DandY

I don't think anyone understands the algorithm VJ uses, but from my analysis I believe it to be a VERY simple averaging of timelines rolling across a rolling multiple month averaging of reported adjudication periods. If I am correct, there is no negative impact from people not updating, there is just less data to go by and less data = less accurate.
 
For those interested in my opinions (precious few I'm sure :P) on the math and have too much time on their hands, read on.
 
It seems like the VJ algorithm is a rolling average og somewhere between 3 and 5 months from what I can tell in seat of the pants analysis. This explains why it did not react to a dramatic shift in timelines and it has been VERY slow to update our timelines constantly pushing them out every week as we were still being "polluted" by pre-April event timelines which resulted in overly optimistic predictions. I believe VJ predictions would have had the same problem when the unusual happened in the DACA slowdown. 
 
I created an algorithm that understands a dramatic shift and historical trends. So, the way mine works is like this:
1) The base is similar to what I think VJ's is (simple rolling average)--I use a 3 month period.
 
2) A rolling 1 year timeline to determine typical trends (i.e. "historically, how much longer or shorter as a percentage is a given month compared to the previous month" ). It then factors this historical trend information into the prediction based upon the predicted average from step 1.
 
3) It then looks at the rolling data for a dramatic shift (did a sample month shift 300% greater than was predicted by the calculations from step 1 and 2 above--e.g. if June is typically 10% longer than May, but all of a sudden June is 30% longer--the algorithm considers this a dramatic shift). 
 
4) If a dramatic shift occurs it attempt to determine "T0" (the date the shift occurs)--it then throws out all data it uses for the averaging before the shift date. And only averages with post T0 data but still applies the trend information from step 2.
 
My algorithm has the following deficiencies:
1) It does not really "understand" different types of dramatic events. It assumes they all behave the same way and are largely able to be correlated to historical trends.
2) Cannot predict a date AT ALL once a shift is detected until there is data available (for instance, if the algorithm predicts March filers should wait 90 days, but we get to 118 days, it pukes and says "oh #######, all bets are off, the world changed and there is no data to base a prediction on". It needs to wait for those March filers to get adjudicated before it can start predicting again.
3) After a dramatic shift, it has a very small data set when it starts predicting again, so it will fluctuate too much for a bit once it starts predicting again. 
 
For over a month, my algorithm has been predicting early June filers to be at an average of 152 days. I have been pretty confident in this calculation +/- a few days for awhile now because I thought the logic was mostly sound and unscientifically it "felt" right. However, I consider it now in question because it seems that the historical trends do not apply. If they did, we we be in the period of adjudication dates starting to get a little quicker. They are not. The trend line is still going up. I likely need to adjust the algorithm to additionally recognize when the historical trend lines are no longer valid. Problem is, if I do it on a simple prediction based upon the slope of the current timeline, it has just as much chance of being right, or being too pessimistic. For now, I'm thinking of possibly recognizing the event that the historical timeline is incorrect, then applying a weighting between the slope of the actual timeline and the historical timeline--but it is unlikely that will be any more accurate.
 
Stay strong my brothers and sisters!

LOL.
 
"Let me see his last year's tax return" <note social security number, birthdate, name, and address>
 
"And his mother's maiden name?" <note maiden name>
 
"Thank you, we have what we need, here is your sticker!" <hehehehehe>

I don't think anyone understands the algorithm VJ uses, but from my analysis I believe it to be a VERY simple averaging of timelines rolling across a rolling multiple month averaging of reported adjudication periods. If I am correct, there is no negative impact from people not updating, there is just less data to go by and less data = less accurate.
 
For those interested in my opinions (precious few I'm sure :P) on the math and have too much time on their hands, read on.
 
It seems like the VJ algorithm is a rolling average og somewhere between 3 and 5 months from what I can tell in seat of the pants analysis. This explains why it did not react to a dramatic shift in timelines and it has been VERY slow to update our timelines constantly pushing them out every week as we were still being "polluted" by pre-April event timelines which resulted in overly optimistic predictions. I believe VJ predictions would have had the same problem when the unusual happened in the DACA slowdown. 
 
