Full text
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 1 of 13 PageID #: 588
UNITED STATES DISTRICT COURT
WESTERN DISTRICT OF KENTUCKY
LOUISVILLE DIVISION
UNITED STATES OF AMERICA PLAINTIFF
v. No. 3:21-cr-89-BJB
SALVADOR ABDUL JONES DEFENDANT
Supplemental Opinion Regarding the Partial Exclusion of Expert
Testimony
“If highly consequential evidence emerges from what looks like an
indecipherable computer program to most non-scientists, non-statisticians, and non-
programmers, it is imperative that qualified individuals explain how the program
works and ensure that it produces reliable information about the case.” United States
v. Gissantaner, 990 F.3d 457, 463 (6th Cir. 2021).
That Sixth Circuit opinion addressed DNA testing in a gun case. This case
concerns cell-phone data in a bank-robbery prosecution. But the reasoning applies
almost directly. On more or less the same grounds articulated in Gissantaner, the
Defendant here objected to testimony offered by a police officer regarding a computer
program and presentation extrapolating from historical cell-site location information.
“CSLI” is the technical and unharmonious term that lawyers and judges appear to
have settled on when discussing a particular chain of evidentiary inferences: the
apparent locations of people, based on the approximate locations of their mobile
phones, derived from the known locations of cell towers.
The law of CSLI, however, remains less settled than the lingo. This chain of
inferences—particularly when aided by software—is perhaps more contestable than
it appears to many, including some who regularly confront such information. Before
this trial, for example, neither side anticipated that the officer’s CSLI testimony
would amount to expert opinion subject to scrutiny under Federal Rule of Evidence
702. So the Court lacked an opportunity to consider in advance a motion to limit or
exclude those opinions under the principles of Daubert v. Merrell Dow
Pharmaceuticals, Inc., 509 U.S. 579, 589 (1993). When the testimony drew an
objection at trial, the Court halted the officer’s direct examination, held a Daubert
hearing, and substantially (but not entirely) limited the scope and content of his
testimony. Given the time-sensitive nature of that mid-trial ruling from the bench,
and the importance of this issue to the parties, the Court issues this supplemental
1
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 2 of 13 PageID #: 589
opinion explaining the reasons for its decision in more detail than was practical while
a jury was waiting for proceedings to resume.1
In short, the Government sought to question a law-enforcement witness about
two types of CSLI. One was permissible; the other was not.
First, the witness testified regarding phone records maintained in the ordinary
course by T-Mobile, the Defendant’s cell-phone provider. Those records reported the
street addresses of Louisville cell towers the Defendant’s phone interacted with at
specific times.
These are straightforward documents that the witness handles on a regular
basis as a member of the Louisville Metro Police Department’s Digital Forensics Unit.
Based on them, the witness could reliably “rule out” areas where the Defendant’s cell
phone would not have been during the time periods surrounding the robberies. He
could also testify that the transmission of various calls or text messages was at least
consistent with the phone’s presence in a particular part of town. And based on his
training and experience, the witness could explain on direct- and cross-examination
the basis for his inferences, confidence, and reservations. This opinion testimony was
appropriate under Rule 702.
Second, and far more problematic, was the witness’s attempt to show the jury
a detailed map purporting to “rule in”—down to the block—the specific areas where
the Defendant’s phone could and could not have been when it made the calls and
messages at issue. As a visual aid to his testimony, the witness planned to represent
these locations using an animated illustration of the purported location of the
Defendant’s phone on a Google Earth Pro map.
The proposed presentation was precise and compelling. Too much so, as it
turned out. That’s because the sponsoring witness was not actually responsible for
calculating or otherwise divining the areas where he said the Defendant’s phone
must’ve been when it transmitted a call or text message through a nearby tower. Nor
could the witness explain how someone or something else charted those perimeters.
