## Use of GAT For Pick 3

### Use of GAT For Pick 3

Hello everyone!

I played around with GAT about a year ago and had some good success with my state's in Pick 3 lottery. For reasons unrelated to GAT I went on "hiatus" from playing the lottery, and now I'm back . . . but I can't seem to remember the how's and why's of customizing GAT to allow use for Pick 3! Would anyone care to share how they use GAT for Pick 3 lotteries? Based on guidance I thought I had found in this forum (I can't seem to find the relevant posts now), here's the process I went through:

Step 1: Create three separate Lottery files for each Pick 3 digit, using the "Powerball" type lottery. The main draw numbers are set to placeholder values (I use 1 2 3 4 5 but the numbers don't matter since you won't be running GAT for the main draw numbers), and the Powerball number is set to the Pick 3 digit value plus 1 (need to add 1 because GAT can't handle 0 in this field, just have to remember to subtract 1 from any resulting predictions later on).

Step 2: For each of the three lottery files created per Step 1, run GAT for the Powerball section (not the Main Number section), with settings of 100, 20, and 466 for Draws, Data, and RF.

Step 3: Upon completion of Step 2, set Panorama Mode to Augmentative, set Numbers Requested to 1, export predictions from the the top 19 GAT Tables to clipboard, and copy to Tab 1 on a spreadsheet. The tab spits out a ranking of all digits based on their frequency of occurrence in the tab.

Repeat Step 3 for Numbers Requested of 2, 3, 4, 5, 6, 7, 8, and 9. Each go into separate tabs as described above, and each tab spits out frequency ranking as described above.

Step 4: it's time for data interpretation! The way I do it is convoluted and the subject of a future post, but at a high level all I'm doing is looking at which frequency ranks consistently point to the next drawn number based on the last 2 weeks of historical data, and playing the trend accordingly (unless I've overcompicated things, it's NOT as simple as merely taking the top 1, 2, or 3 ranked numbers from the frequency ranks spit out by each tab of Step 3).

When I was playing using this method last year, I won straight in two consecutive drawings (the timing was great, it was right before vacation!) playing around 50 combinations for each drawing.

Any thoughts? Comments? Suggestions?

I played around with GAT about a year ago and had some good success with my state's in Pick 3 lottery. For reasons unrelated to GAT I went on "hiatus" from playing the lottery, and now I'm back . . . but I can't seem to remember the how's and why's of customizing GAT to allow use for Pick 3! Would anyone care to share how they use GAT for Pick 3 lotteries? Based on guidance I thought I had found in this forum (I can't seem to find the relevant posts now), here's the process I went through:

Step 1: Create three separate Lottery files for each Pick 3 digit, using the "Powerball" type lottery. The main draw numbers are set to placeholder values (I use 1 2 3 4 5 but the numbers don't matter since you won't be running GAT for the main draw numbers), and the Powerball number is set to the Pick 3 digit value plus 1 (need to add 1 because GAT can't handle 0 in this field, just have to remember to subtract 1 from any resulting predictions later on).

Step 2: For each of the three lottery files created per Step 1, run GAT for the Powerball section (not the Main Number section), with settings of 100, 20, and 466 for Draws, Data, and RF.

Step 3: Upon completion of Step 2, set Panorama Mode to Augmentative, set Numbers Requested to 1, export predictions from the the top 19 GAT Tables to clipboard, and copy to Tab 1 on a spreadsheet. The tab spits out a ranking of all digits based on their frequency of occurrence in the tab.

Repeat Step 3 for Numbers Requested of 2, 3, 4, 5, 6, 7, 8, and 9. Each go into separate tabs as described above, and each tab spits out frequency ranking as described above.

Step 4: it's time for data interpretation! The way I do it is convoluted and the subject of a future post, but at a high level all I'm doing is looking at which frequency ranks consistently point to the next drawn number based on the last 2 weeks of historical data, and playing the trend accordingly (unless I've overcompicated things, it's NOT as simple as merely taking the top 1, 2, or 3 ranked numbers from the frequency ranks spit out by each tab of Step 3).

