 70% count of 1000 active with safety drivers. Registering They must have the vehicle built, the drivers hired and the plates registered. Same for the 479 plates and vehicles registered in Texas. The lack of registration in Florida and finding 23 via sampling indicates there are likely 40+ operating in Florida. Other evidence that matches, is far higher numbers of cybercabs and model Y robotaxi badged in dallas, houston, austin etc... higher Vehicles counts uniquely identiified with plates, higher reported paid miles, higher number of pictures of vehicles. Random traffic cameras showing 0.12% of vehicles in Austin are cybercab. Not just Tesla Y robotaxi.

@ChristopherRandles

Tesla Robotaxi app still has high downloads and high usage.

https://www.ark-invest.com/newsletters/issue-478

Just a moment...

@brianwang more app activity than waymo

robotaxitracker. 281 tesla

@brianwang 7 day active number at https://robotaxitracker.com/ is 116. 30 day number is 248 so some of those 281 / 248 vehicles may not currently be in use. 116 should probably be enough but might a '1 week stunt' or 'temporary high demand' fail to satisfy an "actively operating" test? If the 7 day number is still above 100 by a couple of weeks after Sept 3rd then I think that should be pretty unambiguous to get a mod to resolve the N=100 market.

Trouble with above might be if the 116 7 day number drops below 100 entirely because fewer trips are logged on that site but vehicles are actually still being used as much. So maybe it should resolve on this 116 7 day number?

@ChristopherRandles what is the number of Waymo? Waymo claims over 4000. But the tracker can only spot about 1300 detected by the crowdsource.

@brianwang 439 registered unsupervised in Texas to Tesla 21 unsupervised driving in Florida total is 460. Only 1311 out over 4000 Waymo spotted by crowdsource site . Average driving distances increasing for tesla ark invest tracks over triple the Tesla Robotaxi app downloads and more phone activity 305 active Tesla unsupervised detected

@brianwang If there are around 4000 vehicle operating and you log 1966 trips then quite a few of the trips logged will be in cars previously logged. So 1311 detected vehicles from this number of trips sounds quite plausible especially if some vehicles do less trips than others making them harder to detect.

For tesla the situation is rather different, more dedicated fans that are more likely to log trips. with 7354 logged trips and only 259 vehicles logged in last 30 days, there is over 20 logged trips for each vehicle so you find a large proportion of the vehicles.

For Tesla vehicles identified in
last 60 days 289
last 30 days 259
last 14 days 150
last 7 days 66

The drop from 289 to 259 may well be vehicles that Tesla is no longer using as robotaxis. The declines as we look at shorter more recent period is likely a combination of:

Possible drop of in use of Tesla robotaxis following excitement at the cybercab launch, and
Reduced logging of trip because as the cybercab becomes less novel.

Unfortunately I don't really see how to attribute the dropping numbers between those causes. Maybe the increases in that 7354 trips logged number over the next week together with the how the 289 259 150 and 66 numbers change by a weeks time might be useful and provide more insight?

A week later or nearly.
7421 trips logged so 67 more than 7354 a week ago
311 vehicles up from 305
For Tesla vehicles identified in
last 60 days 290 up 1
last 30 days 204 down 55
last 14 days 83 down 67
last 7 days 22 down 44

If there were over 100 vehicles operating similar mileages/trips and the 67 logged trips were selected at random then you would expect to find more than 22 of the vehicles. However the sampling probably isn't random, some dedicated fans may be logging most of the trips and those trips might predominantly start in certain locations which only a few cars cover.

While the number of vehicles seems to have gone over 100 in a 7 day period it isn't clear whether this has lasted for longer than ~ a week before and after the launch date. Maybe a graph of vehicle miles travelled or better paid vehicle miles travelled will become available mid Oct at 3rd Q earnings call but it wouldn't surprise me if Tesla only show monthly frequency on such a graph to hide the/any decline after launch.

Perhaps being a bit of a pain, but there is plenty of time for more data to come in.

Maybe some breakdown by area might help so recording some more numbers:

Austin
All time tracked 246
In service 198 (community tracked minus not spotted in last 30 days and no longer on Texas DMV register)
cybercabs 42
Trips logged 4375
60 days 189
30 days 135
14 days 43
7 days 8

Dallas
51 unsupervised
492 trips logged
60 days 50
30 days 39
14 days 27
7 days 9

Houston
32 unsupervised
917 trips logged
60 days 29
30 days 13
14 days 1
7 days 1

Tampa
16 unsupervised
86 trips logged
60 days 16
30 days 14
14 days 10
7 days 4

Miami
4 unsupervised
10 trips logged
60 days 3
30 days 2
14 days 2
7 days 0

Orlando
3 unsupervised
12 trips logged
60 days 3
30 days 1
14 days 0
7 days 0

So much for saving the information above, the site seems to have changed. Now seeing:

Trips logged 7718
Unsupervised 314 no safety driver

Texas registered 479 (410 model Y, 69 Cybercab)
30 Days 288 +40 in last 7 days

Matched to tracked plates 268
173 spotted in last 30 days

So that is now 297 more trips than yesterdays and 364 more trips than 8 days ago Sept 15th, but spotted in last 30 days has continued to fall from 259 to 204 to (173Texas +~17 Florida=~190). Strange? This possibly indicates most of these 297 extra trips logged are older than 30 days.

@ChristopherRandles for waymo logging 1900 rides means you are logging 0.5 rides per vehicle out of 4000 vehiciles. Numberically impossible to count 4000 vehicles if you record 1900 rides.

@brianwang After 1966 logged trips you cannot find 1966 vehicles. If the sampling were random you might expect to have discovered 1550 ish vehicles but only 1311 have been found. This 1311 being lower than 1550 could be
a) chance - but this is I think very unlikely
b) the sampling is not random; some cars do fewer trips and or operate less time making them harder to detect (but are still actively operating) or the sample drawn is limited to a few trackers whose range of movement is unevenly distributed over areas the robotaxis operate.
c) Not all of the 4000 vehicles are 'actively operating'.

b) seems to me to be most likely to explain most of the difference between 1550 and 1311 but it is hard to rule out the other explanations playing a part.

While I think the 4000 Waymo vehicles exist, I find it very hard to say whether 0 or 100 or 200 or ... are operating so little that we wouldn't say they are actively operating.

Same applies to 479 registered vehicles in Texas, I don't know how many are 'actively operating'.