The maps the Government wanted to show the jury were generated by a proprietary
software program called TraX. The witness was involved in the maps’ creation—but
in a very limited capacity: he collected the phone-company records and uploaded
them, in spreadsheet format, to the software that spat out the maps. The witness did
not develop, study, or apparently even understand the algorithms or methodology
that concluded a phone must’ve been on one side of a street, and not the other, when
1 Nothing in this opinion is intended to expand or narrow the decision or reasoning the
Court expressed in orally resolving this motion during the trial. The opinion is meant only
to clarify. Should either side interpret the written order differently, it may raise any such
concern with the Court at any time before the sentencing hearing and final judgment.
2
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 3 of 13 PageID #: 590
it transmitted a call or text message. Nor could he discuss the reliability, errors, or
limitations of the software. To the witness, lawyers, and judge—like “most non-
scientists, non-statisticians, and non-programmers”—the output appeared to emerge
from “an indecipherable computer program,” which no one in the courtroom could
examine or explain for the jury. Gissantaner, 990 F.3d at 463. Because this sort of
testimony does not satisfy the standards for reliable expert opinion under Rule 702,
the Court excluded the Government’s use of and testimony about the TraX maps.
A. Procedural Posture
A grand jury charged Salvador Jones with robbing five federally insured banks
in Louisville during the spring of 2021. See DN 1 (citing 18 U.S.C. § 2113(a)). At
trial, the Government offered surveillance videos, witness testimony, recorded
confessions, and incriminating statements (among other evidence) to prove its case.
It also presented the testimony of Louisville Metro Police Officer Timothy O’Daniel.
As indicated by a slide deck he disclosed to the Defendant and the Court, O’Daniel
planned to testify regarding the location of Jones’s phone around the time of each
robbery. The deck contained maps showing shaded areas near the location of cell
towers with which Jones’s phone communicated—implying that Jones (or at least his
phone) must’ve been somewhere within the marked areas. The Government
apparently planned to admit, as exhibits for the jury, the slide deck as well as T-
Mobile records detailing Jones’s cell-tower connections.
Before trial, the Government identified O’Daniel as a witness—but not an
expert witness subject to Federal Rule of Evidence 702 or Federal Rule of Criminal
Procedure 16(a)(1)(G). At the final pretrial conference on September 30, the Court
asked whether both sides agreed that O’Daniel should be treated as a lay witness.
They did. To be sure, defense counsel expressed concern regarding the late disclosure
of discovery material, including emails about O’Daniel’s use of the TraX software.
But his proposed remedy was a continuance (which his client resisted and the Court
denied) rather than exclusion of any related testimony.
When trial began, the Government shared with the defense the exhibits
O’Daniel planned to use during his direct examination. His proposed testimony, as
reflected in the presentation and other exhibits, contained two main components: T-
Mobile spreadsheets and TraX maps.2
The spreadsheets contained cellphone records the Government received (by
warrant) from T-Mobile. This showed the timing of calls and text messages from a
cell number registered to Jones. Rows listed the phone number; the latitude,
longitude, and street address of the cell tower with which the phone connected; the
2 The government intended to admit both the spreadsheets and the maps as trial exhibits.
The spreadsheets were ultimately admitted; the TraX maps were not.
3
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 4 of 13 PageID #: 591
time and duration of the connection; and whether the call or text was ingoing or
outgoing from Jones’s phone.
The series of maps showed the locations of those towers, each surrounded by a
shaded perimeter depicting its purported coverage area. This implied that Jones’s
phone was located within those perimeters when it connected with the tower in
question. These Google Earth Pro maps also noted the locations of the robberies—.
in each instance near the relevant cell tower at the relevant time. Pretty damning
stuff.
The Government neither generated these maps nor received them from T-
Mobile, however. Rather, O’Daniel uploaded the T-Mobile records to a software
program called TraX, which produced the maps for the police department.
TraX is a proprietary program originally developed by a company called ZetX
and later acquired by LexisNexis. In offering TraX to police departments and other
potential customers, Lexis describes TraX as a tool that allows investigators to
“[q]uickly visualize and analyze call detail records obtained through legal process …
and third-party device geolocation information, powered by identity data allowing
law enforcement to find a subject’s location more quickly and efficiently than ever
before.” ACCURINT TRAX, https://risk.lexisnexis.com/products/accurint-trax (last
visited Dec. 2, 2022). In this case, of course, the Government sought to use TraX not
to find a subject, but to prove his proximity to a series of bank robberies.