When I was playing using this method last year, I won straight in two consecutive drawings (the timing was great, it was right before vacation!) playing around 50 combinations for each drawing.

Any thoughts? Comments? Suggestions?

- lottoarchitect
- Site Admin
**Posts:**1493**Joined:**Tue Jan 15, 2008 5:03 pm**Location:**Greece-
**Contact:**

### Re: Use of GAT For Pick 3

Hi mitachoo, this is literally the first time someone posts a Pick 3 strategy and the success they had using GAT. I'm really keen to have more such stories of users. Two straight consecutive hits? That is something notable really!

I want to comment on step 1, you do not need to setup your Pick 3 in the poweball field, this is more complicated than needed. Just create a single one field game for each position of your Pick 3 game. You can define a 1/10 game as the 1st field, during the creation of the game you have to pick "user defined" to allow this. At least till I make GAT natively support Pick style games.

For step 2, 466 RF? I assume you mean the start point parameter (and RF=0). Obviously this value will change from draw draw but the default tested draws/stat.data set to 100/20 are ok.

On step 4, can you elaborate on this "but at a high level all I'm doing is looking at which frequency ranks consistently point to the next drawn number based on the last 2 weeks of historical data, and playing the trend accordingly"?

I want to comment on step 1, you do not need to setup your Pick 3 in the poweball field, this is more complicated than needed. Just create a single one field game for each position of your Pick 3 game. You can define a 1/10 game as the 1st field, during the creation of the game you have to pick "user defined" to allow this. At least till I make GAT natively support Pick style games.

For step 2, 466 RF? I assume you mean the start point parameter (and RF=0). Obviously this value will change from draw draw but the default tested draws/stat.data set to 100/20 are ok.

On step 4, can you elaborate on this "but at a high level all I'm doing is looking at which frequency ranks consistently point to the next drawn number based on the last 2 weeks of historical data, and playing the trend accordingly"?

### Re: Use of GAT For Pick 3

Hi there, and thanks for replying!

Regarding the setup, thanks for the tip. I figured I was making it more complicated than it was!

I will elaborate on my Data Interpretation comments later, we're on Vacation and about to touch down in sunny Florida.

Regarding the setup, thanks for the tip. I figured I was making it more complicated than it was!

I will elaborate on my Data Interpretation comments later, we're on Vacation and about to touch down in sunny Florida.

### Re: Use of GAT For Pick 3

OK, I've unraveled what I did, set up my historical database, and am ready to start predicting. I will post my data interpretation method within the next few days; until then, here's my prediction for today's Connecticut's Pick 3 Day drawing:

Digit 1: 1,4,6,7

Digit 2: 1,2,4

Digit 3: 6,7,8

Total combinations: 36

Digit 1: 1,4,6,7

Digit 2: 1,2,4

Digit 3: 6,7,8

Total combinations: 36

### Re: Use of GAT For Pick 3

And for the Pick 3 Night:

Digit 1: 3,6,7,8,9

Digit 2: 2,3,4

Digit 3: 1,2,4,6,7,9

Digit 1: 3,6,7,8,9

Digit 2: 2,3,4

Digit 3: 1,2,4,6,7,9

### Re: Use of GAT For Pick 3

OK, here's a more detailed explanation of the "data interpretation" I alluded to in my original post.

I've created individual "User Defined" games for each Pick 3 digit, populated each with historical data, and ran GAT (Augmentative, Draws 100, Data 20) for 1 hour for each of the three digits. The run typicality yields just under 1 million GAT Tables tested per game. Then, for "1 number requested out of 10", I export GAT predictions for each of the 19 GAT Tables listed into Excel, into my "X1" tab. I repeat for 2 out of 10, 3 out of 10, etc, up to 9 out of 10.

For each tab, the spreadsheet calculates how many times each of the pick 3 digits shows up in the tab, and ranks the digits by frequency of occurrence, high to low. So for each tab (the X1 "1 number requested out of 10" tab, X2 "2 numbers requested out of 10" tab, up to X9 "9 numbers requested out of 10" tab), I end up with a ranked list of digits.