I created an algorithm that understands a dramatic shift and historical trends. So, the way mine works is like this:
1) The base is similar to what I think VJ's is (simple rolling average)--I use a 3 month period.
 
2) A rolling 1 year timeline to determine typical trends (i.e. "historically, how much longer or shorter as a percentage is a given month compared to the previous month" ). It then factors this historical trend information into the prediction based upon the predicted average from step 1.
 
3) It then looks at the rolling data for a dramatic shift (did a sample month shift 300% greater than was predicted by the calculations from step 1 and 2 above--e.g. if June is typically 10% longer than May, but all of a sudden June is 30% longer--the algorithm considers this a dramatic shift). 
 
4) If a dramatic shift occurs it attempt to determine "T0" (the date the shift occurs)--it then throws out all data it uses for the averaging before the shift date. And only averages with post T0 data but still applies the trend information from step 2.
 
My algorithm has the following deficiencies:
1) It does not really "understand" different types of dramatic events. It assumes they all behave the same way and are largely able to be correlated to historical trends.
2) Cannot predict a date AT ALL once a shift is detected until there is data available (for instance, if the algorithm predicts March filers should wait 90 days, but we get to 118 days, it pukes and says "oh #######, all bets are off, the world changed and there is no data to base a prediction on". It needs to wait for those March filers to get adjudicated before it can start predicting again.
3) After a dramatic shift, it has a very small data set when it starts predicting again, so it will fluctuate too much for a bit once it starts predicting again. 
 
For over a month, my algorithm has been predicting early June filers to be at an average of 152 days. I have been pretty confident in this calculation +/- a few days for awhile now because I thought the logic was mostly sound and unscientifically it "felt" right. However, I consider it now in question because it seems that the historical trends do not apply. If they did, we we be in the period of adjudication dates starting to get a little quicker. They are not. The trend line is still going up. I likely need to adjust the algorithm to additionally recognize when the historical trend lines are no longer valid. Problem is, if I do it on a simple prediction based upon the slope of the current timeline, it has just as much chance of being right, or being too pessimistic. For now, I'm thinking of possibly recognizing the event that the historical timeline is incorrect, then applying a weighting between the slope of the actual timeline and the historical timeline--but it is unlikely that will be any more accurate.
 
Stay strong my brothers and sisters!

I don't think anyone understands the algorithm VJ uses, but from my analysis I believe it to be a VERY simple averaging of timelines rolling across a rolling multiple month averaging of reported adjudication periods. If I am correct, there is no negative impact from people not updating, there is just less data to go by and less data = less accurate.
 
For those interested in my opinions (precious few I'm sure :P) on the math and have too much time on their hands, read on.
 
It seems like the VJ algorithm is a rolling average og somewhere between 3 and 5 months from what I can tell in seat of the pants analysis. This explains why it did not react to a dramatic shift in timelines and it has been VERY slow to update our timelines constantly pushing them out every week as we were still being "polluted" by pre-April event timelines which resulted in overly optimistic predictions. I believe VJ predictions would have had the same problem when the unusual happened in the DACA slowdown. 
 
I created an algorithm that understands a dramatic shift and historical trends. So, the way mine works is like this:
1) The base is similar to what I think VJ's is (simple rolling average)--I use a 3 month period.
 
2) A rolling 1 year timeline to determine typical trends (i.e. "historically, how much longer or shorter as a percentage is a given month compared to the previous month" ). It then factors this historical trend information into the prediction based upon the predicted average from step 1.
 
3) It then looks at the rolling data for a dramatic shift (did a sample month shift 300% greater than was predicted by the calculations from step 1 and 2 above--e.g. if June is typically 10% longer than May, but all of a sudden June is 30% longer--the algorithm considers this a dramatic shift). 
 
4) If a dramatic shift occurs it attempt to determine "T0" (the date the shift occurs)--it then throws out all data it uses for the averaging before the shift date. And only averages with post T0 data but still applies the trend information from step 2.
 