B. Historical Cell-Site Location Information
In recent decades, historical cell-site location information has become a
popular tool for law enforcement officials trying to reconstruct the movements of
suspected criminals. See generally Carpenter v. United States, 138 S. Ct. 2206, 2212–
13 (2018); Torrence v. Commonwealth, 603 S.W.3d 214, 226 (Ky. 2020). CSLI can
serve as an efficient way to learn about a person’s whereabouts in connection with
potential criminal activity. Carpenter, 138 S. Ct. at 2217–18. Courts have generally
allowed testimony regarding CSLI at trial. See, e.g., Torrence, 603 S.W.3d at 226;
United States v. Hill, 818 F.3d 289, 297 (7th Cir. 2016) (“District courts that have
been called upon to decide whether to admit historical cell-site analysis have almost
universally done so.”). Depending on the nature and use of the information, however,
its introduction can pose evidentiary challenges. See Hill, 818 F.3d at 296–97.
Admission of the underlying phone-company records themselves is typically
uncontroversial; if relevant, the records may be admitted consistent with the hearsay
rules. Courts of appeals haven’t treated automatically generated call logs as
“statements” subject to Rule 801’s limitation. See United States v. Lamons, 532 F.3d
1251, 1261–63 (11th Cir. 2008) (call reports “not ‘statements’ within the meaning of
the Confrontation Clause or Federal Rule of Evidence 801(a)”); see also Lyngaas v.
4
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 5 of 13 PageID #: 592
Curaden AG, 992 F.3d 412, 431 (6th Cir. 2021) (citing Lamons and Rule 801 to affirm
the admission of fax records).
And even if such records amounted to statements subject to the hearsay rules,
Federal Rule of Evidence 803(6) exempts records made at the time of the call and
kept in the course of regularly conducted activity by the cellular provider. See United
States v. Yeley-Davis, 632 F.3d 673, 678 (10th Cir. 2011). The records typically
identify the relevant phone number, the name associated with that number and
account, the times or durations of calls or text messages, and the location or address
of towers that the phone used to transmit specific calls and messages. See Carpenter,
138 S. Ct.at 2211–12 (2018) (describing regular collection of CSLI data by phone
companies for business purposes). The T-Mobile records admitted in this case are
typical of the genre, and weren’t challenged on this basis.
This information may prove tremendously revealing, given the nearly
universal relationship between Americans’ cell phones and their hands, ears, and
pockets. See id. at 2218. But the records alone say little about the specific location
of a phone relative to a tower; they simply indicate which tower the phone connected
to. The documents often do not provide details that might pinpoint more precise
locations: for example, which face of a tower transmitted calls or messages to and
from the phone, whether and how the network handed off a call between towers, or
the triangulation of a phone’s location relative to multiple towers. See, e.g., United
States v. Evans, 892 F. Supp. 2d 949, 956 (N.D. Ill. 2012) (noting calls are not always
routed to the nearest tower).
In other words, CSLI data doesn’t equate to GPS tracking in terms of accuracy
and precision, despite some similarities in its production and collection. See
Carpenter, 138 S.Ct. at 2216. GPS is far more specific (and not at issue in this case).
If GPS is the technological equivalent of Hansel and Gretel carefully marking their
trail with crumbs, CSLI is more reminiscent of Johnny Appleseed sporadically tossing
seeds over his shoulder: Both types of debris can help solve or prove a case, but the
latter requires more cautious interpretation. Cf. Carpenter, 132 S. Ct. at 2218, 2232–
33 (Kennedy, J., dissenting) (discussing differences in precision).
Sometimes CSLI is used to estimate a phone’s location based on the
(reasonable but not ironclad) assumption that a cell phone typically connects to the
tower with the strongest signal—usually the closest tower. Hill, 818 F.3d at 295.
Depending on factors like topography, environmental conditions, and the angle and
direction of antennas, a phone may connect to a tower that is further away. Id. at
295–96. Correspondingly, each cell tower generally can reach a geographic area
whose size varies depending on tower density, network traffic, weather, and terrain.