The overlying assumption is that digits that are ranked high (i.e. show up more time than others) are more likely to hit on the next draw. However, if one tracks the ranked numbers for X1 through X9 versus what actually hit, one can quickly ascertain that the top ranked 1, or 2, or 3, etc. do not hit on the next draw with any regularity, so it's not as simple as selecting the top 1, 2, or 3 ranks from each tab and playing those for the next draw.

So what I do track is this: I track the rank of the number drawn within X1 through X9, because this shows some trending that can be taken advantage of. Let me explain with an example: let's say that for the last 7 draws, in 6 cases the actual number drawn was the 4th ranked number in at least one of X1 through X9 based on data analysis after the previous draw. I've found that if for the next draw I consider playing every 4th ranked number from X1 through X9, in most cases I will be correct.

The above by itself usually yields too many numbers: I also eliminate the top 50% of numbers from Ranks that have NOT hit in the past 7 draws.

Backtesting is labor intensive so I'm not 100% certain my 2 consecutive draw success from 2016 was not a fluke. For what it's worth, here's what I would have predicted from the last 4 days for my state's lottery (Connecticut) (please disregard my previous predictions, there was a formula error in my spreadsheet). What's interesting here is the one straight hit came at a VERY low cost (only 24 combinations).

6/2/2015 Day: Predict Digit 1 (045), Digit 2 (7), Digit 3 (023458). 18 total combinations, Hit 1 of 3

6/2/2015 Night: Predict Digit 1 (0147), Digit 2 (034), Digit 3 (23679). 60 total combinations. Hit 2 of 3

6/3/2015 Day: Predict Digit 1 (0679), Digit 2 (38), Digit 3 (259). 24 total combinations, Hit 3 of 3 ***JACKPOT ***

6/3/2017 Night: Predict Digit 1 (0123), Digit 2 (1789), Digit 3 (1478). 64 total combinations, Hit 1 of 3

6/4/2017 Day: Predict Digit 1 (12369), Digit 2 (36), Digit 3 (359). 30 total combinations, Hit 2 of 3

6/4/2017 Night: Predict Digit 1 (135), Digit 2 (12379), Digit 3 (3459). 60 total combinations, Hit 1 of 3

6/5/2017: Day: Predict Digit 1 (016), Digit 2 (357), Digit 3 (268). 27 total combinations, Hit 1 of 3

I've created individual "User Defined" games for each Pick 3 digit, populated each with historical data, and ran GAT (Augmentative, Draws 100, Data 20) for 1 hour for each of the three digits. The run typicality yields just under 1 million GAT Tables tested per game. Then, for "1 number requested out of 10", I export GAT predictions for each of the 19 GAT Tables listed into Excel, into my "X1" tab. I repeat for 2 out of 10, 3 out of 10, etc, up to 9 out of 10.

For each tab, the spreadsheet calculates how many times each of the pick 3 digits shows up in the tab, and ranks the digits by frequency of occurrence, high to low. So for each tab (the X1 "1 number requested out of 10" tab, X2 "2 numbers requested out of 10" tab, up to X9 "9 numbers requested out of 10" tab), I end up with a ranked list of digits.

The overlying assumption is that digits that are ranked high (i.e. show up more time than others) are more likely to hit on the next draw. However, if one tracks the ranked numbers for X1 through X9 versus what actually hit, one can quickly ascertain that the top ranked 1, or 2, or 3, etc. do not hit on the next draw with any regularity, so it's not as simple as selecting the top 1, 2, or 3 ranks from each tab and playing those for the next draw.

So what I do track is this: I track the rank of the number drawn within X1 through X9, because this shows some trending that can be taken advantage of. Let me explain with an example: let's say that for the last 7 draws, in 6 cases the actual number drawn was the 4th ranked number in at least one of X1 through X9 based on data analysis after the previous draw. I've found that if for the next draw I consider playing every 4th ranked number from X1 through X9, in most cases I will be correct.

The above by itself usually yields too many numbers: I also eliminate the top 50% of numbers from Ranks that have NOT hit in the past 7 draws.

