Expert Witness Journal Issue 67 June/July 2026 - Flipbook - Page 57
typically contain time-stamped location data and
limited sensor information but do not retain the
full-range, original sensor streams. Moreover, their
distance calculations between sequential points may
require recalculation. Over greater distances, the
data tends to be fairly accurate, but over shorter
distances – which are often most relevant in
collision reconstruction – errors may become more
pronounced and must be accounted for. In evidential
terms, these types of exported 昀椀les are often used
as secondary data: useful for mapping and general
movement analysis, but limited in precision.
trackers are now commonly owned, contributing
further to the volume of available data. However,
not all GPS-derived data is equal.
Many people subscribe to the familiar phrase, “If it’s
not on Strava, it didn’t happen.”. In practice, however,
even Strava does not always show exactly what
occurred. Di昀昀erences between platform outputs
are not uncommon. Often, the investigator’s initial
task is to understand the distinction between what is
displayed on the platform, what the exported data
indicates, and what actually happened, an important
consideration when assessing evidential reliability.
GPS measurements are inherently subject to error,
often varying by several metres even under good
conditions. To improve usability for consumers,
platforms and devices often apply 昀椀ltering and
smoothing algorithms. While these processes
improve visual clarity, they can obscure shortduration events that may be critical in a collision
scenario. In simpli昀椀ed data sets (GPX 昀椀les for
example), speed is reconstructed from positional
change and can amplify noise potentially resulting
in sporadic speed values being displayed.
As with other forms of digital data, on the surface
the data appears straightforward and simple to
interpret. In reality, uncertainties arising from
the way GPS data is captured, and how those
uncertainties are addressed, introduce signi昀椀cant
complexity. This article sets out three tiers of data
and their implications for forensic analysis.
The Activity View: Interpreted and
Processed Data
The activity page on websites such as Strava presents
a user-friendly visualisation of an activity, including
route mapping, speed, elevation, and segment
e昀昀orts. However, this view represents processed
rather than raw data. When an activity is uploaded,
Strava parses the incoming 昀椀le and independently
recalculates metrics such as distance and speed
(Strava, 2026a). As a result:
•
•
•
The most complete and forensically valuable
dataset is the original activity 昀椀le, typically in FIT
(Flexible and Interoperable Data Transfer) format.
Garmin’s documentation con昀椀rms that the format is
designed to store rich, multi-parameter activity data
at high 昀椀delity (Garmin, n.d.). Strava, for example,
recommends exporting the original activity 昀椀le
where possible (Strava, 2026b). FIT 昀椀les are designed
speci昀椀cally for performance and sensor recording
and may contain:
Displayed speed and distance may di昀昀er from
the recording device
Data may be resampled, averaged, or normalised
Graphs often re昀氀ect smoothed or reconstructed
values
•
•
•
From a collision reconstruction perspective, this
creates an important limitation: website interfaces
are an interpretation layer rather than a primary
data source. This layer is designed with consumers
in mind, and is often processed to be both visually
appealing and to allow users to easily interpret
it. While minor uncertainty or processing errors
may be acceptable for recreational used, they may
be critical in litigation. It is therefore generally
unsuitable as the sole evidential basis, particularly
where second-by-second dynamics, such as braking,
sudden deceleration or sudden acceleration, are
under scrutiny.
•
•
•
A key distinction is that FIT 昀椀les often contain several
independent measurements of movement, whereas
a GPX 昀椀le is typically limited to positional data.
Having the most accurate and reliable data allows
collision investigators to cross validate between data
sources, identify anomalies or signal loss, and obtain
more accurate speed and positional data. Whilst this
data is also subject to the general uncertainty of any
GPS device, having access to a wider range of inputs
allows for independent processing and a more
accurate assessment of uncertainty. In litigation, this
is preferable, as it provides experts with the ability
to be completely transparent in explaining how the
data was processed to achieve the conclusions cited
in reports.
Getting the Most Accurate Data
Many of these websites allow users to export
activities as GPX (Global Positioning XML) 昀椀les.
These are widely used because of their compatibility
and readability. However, GPX 昀椀les exported from
some platforms, including Strava, represent a
reduced dataset (Strava, 2026b). These exports
EXPERT WITNESS JOURNAL
High-resolution timestamped GPS data
Recorded speed (not just derived speed)
Distance streams (e.g. wheel-based
measurements)
Sensor inputs (power, cadence, heart rate)
Device metadata and recording intervals
Event markers (pauses, laps, etc.)
55
JUNE 2026