Different towers will of course have overlapping coverage; otherwise our calls would
drop dead far more often than they do. Id.
5
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 6 of 13 PageID #: 593
Despite this utility and popularity at trial, the reliability of CSLI remains
uncertain. “[W]hile historical cell-site location evidence has been allowed into
evidence routinely in criminal trials, some courts have expressed skepticism
regarding aspects of its reliability.” Paul W. Grimm, Admissibility of Historical Cell
Phone Location Evidence, 44 Litigation 53, 54–55 (2018). Indeed, “[n]o federal court
of appeals has yet said authoritatively that historical cell-site analysis is admissible
to prove the location of a cell phone user.” Hill, 818 F.3d at 297.
C. Expert or Lay Testimony?
Nor have courts definitively resolved an important threshold question
regarding its presentation: when testimony regarding CSLI must be treated as expert
opinion rather than lay testimony. Under Federal Rule of Evidence 602, a lay witness
may testify only to matters within his personal knowledge. Rule 701 also limits the
opinion testimony of a lay witness to statements that are “rationally based on the
witness’s perception” and “not based on scientific, technical, or other specialized
knowledge” that would require expert qualification. At trial, defense counsel argued
that O’Daniel was offering expert opinions and had to be qualified as such. Under
Rule 702, a witness must be “qualified as an expert by knowledge, skill, experience,
training, or education” before he may offer opinion testimony based on his “scientific,
technical, or other specialized knowledge.”
Federal and state courts have split on the question whether a witness who
relies on CSLI to prove a suspect’s location must be qualified as an expert. Some
have allowed officers or phone-company employees to testify as lay witnesses
regarding how CSLI is generated and what it can imply about the location of an
individual’s phone. The Kentucky Supreme Court recently held that the mapping of
cell towers based on call records does not require expert knowledge. See Torrence,
603 S.W.3d at 225–26 (distinguishing expert testimony based on inferences based on
tower locations). And the Fourth Circuit held that a lay witness did not improperly
offer opinion testimony regarding CSLI, despite discussing the factors affecting its
accuracy, because he did not “engage in any analysis comparing the factors”
influencing a cell connection or show how those factors resulted in “any particular
connection.” United States v. Graham, 796 F.3d 332, 363–65 (4th Cir. 2015); see also
United States v. Henderson, 564 F. App’x 352, 363–64 (10th Cir. 2014) (allowing lay
opinion testimony on CSLI to show that an individual was not in a particular
location).
But other courts, including the Sixth Circuit, have required expert
qualification for certain types of CSLI testimony because “[t]estimony ‘concerning
how cell phone towers operate … involve[s] specialized knowledge not readily
accessible to any ordinary person.’” United States v. Reynolds, 626 F. App’x 610, 614
(6th Cir. 2015) (second alteration in original) (quoting Yeley-Davis, 632 F.3d at 684).
6
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 7 of 13 PageID #: 594
The Second and Seventh Circuits appear to agree, holding that “testimony on how
cell phone towers operate constitutes expert testimony.” United States v. Natal, 849
F.3d 530, 533 (2d Cir. 2017); Hill, 818 F.3d at 295–96 (similar).
At this trial, the Government’s attempt to introduce testimony about CSLI
through Detective O’Daniel drew a defense objection. Initially counsel argued the
TraX maps raised a Confrontation Clause problem under Crawford v. Washington,
541 U.S. 36 (2004), and Bullcoming v. New Mexico, 564 U.S. 647 (2011), because
counsel couldn’t cross-examine the technologist who generated the maps. After the
Court inquired about a potentially missing witness, however, it became clear that
none existed: the Government wasn’t aware of any technologist who generated the
maps, at least not directly. Rather, software did the trick. O’Daniel simply uploaded
the T-Mobile cell-phone data to TraX, which used its own algorithms and Google
Earth Pro to produce the animated maps the Government sought to introduce.
That raised a separate but related problem. Even though O’Daniel was subject
to cross-examination, did his testimony about the TraX software program and the
maps it generated “involv[e] specialized knowledge not readily accessible to any
ordinary person?” Reynolds, 626 F. App’x at 614 (quotation omitted). Surely it did:
O’Daniel himself couldn’t even explain the methodology and limits of the software
program. So how could a lay juror be expected to understand algorithmic CSLI? The
TraX mapping is a subject beyond even the broadest construction of lay testimony.