Backtesting is labor intensive so I'm not 100% certain my 2 consecutive draw success from 2016 was not a fluke. For what it's worth, here's what I would have predicted from the last 4 days for my state's lottery (Connecticut) (please disregard my previous predictions, there was a formula error in my spreadsheet). What's interesting here is the one straight hit came at a VERY low cost (only 24 combinations).

6/2/2015 Day: Predict Digit 1 (045), Digit 2 (7), Digit 3 (023458). 18 total combinations, Hit 1 of 3

6/2/2015 Night: Predict Digit 1 (0147), Digit 2 (034), Digit 3 (23679). 60 total combinations. Hit 2 of 3

6/3/2015 Day: Predict Digit 1 (0679), Digit 2 (38), Digit 3 (259). 24 total combinations, Hit 3 of 3 ***JACKPOT ***

6/3/2017 Night: Predict Digit 1 (0123), Digit 2 (1789), Digit 3 (1478). 64 total combinations, Hit 1 of 3

6/4/2017 Day: Predict Digit 1 (12369), Digit 2 (36), Digit 3 (359). 30 total combinations, Hit 2 of 3

6/4/2017 Night: Predict Digit 1 (135), Digit 2 (12379), Digit 3 (3459). 60 total combinations, Hit 1 of 3

6/5/2017: Day: Predict Digit 1 (016), Digit 2 (357), Digit 3 (268). 27 total combinations, Hit 1 of 3

### Re: Use of GAT For Pick 3

OK. I'm working on refining this confusing mess into something that's simpler (and seems to work better based on initial backtesting!) More to come.

### Re: Use of GAT For Pick 3

OK, now I need a sanity check! I've (presumably) fine-tuned my Strategy and would like to know how the following results compare to Random Play. These are from Connecticut twice daily Pick 3 lottery, using GAT and a Strategy that is too cumbersome and convoluted to type (it BEGS for a YouTube video!)

Here are my back-testing results. 2 jackpot hits hits over 27 draws, June 6 and June 8, one with an ultra low pool of only 8 combos! Also 7 near misses (2 of 3 digits hit).

DAY NIGHT

DATE POOL HITS POOL HITS

6/2/17 24 1 8 1

6/3/17 8 0 18 2

6/4/17 6 2 9 2

6/5/17 36 2 36 2

6/6/17 30 0 45 3

6/7/17 36 1 12 1

6/8/17 18 2 8 3

6/9/17 36 0 12 0

6/10/17 4 0 18 0

6/11/17 12 0 8 1

6/12/17 16 1 18 0

6/13/17 12 0 12 0

6/14/17 18 1 16 2

6/15/17 16 1

Here are my back-testing results. 2 jackpot hits hits over 27 draws, June 6 and June 8, one with an ultra low pool of only 8 combos! Also 7 near misses (2 of 3 digits hit).

DAY NIGHT

DATE POOL HITS POOL HITS

6/2/17 24 1 8 1

6/3/17 8 0 18 2

6/4/17 6 2 9 2

6/5/17 36 2 36 2

6/6/17 30 0 45 3

6/7/17 36 1 12 1

6/8/17 18 2 8 3

6/9/17 36 0 12 0

6/10/17 4 0 18 0

6/11/17 12 0 8 1

6/12/17 16 1 18 0

6/13/17 12 0 12 0

6/14/17 18 1 16 2

6/15/17 16 1

- lottoarchitect
- Site Admin
**Posts:**1493**Joined:**Tue Jan 15, 2008 5:03 pm**Location:**Greece-
**Contact:**

### Re: Use of GAT For Pick 3

Hi mitachoo, interesting results indeed. Can you clarify a bit the hits columns? The pool I assume is how many combinations played, the hit refers to one of those combinations had e.g. 2 correct numbers in there? Do you win anything with 2 correct numbers or only a straight win can return an outcome? Given you make all the combinations of the numbers picked by position, I can expect more than a single combination to provide e.g. 2 correct numbers picked.