My algorithm has the following deficiencies:
1) It does not really "understand" different types of dramatic events. It assumes they all behave the same way and are largely able to be correlated to historical trends.
2) Cannot predict a date AT ALL once a shift is detected until there is data available (for instance, if the algorithm predicts March filers should wait 90 days, but we get to 118 days, it pukes and says "oh #######, all bets are off, the world changed and there is no data to base a prediction on". It needs to wait for those March filers to get adjudicated before it can start predicting again.
3) After a dramatic shift, it has a very small data set when it starts predicting again, so it will fluctuate too much for a bit once it starts predicting again. 
 
For over a month, my algorithm has been predicting early June filers to be at an average of 152 days. I have been pretty confident in this calculation +/- a few days for awhile now because I thought the logic was mostly sound and unscientifically it "felt" right. However, I consider it now in question because it seems that the historical trends do not apply. If they did, we we be in the period of adjudication dates starting to get a little quicker. They are not. The trend line is still going up. I likely need to adjust the algorithm to additionally recognize when the historical trend lines are no longer valid. Problem is, if I do it on a simple prediction based upon the slope of the current timeline, it has just as much chance of being right, or being too pessimistic. For now, I'm thinking of possibly recognizing the event that the historical timeline is incorrect, then applying a weighting between the slope of the actual timeline and the historical timeline--but it is unlikely that will be any more accurate.
 
Stay strong my brothers and sisters!

I don't think anyone understands the algorithm VJ uses, but from my analysis I believe it to be a VERY simple averaging of timelines rolling across a rolling multiple month averaging of reported adjudication periods. If I am correct, there is no negative impact from people not updating, there is just less data to go by and less data = less accurate.
 
For those interested in my opinions (precious few I'm sure :P) on the math and have too much time on their hands, read on.
 
It seems like the VJ algorithm is a rolling average og somewhere between 3 and 5 months from what I can tell in seat of the pants analysis. This explains why it did not react to a dramatic shift in timelines and it has been VERY slow to update our timelines constantly pushing them out every week as we were still being "polluted" by pre-April event timelines which resulted in overly optimistic predictions. I believe VJ predictions would have had the same problem when the unusual happened in the DACA slowdown. 
 
I created an algorithm that understands a dramatic shift and historical trends. So, the way mine works is like this:
1) The base is similar to what I think VJ's is (simple rolling average)--I use a 3 month period.
 
2) A rolling 1 year timeline to determine typical trends (i.e. "historically, how much longer or shorter as a percentage is a given month compared to the previous month" ). It then factors this historical trend information into the prediction based upon the predicted average from step 1.
 
3) It then looks at the rolling data for a dramatic shift (did a sample month shift 300% greater than was predicted by the calculations from step 1 and 2 above--e.g. if June is typically 10% longer than May, but all of a sudden June is 30% longer--the algorithm considers this a dramatic shift). 
 
4) If a dramatic shift occurs it attempt to determine "T0" (the date the shift occurs)--it then throws out all data it uses for the averaging before the shift date. And only averages with post T0 data but still applies the trend information from step 2.
 
My algorithm has the following deficiencies:
1) It does not really "understand" different types of dramatic events. It assumes they all behave the same way and are largely able to be correlated to historical trends.
2) Cannot predict a date AT ALL once a shift is detected until there is data available (for instance, if the algorithm predicts March filers should wait 90 days, but we get to 118 days, it pukes and says "oh #######, all bets are off, the world changed and there is no data to base a prediction on". It needs to wait for those March filers to get adjudicated before it can start predicting again.
3) After a dramatic shift, it has a very small data set when it starts predicting again, so it will fluctuate too much for a bit once it starts predicting again. 
 
For over a month, my algorithm has been predicting early June filers to be at an average of 152 days. I have been pretty confident in this calculation +/- a few days for awhile now because I thought the logic was mostly sound and unscientifically it "felt" right. However, I consider it now in question because it seems that the historical trends do not apply. If they did, we we be in the period of adjudication dates starting to get a little quicker. They are not. The trend line is still going up. I likely need to adjust the algorithm to additionally recognize when the historical trend lines are no longer valid. Problem is, if I do it on a simple prediction based upon the slope of the current timeline, it has just as much chance of being right, or being too pessimistic. For now, I'm thinking of possibly recognizing the event that the historical timeline is incorrect, then applying a weighting between the slope of the actual timeline and the historical timeline--but it is unlikely that will be any more accurate.
 
