This article focuses on note cleaning for a larger data set. While time-consuming, this code demonstrates the full application of this package, with a workable example of hybrid note cleaning.
Hybrid Method
The third method attempts to combine the speed of the First N Character method with the accuracy of the Longest Common Substring method. This works by first applying the First N Character method, then applying the Longest Common Substring method for difficult cases - either when the edit distance is too large for the First N Character method to remove notes, or when the length of the cleaned notes is unusually large. Unusually large is defined as more than a set number of standard deviations above the mean clean note length, and is calculated using the extremeid function.
For this method, a larger dataset may be necessary to demonstrate the presence of outliers, in terms of note size.
library(kableExtra)
library(dplyr)
library(knitr)
fnc<-firstnchar(validation_dataset, "notes", 16, "clean_prints", "page_count")
fnc_printable <- fnc %>% slice(-1) %>% select(-c(algorithm,corrected_notes,clean_prints)) %>% head(3)
fnc_printable <- purrr::map_df(fnc_printable, ~ gsub("[\r\n]", " ", .x)) #removing line breaks
fnc_printable %>% kable()%>%
kable_styling(font_size = 8)| page_count | notes | page_notes | edit_distance |
|---|---|---|---|
| 2 | discharge firearm in business- felony, not guilty | discharge firearm in business- felony, not guilty | 0 |
| 3 | discharge firearm in business- felony, not guilty ski mask. no money, no injuries. 9mm, arrested after confiscated and testing. | ski mask. no money, no injuries. 9mm, arrested after confiscated and testing. | 0 |
| 4 | discharge firearm in business- felony, not guilty ski mask. no money, no injuries. 9mm, arrested after confiscated and testing. terry smith- | terry smith- | 0 |
Above demonstrates the validation dataset cleaned using the First N Character method, with a character difference cutoff of 16 (meaning that up to and including a 15 character difference is acceptable). In the above example, the method is successfully applied (as can be seen comparing the “notes” column to the cleaned “page_notes” column).
However, not all notes are as clear-cut as the example above. For the example below, the true value of the cleaned notes should be as shown in the “corrected_notes” column. The First N Character method fails in this case - because the edit distance (94) is larger than the set threshold (16), no notes would be removed.
library(kableExtra)
library(dplyr)
library(knitr)
fnc<-firstnchar(validation_dataset, "notes", 16, "clean_prints", "page_count")
fnc_printable2 <- head(fnc %>% filter(clean_prints==39, page_count==12) %>% select(-c(algorithm,page_notes,clean_prints, page_count)), 1)
fnc_printable2 <- purrr::map_df(fnc_printable2, ~ gsub("[\r\n]", " ", .x)) #removing line breaks
fnc_printable2 %>% kable()%>%
kable_styling(font_size = 8)| notes | corrected_notes | edit_distance |
|---|---|---|
| In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. terry smith: Q: What training is required to become a firearms examiner with the local police department? A: I received my bachelor’s degree in forensic science and in 2015 I transferred to the crime lab from the crime scene unit. I underwent a two-year training program, which was supervised by experienced firearms examiners; I’ve toured manufacturing facilities and saw how firearms and ammunition were produced; and I’ve attended several national and regional meetings of firearms examiners. A: Yes. I received training in the use of a bullet matching algorithm. This is an algorithm that evaluates the characteristics of two fired bullets, in order to produce a score for the similarity of the bullets, where more similar bullets are more likely to have been fired from the same gun. I attended a workshop on the algorithm on 1/11/2020, held by CSAFE - Center for Statistics and Applications in Forensic Evidence. The training taught me to use the algorithm alongside my personal judgement. I found that my conclusion was reflected in the similarity score produced by the algorithm in all 21 cases. Q: How long have the state police been using the bullet matching algorithm? A: They have been using it since January of 2020. Q: Have you testified in court previously using the bullet matching algorithm? A: Yes, I have, approximately 10 times. Q: As a firearms examiner, have you testified about your conclusions, given the results of your testing? A: Yes, I have. A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the next. Prosecution: Your Honor, at this time I would ask that Terry Smith be qualified as an expert in the field of firearms identification subject to cross examination. Court: Any cross on their credentials? Defense: No, Your Honor. Court: This witness is an expert in the area of firearms identification. They can testify to their opinions as well as facts. Go ahead. A: Yes. In the interior of a barrel there are