And to the Government’s credit, it eventually realized as much, conceding that
testimony about the mapping fell within the purview of Rule 702 and Daubert.
What about testimony regarding the T-Mobile records? This is a closer call.
On the one hand, these business records simply state which tower or towers
connected with the phone. Linking a tower address, phone, and crime scene wouldn’t
seem to stretch the reasoning and background knowledge of a lay juror. That is one
reason why some courts, in some similar situations, have ruled that expert opinion is
not necessarily required. See, e.g., Graham, 796 F.3d at 364.
The risk with this approach, at least as applied to testimony like O’Daniel’s, is
that the mere fact that a call connected with a given tower is not by itself directly
relevant to the factual question at hand: where was the caller? Inferring the answer
requires understanding at least a bit about cellular-network technology. How and
why does a phone typically connect with a particular cell tower? What is the typical
range of a tower? Does the reach change based on the density of towers, buildings,
and phones in a given spot? And, ultimately, what can a factfinder reasonably infer
about where a particular call or message originated?
Perhaps by now most American adults understand the basics about how
ubiquitous cell phones communicate through the air with ubiquitous cell towers to
facilitate ubiquitous calls and text messages. We may be approaching a level of lay
7
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 8 of 13 PageID #: 595
understanding approaching matters courts typically expect jurors to follow without
the aid of an expert—like driving a car, Heritage Mut. Ins. Co. v. Reck, 127 F. App’x
194, 199 (6th Cir. 2005), or understanding basic financial statements, United States
v. Madison, 226 F. App’x 535, 544 (6th Cir. 2007).
Much less clear, however, is the proposition that “most jurors and judges”
understand the geolocational byproducts of those communications. Gissantaner, 990
F.3d at 463. As the questions above demonstrate, CSLI relies not only on an
understanding of how mobile-phone technology works, but also on an understanding
of what its use does and doesn’t imply about caller locations. These issues are less
like driving and more like fingerprinting: most of us recognize fingerprints and
understand their investigatory value in the abstract, but that doesn’t mean we
laypersons can reliably derive conclusions about particular prints and matches. See
United States v. Hatcher, 513 F. App’x 581, 584 (6th Cir. 2013) (affirming use of expert
fingerprint testimony). It’s similarly doubtful that most lay jurors could reasonably
perceive the likely coverage area of a cell tower. Explaining when and why a phone
may connect to different towers based on “geography and topography, the angle,
number, and directions of the antennas … the technical characteristics of the relevant
phone, [or] environmental and geographical factors” is not stuff most of us should try
at home. Hill, 818 F.3d at 295–96 (quotation omitted).
Given the technical and unfamiliar nature of location information derived from
use of the cellular-telephone network, therefore, it is “imperative that qualified
individuals explain how the [technology] works and ensure that it produces reliable
information.” Gissantaner, 990 F.3d at 463. So testimony that describes the likely
location of a person (or his or her phone) using CSLI should generally be presented
by a witness who can draw on expertise. That is, “knowledge, skill, experience,
training, or education,” FED. R. EVID. 702, that allows the witness to explain how the
network operates, what locational data it produces, and the relevant limits on the
precision and accuracy of that data, see, e.g., Reynolds, 626 F. App’x at 614.
D. O’Daniel’s Qualifications and the Reliability of his Methods
After exploring these distinctions at length following the defense objection to
O’Daniel’s testimony, the Government agreed: it tendered him as an expert in historic
CSLI. Did his qualifications and methods suffice under Rule 702? Only in part. He
could explain the T-Mobile call records and their general implications regarding the
location of Jones’s phone. He could not, however, explain how the proprietary TraX
software program used those records to generate its specific radii for the location of
Jones’s phone.
Rule 702 requires any expert witness addressing a technical or scientific topic
to be qualified and the testimony to be both reliable and relevant. Daubert, 509 U.S.
at 589. Once challenged, the Government must prove as much by a preponderance.