Apart from that, it is easy to compute odds for a pick-3 game. The straight win has 1/1000 chance to be picked. If you play just one combination, this means on average 1000/1=1000 draws should get drawn to get one straight win. if you play more combinations let's say 5 combinations, the average waiting time is 1000/5=200 draws of playing 5 combinations in each draw to get one straight win. So, your win on 6 June where you had 45 combinations to get that straight win, the average waiting time would be 1000/45 = 22 draws. Similarly for the 8 June draw where 8 combinations we used, 1000/8 = 125 draws. You are well above natural probability here.

Apart from that, it is easy to compute odds for a pick-3 game. The straight win has 1/1000 chance to be picked. If you play just one combination, this means on average 1000/1=1000 draws should get drawn to get one straight win. if you play more combinations let's say 5 combinations, the average waiting time is 1000/5=200 draws of playing 5 combinations in each draw to get one straight win. So, your win on 6 June where you had 45 combinations to get that straight win, the average waiting time would be 1000/45 = 22 draws. Similarly for the 8 June draw where 8 combinations we used, 1000/8 = 125 draws. You are well above natural probability here.

### Re: Use of GAT For Pick 3

I’ve been tweaking the hell out of my Pick 3 GTA System over the last few weeks and I’m almost to the point where I believe my tweaking is done. I’ll now explain in detail what I’m doing and I welcome any feedback. I apologize for the long post (and for adding to the confusion by not being able to add screenshots).

First some preliminaries:

- The goal is to win Straight Pick 3 using GTA.

- I wanted to come up with a system that relied SOLELY on raw prediction data from the top 19 GTA Tables, with no “interpretation” of GTA Charts required. I am horrible at gleaning useful info from these charts and wanted to make the system “data driven” to eliminate any subjectivity associated with chart interpretation.

- I created six Custom Lotteries, one for each Pick 3 Digit for both the Midday and Evening Connecticut Drawings. I have plenty of data to work with as my data set goes back to the early 2000’s.

- My “backtesting window” starts on 5/25/2017. For reasons that will become apparent shortly, backtesting is VERY labor-intensive and I did not go further back as I feared for my sanity.

Here’s my process at a high level: after the days drawings, I update each of the 6 Custom Lotteries and run GAT for a minimum of 60 minutes and save the results (this typically yields around 1 million GAT’s tested). I then export the top 19 GAT Table predilections for each of the 1 through 9 number prediction options into my Excel Spreadsheet, run my SDM (Super Duper Macro), and voila . . . I now have a pool of numbers to play for the next drawing. Theoretically, the actual drawn number has a higher likelihood of being within this pool than that associated with the size of the pool itself because, again theoretically, we’ve gleaned some improvement over random play for each of the 6 digits.

A side question before I discuss the SDM in detail: with the Custom Lotteries I’ve created being based on the relatively high odds of 1 in 10 (relatively high comparted to Powerball, Mega Millions, etc), is there even any need to run GAT for so long? Can I reduce the run time to something significantly less than 60 minutes without compromising accuracy? The reason I ask is this run time is the biggest hurdle to more backtesting; if it can be “safely” be reduced to 10 minutes or so I would be more apt to incorporate more backtesting.

And now the SDM discussion. Note that everything I describe for the SDM is done once for each digit (6 runs total), and most of the legwork is done by the SDM with one notable exception.

We already ran GAT after the latest draw as discussed previously. The next step is to export the top 19 GAT tables into the SDM for each of the 1 through 9 digit prediction options. There are 9 tabs each dedicated to the1 through 9 digit prediction options, and when predictions are pasted in, the spreadsheet automatically calculates the frequency of occurrence of each digit 0 through 9 in each tab. The overlying assumption is that the digits that show up more than others are more likely to hit in the next draw.

A side comment: copying and pasting 9 times for each of the 6 Custom Lotteries is the second biggest time driver (behind running GAT for 60 minutes). If future GAT updates can include a way to instantly export ALL predications at once that would be fabulous!

So now we have a “prioritized” list of digits for each of the 1 through 9 digit prediction options. Unfortunately it’s NOT as simple as selecting the top 3 or so ranked digits from any of the 9; in general there is not enough consistency as to where the next draw hits.