raised portions called lands and depressed areas called grooves. When a bullet passes down the barrel, a bullet will spin and that gives it stability and accuracy over a distance. Those raised areas are designed by the manufacturer. They’re cut into the barrel. And each particular file has a different combination of lands and grooves. But essentially what those lands do is grip a bullet and spin it, and as that bullet passes down the barrel, it scratches the random imperfections of that barrel into the bullet. A: I place them under the comparison microscope, and I roll the bullet around until I can see the agreement in a particular area: unique surface contour that has sufficient agreement. At that point, when I’ve seen that, I start to rotate the bullets around and I look at all the different lands and grooves, impressions, for that unique detail. When I can see those, that agreement on multiple areas of the bullet, I identify the bullet as having sufficient agreement. A: The algorithm uses 3D measurements to make a comparison between the surface contours of each of the lands on each bullet. These comparisons result in a match score between 0 and 1, where 1 indicates a clear match, and 0 indicates that there is not a match. The bullet is aligned based on the maximum agreement between the lands, and the average match score for the lands is computed. This average score gives an overall match score for the entire bullet. Q: What was the match score between the two test-fired bullets? A: The match score was 0.976. Q: What was the match score between the better-marked test fire bullet and the fired evidence? A: The match score was 0.989. Q: What does this match score indicate about the bullets? A: The match score indicates that there is substantial similarity between the two bullets, which suggests that they were most likely fired from the same barrel. Q: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct? A: Yes. Q: That’s what we call peer review? A: Peer review, yes. Q: Is there something fixed about the amount of what has to be found to constitute sufficient agreement? A: No, there is not a fixed amount or a numerical value for my visual comparison. For the algorithm, however, a score above 0.8 is a general indicator of sufficient agreement. | Q: Is there something fixed about the amount of what has to be found to constitute sufficient agreement? A: No, there is not a fixed amount or a numerical value for my visual comparison. For the algorithm, however, a score above 0.8 is a general indicator of sufficient agreement. | 94 |
In order to determine if there are any extremely long note pages that should be considered for the Longest Common Substring method, the extremeid function is applied. Here, extreme values are defined as those that are larger than 4 standard deviations above the mean page length, based on a condition in the dataframe (“algorithm”), which resulted in different information being presented to note-takers, thus providing different notes. In the example below, the note length is beyond the calculated outlier value, indicating that the notes represent an extreme value.
extreme_dataset <- extremeid(dataset=fnc, extreme=4, clean_notes="page_notes",
pageid="page_count", group_list = c("algorithm"))
extreme_printable <- head(extreme_dataset %>% filter(extreme_value==TRUE, clean_prints==39) %>% subset(select=c("page_notes", "note_length", "outlier", "extreme_value")), 1)
extreme_printable <- purrr::map_df(extreme_printable, ~ gsub("[\r\n]", " ", .x))
extreme_printable %>% kable()%>%
kable_styling(font_size = 8)| page_notes | note_length | outlier | extreme_value |
|---|---|---|---|
| In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. terry smith: Q: What training is required to become a firearms examiner with the local police department? A: I received my bachelor’s degree in forensic science and in 2015 I transferred to the crime lab from the crime scene unit. I underwent a two-year training program, which was supervised by experienced firearms examiners; I’ve toured manufacturing facilities and saw how firearms and ammunition were produced; and I’ve attended several national and regional meetings of firearms examiners. A: Yes. I received training in the use of a bullet matching algorithm. This is an algorithm that evaluates the characteristics of two fired bullets, in order to produce a score for the similarity of the bullets, where more similar bullets are more likely to have been fired from the same gun. I attended a workshop on the algorithm on 1/11/2020, held by CSAFE - Center for Statistics and Applications in Forensic Evidence. The training taught me to use the algorithm alongside my personal judgement. I found that my conclusion was reflected in the similarity score produced by the algorithm in all 21 cases. Q: How long have the state police been using the bullet matching algorithm? A: They have been using it since January of 2020. Q: Have you testified in court previously using