8
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 9 of 13 PageID #: 596
Pride v. BIC Corp., 218 F.3d 566, 578 (6th Cir. 2000). And the trial court must
exercise its considerable discretion as a “gatekeeper” in making these determinations
according to the facts and circumstances of the case. Kumho Tire Co. v. Carmichael,
526 U.S. 137, 142, 145 (1999).
The Defendant argued generally that O’Daniel’s proposed CSLI testimony was
not reliable. Jones also specifically challenged the reliability of O’Daniel’s planned
use of the animated maps charted by the TraX program.
O’Daniel’s qualifications and the reliability of his proffered testimony both
turned on two related issues: (1) his understanding of the underlying methodology
and information, and (2) the degree of precision with which he proposed to convey his
opinions to the jury.
1. Maps. The Daubert hearing addressed the TraX maps first. The
Government sought to introduce the maps and his testimony about them to
affirmatively prove that Jones was located near each of the five robberies at the time
that they occurred. But O’Daniel’s knowledge was limited: he could not describe how
TraX produces estimated location ranges for particular cell sites. He described
attending some TraX-provided training on how to use the software (and, notably, how
to testify effectively about it). But O’Daniel conceded that his use of the data in this
investigation was limited to uploading the T-Mobile cell-site data into the TraX
program. The software then generated the file that mapped estimated location
ranges onto the map supplied by Google Earth Pro.
Nor did O’Daniel conduct a so-called “drive test” in this case, though he has in
others. A drive test is a corroboration method used by officers to verify the connection
ranges associated with a given tower. See, e.g., United States v. Morgan, 292 F. Supp.
3d 475, 478–82 (D.D.C. 2018) (describing the uses and mechanics of drive testing);
United States v. Reynolds, No. 1:20-cr-24, 2021 WL 3750156, at *3 (W.D. Mich. Aug.
25, 2021) (drive testing can bolster reliability of CSLI testimony under Daubert).
Rather than simply trusting information from a third-party’s software program or
database, officers sometimes travel to the locations at issue and attempt to connect
to the relevant tower. If they can, that supports their understanding of the tower’s
range. But that corroboration was unavailable here.
O’Daniel’s maps and proposed testimony nevertheless purported to identify
where Jones’s phone was—within a precisely marked area that ruled in or out specific
blocks and streets—near in time to the robberies. This meant its proponent had to
ensure that the witness was prepared to tell the jury how CSLI works and explain
not only … the technique’s potential pitfalls, but also … the relative imprecision of
the information he gleaned.” Hill, 818 F.3d at 299.
9
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 10 of 13 PageID #: 597
O’Daniel was unable to make any such showing. He testified (outside the
hearing of the jury) that the tower ranges shown in his presentation were calculated
and illustrated by the TraX software, rather than by LMPD or T-Mobile. But he could
not demonstrate that he understood the TraX algorithm’s basis for placing a phone
within a particular geographic perimeter when it connected with a nearby tower. It
was not at all clear, therefore, whether or how the TraX interpretation of CSLI could
place a phone within the radius shown in O’Daniel’s original slide deck. See Hill, 818
F.3d at 298–99; Reynolds, 626 F. App’x at 617.
But part of his maps made sense: he could explain why he knew Jones’s phone
was at least in Louisville. The tower locations and call times, O’Daniel explained,
were at least consistent with—if not conclusive of—his phone’s presence near the
robberies. This implicitly rejected any notion that the phone was somewhere far
afield.
That approach is consistent with a recent unpublished Sixth Circuit decision.
Reynolds approved the use of CSLI to rule out particular locations of a phone (and
therefore the user of that phone) based on the phone’s connections with towers distant
from a given neighborhood or area. But the Court of Appeals expressed great
skepticism about the use of CSLI to rule in—or affirmatively establish—a person’s
location in a particular spot at a particular time. The Reynolds court affirmed the
admission of CSLI to show that other potential perpetrators were not in the area of a
crime, based on the connections of their phones to towers far away from the scene
during the relevant times. 626 F. App’x at 617–18. The panel cast doubt, however,
on the admissibility of CSLI to show that a caller was located in a specific sector
associated with a single cell tower. Id. at 617.
