Mount Vernon Trail · Arlington & Alexandria, VA
The backlog became a standing population
Three weeks after my first survey, I rode the same corridor with the same method. Fewer new devices appeared, but the ones already there barely moved. A quarter stayed a week or more, and six devices spanned the full 10-day date window.
- reconstructed devices, at minimum
- 29
- stayed a week or more
- 24%
- spanned the 10-day date window
- 6
- device-days of obstruction
- 83
- new devices per pass
- 1.3
The result
Fewer arrived. The ones there did not leave.
Across 16 passes on eight riding days, 150 GPS-tagged sightings resolve to at least 29 devices. New arrivals fell from 3.3 per pass in period 1 to 1.3, yet between six and 13 devices stood in the corridor on every pass.
The mean observed stay nearly doubled from 1.5 to 2.9 days. Seven devices, 24% of the reconstructed total, stayed a week or more. Six spanned all 10 calendar days, and four were recorded on both the first and last pass. Despite 41% fewer devices than period 1, this window produced more cumulative obstruction: 83 device-days instead of 73.
On 50 occasions I saw a device in the morning and again that evening, still in place as much as 7.9 hours later.
Live map
Where the long stays concentrated
Circle area tracks recorded sightings. Color shows the minimum observed stay: red 6+ days, orange 3–5 days, ochre 1–2 days, gray under a day. Basemap: OpenFreeMap, © OpenMapTiles, data © OpenStreetMap contributors.
Circle area follows sightings and color follows minimum observed dwell. The public map uses clustered locations, never the timestamped commute log.
How I counted
From 150 sightings to a conservative device floor
I deliberately re-recorded each obstruction on every pass. Sightings within 40 metres are grouped into a location, and an unbroken run across the ordered 16 passes is treated as one occupancy episode.
A one-pass gap inside a run is treated as a missed observation. I observed almost no evidence of retrievals, and the pattern supports that account: 14 one-pass internal gaps versus five gaps of two or more. Longer gaps are still treated as departures, which, if anything, overcounts turnover. Splitting every gap produces a strict upper bound of 43 devices.
The morning pass on August 28 was incomplete. Sightings from it remain, but its absences never count as evidence that a device left.
What the counts do and do not represent
Twenty-nine is a floor. On 11 passes, one location produced multiple simultaneous sightings, implying at least 12 additional devices that location-based reconstruction cannot separate. Devices that arrived and left between passes are invisible.
Dwell times are minimums. Brand is my visual observation, not a field in the data, so the map and location table make no operator claim.
What the ambiguity means
Either reading requires action.
If each occupancy episode is one device, multiple units sat untouched for up to 10 days. Collection and idle-device monitoring failed.
If successive devices cycled through the same positions, turnover and bad parking were higher than measured. Geofencing and end-of-ride controls failed.
Requested action
Treat the coordinates as a work order.
- Clear the 12 recurring locations now and explain how seven devices went unflagged for a week or more.
- Account for collection activity across both measured windows using operator telemetry.
- Geofence the trail and bridge crossing, require end-of-ride parking verification, and add the corridor to a daily sweep.
- Make removal standards enforceable and measure dwell time rather than complaint volume across jurisdictions.
The series continues
The effect of intervention can be measured.
The complete timestamped datasets for both periods are available on request to an operator or agency that wants to compare them with internal parking and collection records. They are not posted publicly because they reveal a repeat commute pattern.
Contact Jordan about the studyData notes & limitations
Read these alongside the findings.
- Sampling: I made two passes per riding day on a fixed route. Devices outside my line of sight, or placed and removed between passes, are absent. This biases every count downward.
- Device count: 29 is reconstructed from 150 sightings, not directly counted. Splitting every gap gives 43. Both miss devices parked together and anything that came and went between passes.
- Incomplete pass: the morning of August 28 was under-logged. Its sightings remain, but its absences are not evidence and never split an occupancy episode.
- Positional accuracy: phone GPS fixes were taken while moving, usually within a few metres but less precise near the bridge. The 40-metre clustering radius may merge nearby parking spots.
- Episode identity: sightings at one coordinate cannot distinguish one stationary device from successive devices left in the same place.
- Cross-period matching: locations are matched within 60 metres. A match means the place recurred, never that the same device did; the 10-day gap between periods was not observed.
- Dwell times are lower bounds, measured from first to last sighting. A device may have arrived earlier or remained later.
- Operator attribution: brand is my contemporaneous visual observation and is not encoded in the data. No per-location table attributes a device to an operator.
- Period: August 24 to September 3, 2026, covering 11 calendar days and eight riding days; there were no rides August 29–31.
- Extent: sightings span 38.84107° to 38.87680° N and 77.03802° to 77.04994° W.