Elara-Cortex Technical Note · 20 July 2026

Dead-zone routing: navigation that survives the loss of a mobile signal.

K. Lekola. Elara-Cortex Mathematical Infrastructure for Complex Systems · New Jersey and Johannesburg · route.elara-cortex.com

Abstract. Most navigation depends on a network round-trip: the device sends its position to a routing server and receives the next instruction. When the mobile signal drops, in an urban canyon, a tunnel, a rural area, or during a positioning outage, that round-trip fails and turn-by-turn guidance stops. This note sets out the dead-zone problem, its measured operational cost from published sources, and the approach Elara takes to it: plan the route once, then keep re-optimising it on the device, from the satellite position alone, with no live connectivity. The routable index is small enough to sit on the device and the recomputation is local and sub-millisecond, so the route holds through the dead zone and resumes cleanly when signal returns. The method is proprietary; the capability is demonstrable live, and the supporting latency and footprint measurements are reported in the companion research reports.

1 · The problem

A dead zone is any place where a device cannot reach the network well enough to complete a routing request: dense building clusters that scatter the signal, tunnels and underpasses, mountainous or rural terrain with thin coverage, and short-lived positioning outages caused by interference. In each case the satellite receiver may still know where the device is, but the route cannot be recomputed because the server cannot be reached. Conventional turn-by-turn navigation treats connectivity as always present, so it has no answer when connectivity is the thing that fails.

The cost of this is not marginal. In 2023 a positioning-interference event around Nanjing was reported to have reduced food deliveries by roughly forty per cent and ride-hailing by roughly sixty per cent over about six hours, because routing across the affected area stopped working. At national scale, a study prepared by RTI International for the United States National Institute of Standards and Technology estimated the economic cost of a positioning-system outage at on the order of one billion dollars per day. At the level of a single delivery, industry benchmarks put the cost of a failed drop, the re-attempt, the driver time, and the customer churn it invites, at roughly seventeen to twenty dollars, and in some emerging markets the underlying failure rate is reported above thirty per cent.

Table 1. The operational cost of dead zones, from published sources. Each figure is external and attributed to its source in the right-hand column; none is asserted.
Observation	Figure	Source
Nanjing positioning-interference event, 2023	food delivery −40%, ride-hailing −60% over ~6 h	South China Morning Post
Economic cost of a positioning-system outage	~$1 billion / day	RTI International for NIST, 2019
Cost of a single failed delivery	~$17–20	SmartRoutes / Statista; corroborated by GoBolt and Locus.sh
Reported delivery-failure rate, some emerging markets	30%+	iCargos last-mile operational data
Smartphone radio power draw rises in weak signal, relevant because a device that repeatedly retries a failed round-trip in a dead zone burns battery for no delivered route	modem works harder as signal weakens	Carroll & Heiser, USENIX ATC 2010

2 · Why conventional navigation breaks

The structural reason is the network round-trip. When routing lives on a server, every re-plan is a request and a reply across the mobile network, and the quality of navigation is bounded by the quality of the connection. A route computed once at the start of a trip degrades as conditions change, and the moment the signal drops, it cannot be refreshed at all. The companion report TR-2026-04 measures this directly: the same route that is computed locally in well under a millisecond takes on the order of hundreds of milliseconds when it is fetched across a network round-trip, and in a dead zone the round-trip does not complete at any latency. Navigation that depends on the network inherits the network's dead zones.

3 · The approach: plan once, re-route on the device

Elara moves the routable knowledge onto the device. The route is planned once, and it is then re-optimised locally, from the satellite position alone, whenever the road situation changes, with no live connectivity required. Two properties make this practical. First, the index of routable places is compact: a gazetteer of 180,474 towns is carried in a few megabytes, and its integrity is verified byte-for-byte by a SHA-256 digest, so the device holds a small, exact copy rather than streaming a large map. Second, the recomputation is local and fast enough to run on every change: the local route computation reported in TR-2026-04 completes in about two-thirds of a millisecond on a consumer machine, so re-planning on the device is imperceptible. When the signal returns, the device reconciles with the network cleanly; nothing was lost while it was offline.

The internal mathematics of the engine is a trade secret of Elara-Cortex and is not disclosed here. The claims in this note are constructed to be checkable without it: the footprint and the byte-exact reconstruction are verifiable from the published digest, the latency figures are reported with their protocol in the companion reports, and the routing itself can be exercised live on this site.

4 · Where it applies, and where it does not

The benefit is largest exactly where connectivity is least reliable: emerging-market last-mile delivery, rural and remote routes, dense urban canyons, tunnels and underground sections, and any event that degrades positioning for a period. It is smaller where coverage is continuous and a network route is always available, though even there the local path removes the per-request latency and the dependence on a metered external service. The index is town-level rather than full street-level in every location, and the approach addresses the routing layer; it does not improve the underlying satellite fix, and it is not a substitute for coverage where none exists. These boundaries are stated so the claim is not read more widely than it is meant.

5 · A note on the wider trend

The pressure that makes an on-device, compact approach valuable is not going away. The volume of data the world stores is growing at roughly twenty-three per cent a year, which doubles it about every three and a half years (IDC Global DataSphere), and the reflex response has been to build more storage and more compute to hold larger maps and models. Carrying the routable knowledge in a small, exact footprint is the opposite move: it keeps the working set on the device rather than in a data centre, and its value grows as the raw data grows rather than shrinking. The same compression discipline is documented, with its exact measurement basis, in the compression sections of TR-2026-02.

6 · Live demonstration

The routing engine described here runs live on this site. A route can be computed now at route.elara-cortex.com/elara-route/app, and the measured latency, footprint and solution-quality claims that stand behind this note are reported in full, with their protocols and receipts, in the research series linked below.

References

RTI International (2019). Economic Benefits of the Global Positioning System (GPS). Report prepared for the U.S. National Institute of Standards and Technology (NIST), RTI Report No. 0215471.
South China Morning Post (2023). Reporting on the Nanjing satellite-positioning interference incident and its effect on ride-hailing and food delivery.
Carroll, A., & Heiser, G. (2010). An analysis of power consumption in a smartphone. In Proc. USENIX Annual Technical Conference (ATC).
International Data Corporation (2023). Global DataSphere Forecast.
Elara-Cortex (2026). TR-2026-04: Local computation versus a network round-trip in navigation. route.elara-cortex.com/elara-route/nav-moat.
Elara-Cortex (2026). TR-2026-02: Measured performance of the production engine. route.elara-cortex.com/elara-route/benchmarks.

Research series: TR-2026-01 · TR-2026-02 · TR-2026-04 · EL-2026-01
© 2026 Elara-Cortex Mathematical Infrastructure for Complex Systems · New Jersey · Johannesburg · Every external figure in this note is cited to its published source; the routing method is proprietary and the capability is demonstrable live.