This distinction relied on the different methodologies or deductions involved:
affirmatively placing a caller in a particular location depended on the “questionable
assumption that each call connected to one of the nearest towers.” Id. As prior
decisions have recognized, a cell phone generally connects to the tower that provides
the strongest signal; for a variety of reasons—including weather, obstructions, and
cell-network volume—that might not be the closest tower. Id. at 615; see also Hill,
818 F.3d at 295–96. So the Sixth Circuit cautioned that a prosecution witness likely
could not rely on a single cell-tower connection, without more, to show a person’s
phone was near that tower.
In Hill, the Seventh Circuit applied a similar analysis to admit CSLI to prove
that an individual was in a broad area, but rejected its use to prove that the defendant
was in a narrower range. 818 F.3d at 298–99. “Historical cell-site analysis can show
with sufficient reliability that a phone was in a general area, especially in a well-
populated one,” but testimony to that effect is admissible only if the expert makes
clear “the relative imprecision of the information he gleaned.” Id.
10
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 11 of 13 PageID #: 598
In contrast, using CSLI to show that a caller is not in a particular area, or that
a phone was likely within a broad area surrounding a tower, is generally acceptable
under the reasoning of the Reynolds court. While a call might connect through a
nearby tower that is not the nearest, it’s typically implausible that a call or message
would’ve moved through a far-away tower. “[I]t is reliable to assume that a call would
not connect to a tower that was many sectors away.” Reynolds, 626 F. App’x at 617.
And the difference between near and far towers is not merely a matter of conjecture:
“there are identifiable, measurable, and scientifically accepted factors that determine
a cell tower’s maximum coverage.” Id.; cf. Gissantaner, 990 F.3d at 463 (citing
Daubert, 509 U.S. at 593–94). This distinction also accords with the Seventh Circuit’s
thoughtful decision in Hill, 818 F.3d at 298, which described as “uncontroversial” the
use of CSLI to “prov[e] where [the defendant] was not.”
The Government’s initial slide deck included a slick animated presentation
produced not by O’Daniel but by TraX. Nor did O’Daniel understand much about how
the software produced it. His role was to feed the software the historical cell-phone
records supplied by T-Mobile (quite helpfully in a file type susceptible to uploading to
TraX) and to testify about the PowerPoint display that TraX and Google Earth Pro
then generated. In discussing where a phone may have been found during a
particular call or message, he could not explain—even in broad strokes—why one side
of a street was shaded and the other not. Correspondingly, he could not testify about
how or why the shaded areas reflected any degree of uncertainty or qualification. To
understand this material, a jury would’ve benefitted from hearing knowledgeable
testimony about what the software can and can’t do well—and when it’s necessary to
view the output in shades of gray.
Yet the map’s perimeters were marked in stark black and white. This raises
concerns under Daubert; it’s the sort of technical testimony that could easily mislead
jurors about the precision and accuracy of tools they don’t understand. Gissantaner,
990 F.3d at 463. As discussed above, no established record or literature available to
the Court shows that CSLI can precisely locate a user’s cell phone. The Government
didn’t use GPS, “pinging,” or triangulation—more precise and reliable methods for
assessing historical geolocation data. Because the ranges were produced by the TraX
software, the Court lacked any information on how the locations were generated.
Other courts presented with map ranges generated by TraX have confronted similar
issues. See People v. Valdez, C087046, 2022 WL 556833, at *19 n.24 (Cal. Ct. App.
Feb. 24, 2022) (“exactly what TraX does” to map coverage areas of particular towers
“was not clear”). During the hearing, O’Daniel indicated that the TraX software uses
an algorithm to calculate the range—and that the algorithm relied on more than just
the call and tower data obtained from the cell-phone provider. Those other factors,
of course, are proprietary and (at least in this case) unavailable for scrutiny or cross-
examination.