But if we look at it from a Rank standpoint across ALL 9 instead, there’s some trending that can be taken advantage of. Let me explain; if we look at, for example, for the 7 draws, how many times the next winning number was found in at least one of a given Rank across the 9, we can clearly see where some Ranks are “Hot” (i.e. the next winning number was found in one or more of that Rank across the 9 for most of the last 7 days), and where some of the Ranks are “Cold” (i.e. the none of the numbers from that Rank across the 9 showed up in the next draw for most of the last 7 days). Put the two together and you have the basis from SDM: presume the next draw is likely to come from the list of numbers compiled from the “Hottest” Rank, and/or is unlikely to come the list of numbers compiled from the “Coldest” Ranks. This assumes the 10 lists are totally independent statistical entities. Clearly they are not, and this is an approximation.

The above was the basis for my two consecutive wins last year that I alluded to earlier. But I’ve improved upon it significantly! I had gotten frustrated when, due to “overlap” between “Hot” and “Cold” lists, a number initially on the “Hot” list was deleted because it was also on the “Cold” list. And of course, in numerous cases, that’s the number that came up in the next draw! So my improvement was this: instead of just marrying the two lists as described above, one at a time I bump up the 3 “Coldest” lists against all 7 remaining lists and look for common numbers! I end up with 9 sets of numbers, then I do a Frequency Count for each digit (more on this below), and lastly I compare the 3 sets of numbers and look for common numbers between the three. With the SDM being finalized I can now run backtests without any human error ; I will post these numbers as soon as I can figure out how to do so in a manner that doesn’t jumble everything up (like a few posts back).

What I do with the aforementioned Frequency Count is evolving. I had found previously that digits that showed up at least once but occurred well below those that showed up the most, were more likely to hit on the next draw. But what was odd is that the behavior was drastically different for the Midday versus Evening drawings. The Evening performed MUCH better (3 straight hits in 18 draws, and the pool size for the 3 hits were 45, 24, and 24). How can that be when they are using the same machine?

So I played around with this Frequency Count Threshold and found that if I only keep those that showed up the most, the pendulum swings in the other direction . . . now the Evening suffers and Midday performs better . . . MUCH better . . . 5 straight hits in 18 draws (pool sizes 24, 30, 36, 30, 40)!

So here’s the question: do I treat them separately and use whichever Frequency Count method works best for each? Or do I stack the two together, treat them as one (since they use the same machine), and fine time the threshold that way?

Wow. This was a mouthful. My apooligies.

While I was typing I found I didn’t hit the Midday draw. Ironically, amongst the 3 pools of numbers SDM identified ( 4 36 126789), the 4 was the only that hit (490 was drawn).

First some preliminaries:

- The goal is to win Straight Pick 3 using GTA.

- I wanted to come up with a system that relied SOLELY on raw prediction data from the top 19 GTA Tables, with no “interpretation” of GTA Charts required. I am horrible at gleaning useful info from these charts and wanted to make the system “data driven” to eliminate any subjectivity associated with chart interpretation.

- I created six Custom Lotteries, one for each Pick 3 Digit for both the Midday and Evening Connecticut Drawings. I have plenty of data to work with as my data set goes back to the early 2000’s.

- My “backtesting window” starts on 5/25/2017. For reasons that will become apparent shortly, backtesting is VERY labor-intensive and I did not go further back as I feared for my sanity.

Here’s my process at a high level: after the days drawings, I update each of the 6 Custom Lotteries and run GAT for a minimum of 60 minutes and save the results (this typically yields around 1 million GAT’s tested). I then export the top 19 GAT Table predilections for each of the 1 through 9 number prediction options into my Excel Spreadsheet, run my SDM (Super Duper Macro), and voila . . . I now have a pool of numbers to play for the next drawing. Theoretically, the actual drawn number has a higher likelihood of being within this pool than that associated with the size of the pool itself because, again theoretically, we’ve gleaned some improvement over random play for each of the 6 digits.