Stay strong my brothers and sisters!

Yeah, the K1 section of the website should be renamed "Separated Lovers Emotional Support Group".

I have read posts by this poster and they seem to follow the info from that region pretty closely so I am inclined to accept their data. All they said was K1 approval rate from the region was exceptionally low 5-6 years ago. 2-3 years ago it become a little higher. Then this year it dropped back to extremely low. The drop correlates with the new President and his rhetoric. Given their other posts, I don't believe their intent was to "blame" the president and start a flame war.

I have read posts by this poster and they seem to follow the info from that region pretty closely so I am inclined to accept their data. All they said was K1 approval rate from the region was exceptionally low 5-6 years ago. 2-3 years ago it become a little higher. Then this year it dropped back to extremely low. The drop correlates with the new President and his rhetoric. Given their other posts, I don't believe their intent was to "blame" the president and start a flame war.

I have read posts by this poster and they seem to follow the info from that region pretty closely so I am inclined to accept their data. All they said was K1 approval rate from the region was exceptionally low 5-6 years ago. 2-3 years ago it become a little higher. Then this year it dropped back to extremely low. The drop correlates with the new President and his rhetoric. Given their other posts, I don't believe their intent was to "blame" the president and start a flame war.

I have read posts by this poster and they seem to follow the info from that region pretty closely so I am inclined to accept their data. All they said was K1 approval rate from the region was exceptionally low 5-6 years ago. 2-3 years ago it become a little higher. Then this year it dropped back to extremely low. The drop correlates with the new President and his rhetoric. Given their other posts, I don't believe their intent was to "blame" the president and start a flame war.

The people to blame are the many Moroccans who have come before them and lied in order to get visas, not the administration. As much as I hate Trump, he isn't the one to blame here.

I have read posts by this poster and they seem to follow the info from that region pretty closely so I am inclined to accept their data. All they said was K1 approval rate from the region was exceptionally low 5-6 years ago. 2-3 years ago it become a little higher. Then this year it dropped back to extremely low. The drop correlates with the new President and his rhetoric. Given their other posts, I don't believe their intent was to "blame" the president and start a flame war.

Yes, we know they don't work them in order. People with the same NOA1 date will get NOA2 a month apart! Like the others, it is either a fluke (file slipped into wrong box) or an expedite. There are still April cases being work, they are not even close to June yet.

Yeah, the K1 section of the website should be renamed "Separated Lovers Emotional Support Group".

Don’t give up but you also have to be realistic. Right now your anticipated NOA2 dat is Nov 12. Add 3 to 4 weeks for NVC and that puts you at Dec 12. Then scheduling medical and interview can be another month out. 
 
These are obviously very conservative numbers but by how much?  A few weeks?  It has been said that is it possible to get an interview scheduled as quick as 45 days after NOA2. If your NOA2 happens on Nov 12, 45 days later will be the week of January 1. 
 
If if you use the 45 days, which is very aggressive, you would need to receive your NOA2 by Oct 30 to have the interview on Dec 16. So to give yourself the 10 days to get the visa after the interview and a week to travel the NOA2 would need to be approved by Oct 6 in order to make it by Dec 16. 
 
I dont want to be the bearer of bad news but these are how the raw numbers work for all of us right now. 
 
My anticipated NOA2 date is now Nov 4. When I originally joined it was September 6. It is between 5 and 6 months for NOA2. 
 
Please dont buy tickets or plan a wedding before you have visa on hand. There are too many variables including AP that can eat another month or two!!!  Ask those February and March filers who are still waiting!
 
I hope all goes well for you but do this with open eyes. This is very slow right now. 

Clearly I said something so that statement is false. You do not know if I know of what I speak since you do not know me or my experience. Your assertion that the only qualified process improvement opinion requires 20 years in a job is demonstrably false.
 
