Open-pit mine operations below the radio horizon.

The mission

A copper pit, three hundred metres deep and growing. Every morning a survey drone flies the benches and stockpiles for volumetrics; blast crews need monitoring before and after every shot; a pilot programme is putting telemetry on the haul trucks; and the operations centre that consumes all of it sits in an office block outside the rim. The pit is one of the most instrumented industrial sites imaginable — and its own geometry is the reason half the instruments cannot phone home.

What breaks

Radio does not bend over a rock wall. A drone or a truck below the rim is invisible to every ground station and every commercial cell outside the pit — not weakly connected, but geometrically cut off, and no transmitter power changes that, because the problem is the horizon, not the link budget. Networks built inside the pit fare little better over time: the pit is a shape that changes. Every blast moves the terrain the coverage plan assumed; a repeater that covered a bench in March shadows it by June; and each repeater is a single point whose failure takes its whole sector down. The coverage engineering never stops, because the mine never stops digging the map out from under it.

The architecture on this mission

The pattern puts relays where the geometry is won or lost: on the rim. A handful of nodes — mast-mounted at fixed rim sites, hosted on vehicles that park on the crest, or aboard an aircraft orbiting above the pit during intensive operations — see both worlds at once: down every face of the pit and out to the operations centre. Everything below the horizon runs the same node and reaches the nearest relay over one short hop up; the mesh does the rest, relaying up to eight hops with every leg separately encrypted, and relays cannot read the traffic they carry.

Redundancy is the point of using several. With two or three rim relays up, most of the pit sees more than one of them, so a relay lost — powered down for maintenance, shadowed by a new bench, its host vehicle driven off — is routed around in under a second, from the continuous probing every node already does. The blast cycle gets the same honesty: when the pit is cleared for a shot and nodes are withdrawn or powered down, survey and telemetry data queues store-carry-forward wherever it is, and drains automatically after the all-clear, priority first. At the operations-centre edge, the rim nodes bond whatever the site offers outward — fibre at the office, a carrier link as backup — into the same measured fabric.

How the shift unfolds

1. Shift start. The rim relays are already up — they are infrastructure, not per-sortie kit. The survey drone launches for the morning volumetrics run; its node joins the mesh and the link state floods to every member.
2. Below the horizon. The drone works the lower benches, invisible to the outside world, streaming imagery one short hop up to the nearest rim relay and onward to the operations centre. The haul-truck telemetry rides the same fabric.
3. Equipment moves. The pickup hosting a temporary crest relay drives off to the workshop. Routes recompute around the loss in under a second; the drone’s stream continues through the remaining rim sites and nobody in the operations room notices.
4. The blast window. The pit is cleared for the midday shot. Withdrawn nodes queue what they hold; the shot fires; the geometry that every coverage plan depended on changes by exactly one bench.
5. After the all-clear. The post-blast survey flies the new face immediately — the mesh does not care that the map changed, only which relays each bench can still see. Queued data from before the window drains behind the live traffic.
6. The debrief artefact. The journal shows, per relay and per bench, what actually connected where — the evidence for which rim site to move as the pit deepens, produced free by the day’s traffic.

What each mechanism contributes

• Mesh relaying — up to 8 hops, per-leg encryption, relays blind to the traffic they carry; the rim geometry becomes routing, not engineering debt. Shipped.
• Sub-second rerouting — links probed several times a second; a lost relay abandoned and routed around in under a second. Shipped.
• Store-carry-forward — survey and telemetry data queues through blast windows and outages, draining priority-first on reconnection. Shipped, opt-in.
• Bonding at the edge — the rim-to-office leg aggregates fibre and carrier links into one measured, encrypted fabric. Shipped.
• External parts — the rim sites themselves: masts, power, lightning protection and the emplacement decisions are the mine’s. That is what makes this a pattern rather than a boxed capability.

The honest boundary: Atlas contributes the mesh, the rerouting and the queue — the rim sites are the mine’s. Powering a mast on a crest that moves every year, hardening it against blasting and lightning, and choosing where the next one goes are emplacement engineering Atlas informs (the journal shows exactly where coverage is thinning) but does not do. And no software sees through rock: a bench shadowed from every relay is offline until a relay moves or the geometry does — the system’s contribution there is queueing honestly instead of pretending.

What a pilot should prove

• Coverage of defined working benches through two rim relays, against the direct-path baseline, from the built-in measurement.
• Continuity while a relay is removed mid-survey: routes re-form without operator action, quantified from the journal.
• Survey-data completeness across a real blast window — everything captured before the clearance delivered after the all-clear.
• A relay-placement recommendation for the next push-back, generated from one month of per-bench journal data.

One pit, two rim sites, one survey cycle. The evaluation format covers the structure.