A side question before I discuss the SDM in detail: with the Custom Lotteries I’ve created being based on the relatively high odds of 1 in 10 (relatively high comparted to Powerball, Mega Millions, etc), is there even any need to run GAT for so long? Can I reduce the run time to something significantly less than 60 minutes without compromising accuracy? The reason I ask is this run time is the biggest hurdle to more backtesting; if it can be “safely” be reduced to 10 minutes or so I would be more apt to incorporate more backtesting.

And now the SDM discussion. Note that everything I describe for the SDM is done once for each digit (6 runs total), and most of the legwork is done by the SDM with one notable exception.

We already ran GAT after the latest draw as discussed previously. The next step is to export the top 19 GAT tables into the SDM for each of the 1 through 9 digit prediction options. There are 9 tabs each dedicated to the1 through 9 digit prediction options, and when predictions are pasted in, the spreadsheet automatically calculates the frequency of occurrence of each digit 0 through 9 in each tab. The overlying assumption is that the digits that show up more than others are more likely to hit in the next draw.

A side comment: copying and pasting 9 times for each of the 6 Custom Lotteries is the second biggest time driver (behind running GAT for 60 minutes). If future GAT updates can include a way to instantly export ALL predications at once that would be fabulous!

So now we have a “prioritized” list of digits for each of the 1 through 9 digit prediction options. Unfortunately it’s NOT as simple as selecting the top 3 or so ranked digits from any of the 9; in general there is not enough consistency as to where the next draw hits.

But if we look at it from a Rank standpoint across ALL 9 instead, there’s some trending that can be taken advantage of. Let me explain; if we look at, for example, for the 7 draws, how many times the next winning number was found in at least one of a given Rank across the 9, we can clearly see where some Ranks are “Hot” (i.e. the next winning number was found in one or more of that Rank across the 9 for most of the last 7 days), and where some of the Ranks are “Cold” (i.e. the none of the numbers from that Rank across the 9 showed up in the next draw for most of the last 7 days). Put the two together and you have the basis from SDM: presume the next draw is likely to come from the list of numbers compiled from the “Hottest” Rank, and/or is unlikely to come the list of numbers compiled from the “Coldest” Ranks. This assumes the 10 lists are totally independent statistical entities. Clearly they are not, and this is an approximation.

The above was the basis for my two consecutive wins last year that I alluded to earlier. But I’ve improved upon it significantly! I had gotten frustrated when, due to “overlap” between “Hot” and “Cold” lists, a number initially on the “Hot” list was deleted because it was also on the “Cold” list. And of course, in numerous cases, that’s the number that came up in the next draw! So my improvement was this: instead of just marrying the two lists as described above, one at a time I bump up the 3 “Coldest” lists against all 7 remaining lists and look for common numbers! I end up with 9 sets of numbers, then I do a Frequency Count for each digit (more on this below), and lastly I compare the 3 sets of numbers and look for common numbers between the three. With the SDM being finalized I can now run backtests without any human error ; I will post these numbers as soon as I can figure out how to do so in a manner that doesn’t jumble everything up (like a few posts back).

What I do with the aforementioned Frequency Count is evolving. I had found previously that digits that showed up at least once but occurred well below those that showed up the most, were more likely to hit on the next draw. But what was odd is that the behavior was drastically different for the Midday versus Evening drawings. The Evening performed MUCH better (3 straight hits in 18 draws, and the pool size for the 3 hits were 45, 24, and 24). How can that be when they are using the same machine?

So I played around with this Frequency Count Threshold and found that if I only keep those that showed up the most, the pendulum swings in the other direction . . . now the Evening suffers and Midday performs better . . . MUCH better . . . 5 straight hits in 18 draws (pool sizes 24, 30, 36, 30, 40)!

So here’s the question: do I treat them separately and use whichever Frequency Count method works best for each? Or do I stack the two together, treat them as one (since they use the same machine), and fine time the threshold that way?

Wow. This was a mouthful. My apooligies.

While I was typing I found I didn’t hit the Midday draw. Ironically, amongst the 3 pools of numbers SDM identified ( 4 36 126789), the 4 was the only that hit (490 was drawn).

### Who is online

Users browsing this forum: No registered users and 1 guest