As for why I made my claim--what I do for a living is automate document centric workflows like the USCIS ones and have done so for many state governments and large financial institutions so am very familiar with the technology and techniques. I have not performed in depth analysis of the USCIS and I am only going by descriptions of the process provided by people that work there. I have no reason to believe they have misrepresented the process, so I used a rough time on task calculation based upon their description and benchmarked against very similar types of work which average a 37% reduction if effort from the types of automation I am proposing. 
 
To claim that there is no way to automate any portion of that process and reap savings is patently absurd.  
 

Clearly I said something so that statement is false. You do not know if I know of what I speak since you do not know me or my experience. Your assertion that the only qualified process improvement opinion requires 20 years in a job is demonstrably false.
 
As for why I made my claim--what I do for a living is automate document centric workflows like the USCIS ones and have done so for many state governments and large financial institutions so am very familiar with the technology and techniques. I have not performed in depth analysis of the USCIS and I am only going by descriptions of the process provided by people that work there. I have no reason to believe they have misrepresented the process, so I used a rough time on task calculation based upon their description and benchmarked against very similar types of work which average a 37% reduction if effort from the types of automation I am proposing. 
 
To claim that there is no way to automate any portion of that process and reap savings is patently absurd.  
 

This is a better way.
 
If you look at what the adjudicator does in the 15 minutes of adjudication, at least 30% (and possibly 50%) of it could be automated as could the queue management to make sure they actually work them in order (which they don't). Additionally savings could be add in the overhead of file management.

Yeah, the K1 section of the website should be renamed "Separated Lovers Emotional Support Group".

Yeah, the K1 section of the website should be renamed "Separated Lovers Emotional Support Group".

Right now VJ is predicting 146-149 days for us. I'm going to grab database this weekend I think and see what it is looking like. I still think that is optimistic as late April/Early May filers look to average REALLY close to 150 (Many May 1 at exactly 150 and some late April's getting up to 160 and still not adjudicated). I don't think they are working them at the rate they came in though so I don't think we can assume the velocity will be constant, so we'll see if I stick with my 152 estimate for early June filers after the weekend analysis. 

Right now VJ is predicting 146-149 days for us. I'm going to grab database this weekend I think and see what it is looking like. I still think that is optimistic as late April/Early May filers look to average REALLY close to 150 (Many May 1 at exactly 150 and some late April's getting up to 160 and still not adjudicated). I don't think they are working them at the rate they came in though so I don't think we can assume the velocity will be constant, so we'll see if I stick with my 152 estimate for early June filers after the weekend analysis. 

One thing I meant to add when I posted this: the fiscal year for USCIS is Oct 1 to Sept 30 so they are shifted exactly a quarter from calendar year. So 3rd quart in these reports is 2nd quarter calendar year. It takes them 2-3months to publish the data so (hopefully!) we will all be approved before 4th quarter (July-Sept) number are published. 
 
And because my role here is that of pessimism, it looks like we are on track for early May filers to average over 150 days which means our timelines are still getting longer. Unless something significant changes, June 2017 filers will have the longest average wait time in 4 years (longest since DACA in 2013).
 
I have been pretty confident for almost a month in my algorithm which predicted an average wait time of 152 days for early June filers (which is longer than VJ has ever predicted for us). If this week finishes out like it is starting, my algorithm will bump to somewhere between 156 and 158 days for early June filers. Yay us! (NOT!)

Does it show?  But not a "real" engineer (but I play one on TV)--I do software.
 
In fact, I automate document-centric workflows very much like the USCIS one! In my analysis of the process (as explained by the former adjudicator), I believe I could implement a system at USCIS that would reduce the 15 minute adjudication process to between 7-9 minutes and add additional savings on initial processing and file routing. So watching this process REALLY pains me because I know how to make it faster and cheaper with not that complex of a software system.
 
 

Does it show?  But not a "real" engineer (but I play one on TV)--I do software.
 
In fact, I automate document-centric workflows very much like the USCIS one! In my analysis of the process (as explained by the former adjudicator), I believe I could implement a system at USCIS that would reduce the 15 minute adjudication process to between 7-9 minutes and add additional savings on initial processing and file routing. So watching this process REALLY pains me because I know how to make it faster and cheaper with not that complex of a software system.