11
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 12 of 13 PageID #: 599
O’Daniel’s proffered testimony compares unfavorably with that of a TraX
developer admitted in a similar case within this circuit. See Reynolds, 2021 WL
3750156, at *1 (W.D. Mich. Aug. 25, 2021). The judge allowed testimony about a TraX
presentation after the prosecution presented testimony from the program’s creator
about its testability and reliability. Id. Here, the Government offered no testimony
on how the TraX program operates or how it has been received by the scientific
community. Given the Court’s inability to assess the reliability of the disputed
locations and ranges produced by TraX, Rule 702 demands the exclusion of the results
of an “indecipherable computer program.” Gissantaner, 990 F.3d at 464 (“The
question on the table is whether a method can be ‘assessed for reliability[.]’”); cf.
Jones, 918 F. Supp. 2d at 6 (excluding cell-phone location diagram depicting precise
wedge-shaped areas that would mislead the jury regarding the precision of estimated
CSLI location information).
As a result of this decision to grant in part Jones’s motion to exclude the
officer’s testimony, the Court barred O’Daniel from testifying about TraX, explained
that limitation to him outside the hearing of the jury, and required him to submit a
new map based on the basic information reflected in the T-Mobile phone records.
2. T-Mobile records. O’Daniel’s inferences from the T-Mobile spreadsheets,
derived without the use of software, fared far better. At the Daubert hearing, he
explained his knowledge of the cellular network, the call-log records provided by T-
Mobile, and the inferences he could draw regarding a phone’s location based on such
records. As described above, these list the cell towers actually used by the
Defendant’s phone for the calls and messages in question. O’Daniel then described
how these records showed the locations of cell towers to which Jones’s phone
connected. He also discussed how he verified the T-Mobile cell-site data, at least in
part, by checking each of the cell-tower locations against a database of sites provided
by the Department of Justice and the FBI. That database, however, is not publicly
available, so its reliability cannot readily be assessed.
Importantly, O’Daniel also discussed the limitations on his use of call-report
records. His methodology entailed comparing tower locations and connection times
from paper call records with addresses and windows of time relevant to criminal
investigations. This did not allow, he conceded, for inferences that the phone was
located within an unduly precise area at a specific time. While no error rate for this
technique has been calculated in the literature the parties and Court were aware of,
this use of static, historical records to derive locational information has been
subjected to publication and peer criticism. Hill, 818 F.3d at 298 (listing
publications). So this testimony was admissible under Rule 702 to support the
limited proposition that a cell phone was “in a general area.” Id.
12
Case 3:21-cr-00089-BJB-RSE Document 87 Filed 12/23/22 Page 13 of 13 PageID #: 600
Once his testimony resumed at trial, O’Daniel and the prosecutor again took
care to not overstate his conclusions. He testified that Jones’s phone actually
connected to the cell towers listed on the T-Mobile spreadsheet. He also spoke about
his experience as a member of the LMPD’s Digital Forensics Unit, the team charged
with requesting and analyzing cell-phone records.
Defense counsel cross-examined O’Daniel on this evidence and testimony,
effectively illustrating CSLI’s limitations for the jury. Jones, moreover, took the
stand and testified in a manner consistent with O’Daniel’s testimony: he didn’t
dispute that his phone was located nearby the robberies in question, but instead laid
responsibility for those calls (and indeed the robberies themselves) at the feet of his
stepson Khaleel.3
Conclusion
Granting in part the Defendant’s motion to exclude CSLI evidence shielded the
jury from potentially misleading and prejudicial technical testimony that no witness
was positioned to explain and defend. The limited CSLI that did come in, by contrast,
was not manipulated or manipulable by the witness. And his methodology was
consistent with O’Daniel’s experience, training, and qualification.
The CSLI information, moreover, was not a huge part of the prosecution’s case.
The location evidence was consistent with Jones’s own testimony—as well as other
overwhelming evidence of his guilt. This included a lengthy recorded confession to
police and multiple recorded jail calls that discussed the robberies and corroborated
important aspects of the Government’s proof. Once the Court sawed off the TraX
information, the CSLI evidence offered and admitted here amounted to a relatively
small piece of a much larger puzzle that the jury quickly solved.
December 23, 2022
3 Khaleel Gibbs unfortunately died during the period between the robberies and the trial,
preventing the jury from hearing his own narrative and deciding whether it was consistent
with Jones’s.
13