the bullet matching algorithm? A: Yes, I have, approximately 10 times. Q: As a firearms examiner, have you testified about your conclusions, given the results of your testing? A: Yes, I have. A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the next. Prosecution: Your Honor, at this time I would ask that Terry Smith be qualified as an expert in the field of firearms identification subject to cross examination. Court: Any cross on their credentials? Defense: No, Your Honor. Court: This witness is an expert in the area of firearms identification. They can testify to their opinions as well as facts. Go ahead. A: Yes. In the interior of a barrel there are raised portions called lands and depressed areas called grooves. When a bullet passes down the barrel, a bullet will spin and that gives it stability and accuracy over a distance. Those raised areas are designed by the manufacturer. They’re cut into the barrel. And each particular file has a different combination of lands and grooves. But essentially what those lands do is grip a bullet and spin it, and as that bullet passes down the barrel, it scratches the random imperfections of that barrel into the bullet. A: I place them under the comparison microscope, and I roll the bullet around until I can see the agreement in a particular area: unique surface contour that has sufficient agreement. At that point, when I’ve seen that, I start to rotate the bullets around and I look at all the different lands and grooves, impressions, for that unique detail. When I can see those, that agreement on multiple areas of the bullet, I identify the bullet as having sufficient agreement. A: The algorithm uses 3D measurements to make a comparison between the surface contours of each of the lands on each bullet. These comparisons result in a match score between 0 and 1, where 1 indicates a clear match, and 0 indicates that there is not a match. The bullet is aligned based on the maximum agreement between the lands, and the average match score for the lands is computed. This average score gives an overall match score for the entire bullet. Q: What was the match score between the two test-fired bullets? A: The match score was 0.976. Q: What was the match score between the better-marked test fire bullet and the fired evidence? A: The match score was 0.989. Q: What does this match score indicate about the bullets? A: The match score indicates that there is substantial similarity between the two bullets, which suggests that they were most likely fired from the same barrel. Q: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct? A: Yes. Q: That’s what we call peer review? A: Peer review, yes. Q: Is there something fixed about the amount of what has to be found to constitute sufficient agreement? A: No, there is not a fixed amount or a numerical value for my visual comparison. For the algorithm, however, a score above 0.8 is a general indicator of sufficient agreement. | 6368 | 6013.49161283854 | TRUE |
The comments that should then be considered using the Longest Common Substring method are those with an extremely large amount of notes (extreme_value=TRUE), and those with an edit distance above the cutoff used in the First N Character method (larger than 15). Note that the previous page of notes is also necessary for comparison in the Longest Common Substring method.
extreme_dataset <- extreme_dataset %>% mutate(apply_lcs = ifelse(!is.na(edit_distance) & (extreme_value==TRUE | edit_distance > 15), TRUE, FALSE))
extreme_printable2 <- head(extreme_dataset %>% filter(apply_lcs==TRUE, clean_prints==39) %>% subset(select=c("page_notes", "edit_distance", "apply_lcs")), 1)
extreme_printable2 <- purrr::map_df(extreme_printable2, ~ gsub("[\r\n]", " ", .x))
extreme_printable2 %>% kable()%>%
kable_styling(font_size = 8)| page_notes | edit_distance | apply_lcs |
|---|---|---|
| In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. terry smith: Q: What training is required to become a firearms examiner with the local police department? A: I received my bachelor’s degree in forensic science and in 2015 I transferred to the crime lab from the crime scene unit. I underwent a two-year training program, which was supervised by experienced firearms examiners; I’ve toured manufacturing facilities and saw how firearms and ammunition were produced; and I’ve attended several national and regional meetings of firearms examiners. A: Yes. I received training in the use of a bullet matching algorithm. This is an algorithm that evaluates the characteristics of two fired bullets, in order to produce a score for the similarity of the bullets, where more similar bullets are more likely to have been fired from the same gun. I attended a workshop on the algorithm on 1/11/2020, held by CSAFE - Center for Statistics and Applications in Forensic Evidence. The training taught me to use the algorithm alongside my personal judgement. I found that my conclusion was reflected in the similarity score produced by the algorithm in all 21 cases. Q: How long have the state police been using the bullet matching algorithm? A: They have been using it since January of 2020. Q: Have you testified in court previously using the bullet matching algorithm? A: Yes, I have, approximately 10 times. Q: As a firearms examiner, have you testified about your conclusions, given the results of your testing? A: Yes, I have. A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the next. Prosecution: Your Honor, at this time I would ask that Terry Smith be qualified as an expert in the field of firearms identification subject to cross examination. Court: Any cross on their credentials? Defense: No, Your Honor. Court: This witness is an expert in the area of firearms identification. They can testify to their opinions as well as facts. Go ahead. A: Yes. In the interior of a barrel there are raised portions called lands and depressed areas called grooves. When a bullet passes down the barrel, a bullet will spin and that gives it stability and accuracy over a distance. Those raised areas are designed by the manufacturer. They’re cut into the barrel. And each particular file has a different combination of lands and grooves. But essentially what those lands do is grip a bullet and spin it, and as that bullet passes down the barrel, it scratches the random imperfections of that barrel into the bullet. A: I place them under the comparison microscope, and I roll the bullet around until I can see the agreement in a particular area: unique surface contour that has sufficient agreement. At that point, when I’ve seen that, I start to rotate the bullets around and I look at all the different lands and grooves, impressions, for that unique detail. When I can see those, that agreement on multiple areas of the bullet, I identify the bullet as having sufficient agreement. A: The algorithm uses 3D measurements to make a comparison between the surface contours of each of the lands on each bullet. These comparisons result in a match score between 0 and 1, where 1 indicates a clear match, and 0 indicates that there is not a match. The bullet is aligned based on the maximum agreement between the lands, and the average match score for the lands is computed. This average score gives an overall match score for the entire bullet. Q: What was the match score between the two test-fired bullets? A: The match score was 0.976. Q: What was the match score between the better-marked test fire bullet and the fired evidence? A: The match score was 0.989. Q: What does this match score indicate about the bullets? A: The match score indicates that there is substantial similarity between the two bullets, which suggests that they were most likely fired from the same barrel. Q: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct? A: Yes. Q: That’s what we call peer review? A: Peer review, yes. | 1062 | TRUE |
Observations where “apply_lcs” is true indicate those difficult cases. They can then be run through the hybrid application of lcs to clean the remaining notes. Due to time, this process was only completed on a single participant.
hybrid_dataset <- lcsclean_hybrid(dataset=subset(extreme_dataset, clean_prints==39), notes="notes", propor=0.333, identifier="clean_prints", pageid="page_count", toclean="apply_lcs")
hybrid_printable1 <- head(hybrid_dataset[hybrid_dataset$apply_lcs==TRUE,] %>% subset(select=c("notes", "page_notes", "lcs_notes", "hybrid_notes")), 1)
hybrid_printable2 <- head(hybrid_dataset[hybrid_dataset$apply_lcs==FALSE & !(hybrid_dataset$page_count %in% c(1,2)),] %>% subset(select=c("notes", "page_notes", "lcs_notes", "hybrid_notes")), 1)
hybrid_printable <- rbind(hybrid_printable2,hybrid_printable1)
hybrid_printable <- purrr::map_df(hybrid_printable, ~ gsub("[\r\n]", " ", .x))
hybrid_printable %>% kable()%>%
kable_styling(font_size = 8)| notes | page_notes | lcs_notes | hybrid_notes |
|---|---|---|---|
| In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. | Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. | NA | Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. |
| In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. terry smith: Q: What training is required to become a firearms examiner with the local police department? A: I received my bachelor’s degree in forensic science and in 2015 I transferred to the crime lab from the crime scene unit. I underwent a two-year training program, which was supervised by experienced firearms examiners; I’ve toured manufacturing facilities and saw how firearms and ammunition were produced; and I’ve attended several national and regional meetings of firearms examiners. A: Yes. I received training in the use of a bullet matching algorithm. This is an algorithm that evaluates the characteristics of two fired bullets, in order to produce a score for the similarity of the bullets, where more similar bullets are more likely to have been fired from the same gun. I attended a workshop on the algorithm on 1/11/2020, held by CSAFE - Center for Statistics and Applications in Forensic Evidence. The training taught me to use the algorithm alongside my personal judgement. I found that my conclusion was reflected in the similarity score produced by the algorithm in all 21 cases. Q: How long have the state police been using the bullet matching algorithm? A: They have been using it since January of 2020. Q: Have you testified in court previously using the bullet matching algorithm? A: Yes, I have, approximately 10 times. Q: As a firearms examiner, have you testified about your conclusions, given the results of your testing? A: Yes, I have. A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the next. Prosecution: Your Honor, at this time I would ask that Terry Smith be qualified as an expert in the field of firearms identification subject to cross examination. Court: Any cross on their credentials? Defense: No, Your Honor. Court: This witness is an expert in the area of firearms identification. They can testify to their opinions as well as facts. Go ahead. A: Yes. In the interior of a barrel there are raised portions called lands and depressed areas called grooves. When a bullet passes down the barrel, a bullet will spin and that gives it stability and accuracy over a distance. Those raised areas are designed by the manufacturer. They’re cut into the barrel. And each particular file has a different combination of lands and grooves. But essentially what those lands do is grip a bullet and spin it, and as that bullet passes down the barrel, it scratches the random imperfections of that barrel into the bullet. A: I place them under the comparison microscope, and I roll the bullet around until I can see the agreement in a particular area: unique surface contour that has sufficient agreement. At that point, when I’ve seen that, I start to rotate the bullets around and I look at all the different lands and grooves, impressions, for that unique detail. When I can see those, that agreement on multiple areas of the bullet, I identify the bullet as having sufficient agreement. A: The algorithm uses 3D measurements to make a comparison between the surface contours of each of the lands on each bullet. These comparisons result in a match score between 0 and 1, where 1 indicates a clear match, and 0 indicates that there is not a match. The bullet is aligned based on the maximum agreement between the lands, and the average match score for the lands is computed. This average score gives an overall match score for the entire bullet. Q: What was the match score between the two test-fired bullets? A: The match score was 0.976. Q: What was the match score between the better-marked test fire bullet and the fired evidence? A: The match score was 0.989. Q: What does this match score indicate about the bullets? A: The match score indicates that there is substantial similarity between the two bullets, which suggests that they were most likely fired from the same barrel. Q: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct? A: Yes. Q: That’s what we call peer review? A: Peer review, yes. | In this case, the defendant - Richard Cole - has been charged with willfully discharging a firearm in a place of business. This crime is a felony. Mr. Cole has pleaded not guilty to the charge.Police received a 911 call from a convenience store clerk stating that a man had entered the store, pulled out a firearm, fired a shot into the ceiling, and then demanded money. The man ended up leaving the store without receiving any money and no one was hurt by the gunshot. The store clerk took the stand and stated that she did not get a close enough look at the robber’s face to make an identification, because the robber wore a ski mask. As the cashier was pulling out the money to hand to him, she pressed a hidden button that activated an alarm and called the police. Startled by the alarm bells, the robber rushed out of the store. Next, the detective testified that he arrived at the convenience store and interviewed the clerk. The detective recovered the 9mm bullet from the ceiling of the store, which was collected for forensic analysis. Two days later, a police officer pulled over Richard Cole for speeding. During a search of the Defendant’s vehicle, the detective located a 9mm handgun, which was legally licensed to the Defendant. Because this gun was the same caliber as the one used in the convenience store shooting, it was confiscated for testing. Richard Cole was subsequently arrested and charged with willfully discharging a firearm in a place of business. terry smith: Q: What training is required to become a firearms examiner with the local police department? A: I received my bachelor’s degree in forensic science and in 2015 I transferred to the crime lab from the crime scene unit. I underwent a two-year training program, which was supervised by experienced firearms examiners; I’ve toured manufacturing facilities and saw how firearms and ammunition were produced; and I’ve attended several national and regional meetings of firearms examiners. A: Yes. I received training in the use of a bullet matching algorithm. This is an algorithm that evaluates the characteristics of two fired bullets, in order to produce a score for the similarity of the bullets, where more similar bullets are more likely to have been fired from the same gun. I attended a workshop on the algorithm on 1/11/2020, held by CSAFE - Center for Statistics and Applications in Forensic Evidence. The training taught me to use the algorithm alongside my personal judgement. I found that my conclusion was reflected in the similarity score produced by the algorithm in all 21 cases. Q: How long have the state police been using the bullet matching algorithm? A: They have been using it since January of 2020. Q: Have you testified in court previously using the bullet matching algorithm? A: Yes, I have, approximately 10 times. Q: As a firearms examiner, have you testified about your conclusions, given the results of your testing? A: Yes, I have. A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the next. Prosecution: Your Honor, at this time I would ask that Terry Smith be qualified as an expert in the field of firearms identification subject to cross examination. Court: Any cross on their credentials? Defense: No, Your Honor. Court: This witness is an expert in the area of firearms identification. They can testify to their opinions as well as facts. Go ahead. A: Yes. In the interior of a barrel there are raised portions called lands and depressed areas called grooves. When a bullet passes down the barrel, a bullet will spin and that gives it stability and accuracy over a distance. Those raised areas are designed by the manufacturer. They’re cut into the barrel. And each particular file has a different combination of lands and grooves. But essentially what those lands do is grip a bullet and spin it, and as that bullet passes down the barrel, it scratches the random imperfections of that barrel into the bullet. A: I place them under the comparison microscope, and I roll the bullet around until I can see the agreement in a particular area: unique surface contour that has sufficient agreement. At that point, when I’ve seen that, I start to rotate the bullets around and I look at all the different lands and grooves, impressions, for that unique detail. When I can see those, that agreement on multiple areas of the bullet, I identify the bullet as having sufficient agreement. A: The algorithm uses 3D measurements to make a comparison between the surface contours of each of the lands on each bullet. These comparisons result in a match score between 0 and 1, where 1 indicates a clear match, and 0 indicates that there is not a match. The bullet is aligned based on the maximum agreement between the lands, and the average match score for the lands is computed. This average score gives an overall match score for the entire bullet. Q: What was the match score between the two test-fired bullets? A: The match score was 0.976. Q: What was the match score between the better-marked test fire bullet and the fired evidence? A: The match score was 0.989. Q: What does this match score indicate about the bullets? A: The match score indicates that there is substantial similarity between the two bullets, which suggests that they were most likely fired from the same barrel. Q: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct? A: Yes. Q: That’s what we call peer review? A: Peer review, yes. | A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the nextQ: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct?A: Yes.Q: That’s what we call peer review?A: Peer review, yes. | A: No. When firing a firearm there is a dynamic process because it is a contained explosion. When the firing pin hits the primer, which is basically the initiator, what gets it going, it will explode, burn the gun powder inside the casing, and the bullet will travel down the barrel, picking up the microscopic imperfections of the barrel, and the cartridge case will slam rearward against the support mechanism. During that dynamic process, each time it happens, a bullet will be marked slightly differently from one to the nextQ: Is it the local police department’s protocol to have somebody else who’s a firearms tool mark examiner in your lab review that report, review your work, and determine if it’s correct?A: Yes.Q: That’s what we call peer review?A: Peer review, yes. |
The above notes show an easy and a difficult case for the First N Character method (indicated as “page_notes”). In the first row, the FNC method is able to correctly distinguish between old and new notes. In this case, LCS does not need to be applied, and the FNC notes are copied into the “hybrid_notes” column. In the second row, the FNC method fails, leading to the application of LCS. These notes are then included in the “hybrid_notes” column, so that this represents the full dataset of cleaned notes.