A temporary network where none survives.
After a major disaster the question is not whether the network is down — it is that there is no network left to be down: the field hospital, the shelters and the headquarters each improvise an island. This mission is about linking those islands into one working network on the vehicles and satellite terminals the response already brings — a pattern in which Atlas carries the traffic between the sites and your access equipment serves the people at each of them.
Civil protection agencies, telecom emergency units and humanitarian organisations restoring coordination after a disaster.
An area can be without any access network for weeks; the field hospital, the shelters and the headquarters each improvise their own island.
A bonded backhaul per site — satellite plus whatever cellular survived — and a mesh that links the islands into one network, coordination first.
Vehicle-carried computers, drone companion boards and the headquarters’ workstation. One Linux binary per node, one configuration file.
The mission
An earthquake, a catastrophic flood, a storm that took the grid and the towers with it. Within days the response has shape on the ground — a field hospital at the edge of town, shelters in the school and the sports hall, a headquarters at the municipal depot — but no shape on any network, because there is none. Patient transfers are coordinated by sending a vehicle. Supply requests travel as photographed paper. Each site has scraped together something local — a generator, a Wi-Fi router, a satellite phone — and none of it connects to anything else. The mission is to put one working network over the response area, fast, out of what the convoy brought.
What breaks
What fails here is not a link but the assumption behind every link: that there is infrastructure to connect to. The cell sites are dark, damaged or on batteries that died on day two; the fibre is cut where the ground moved; whatever coverage flickers back is instantly saturated by everyone left in the area. Restoration is measured in weeks, and the response cannot wait for it.
The classical answer is a truck-mounted emergency cell site — scarce, slow to arrive, and covering one site, not the relationships between sites. Satellite terminals at every site solve each island’s uplink and nothing else: hospital-to-headquarters traffic crosses two satellite hops and a foreign data centre, is metered per megabyte, and dies with the weather. The thing the response actually needs — the sites talking to each other, locally, with the expensive uplinks shared and spent deliberately — is precisely what a pile of independent terminals does not provide.
The architecture on this mission
Each site gets a node — a vehicle-carried computer, or the board already in a drone — and the nodes form one encrypted mesh with no central server and no configuration ceremony: they authenticate by key and join. Site-to-site traffic travels locally over line-of-sight hops, relayed up to eight hops deep, so the hospital reaches the headquarters directly rather than via space. Where two sites cannot see each other across rubble or distance, a drone node bridges the gap; a vehicle parked on high ground does the same job for the price of parking it there.
Each site’s node also bonds whatever reaches outward — the satellite terminal, the one cell sector that came back on a generator — into the same fabric, measured several times a second like every other link. The mesh shares those uplinks: the headquarters’ terminal serves every site, the metered links are spent deliberately rather than accidentally, and when a surviving cellular path dies again, traffic is reweighted onto what remains in under a second. Six priority classes keep the order honest under scarcity — coordination and the hospital’s traffic first, bulk transfers queued for the quiet hours — and every hop is encrypted, with relays unable to read the traffic they carry.
What Atlas is not, on this mission, is the network in anyone’s hand. People at each site connect to ordinary access equipment — Wi-Fi access points, the responders’ own radios, the hospital’s telephony — and that equipment hangs off the site’s node as its route to everywhere else. That division of labour is why this page carries a Pattern badge rather than a Capability badge: Atlas’s share is shipped product, and the bubble as a whole is Atlas plus equipment and procedure that are necessarily yours.
How the deployment unfolds
- Day one. The response arrives and improvises. Three sites, three islands, and a vehicle doing the work a network should — carrying patient lists and supply requests as photographed paper.
- Nodes out. A vehicle node is dropped at each site and keyed in; the headquarters raises its satellite terminal. Within the hour the three islands appear on one dashboard, each with its links’ measured state.
- The bubble closes. Line-of-sight hops link headquarters to hospital; the far shelter, blocked by collapsed buildings, is bridged by a drone node — later swapped for a vehicle parked on the overpass. Hospital-to-headquarters traffic now moves locally, in one encrypted mesh, not via space.
- Scarcity, ordered. In the evening the one surviving cell sector dies under load — again. Traffic reweights onto the satellite path in under a second, and the class order takes over: coordination and medical traffic first, the day’s photo documentation queued behind them for the night.
- The bubble grows. A fourth site opens in the next village. Its node authenticates by key and joins; routes recompute; nobody reconfigures anything at the other three sites.
- Handover. The carriers restore a sector, then another; each restored path joins the bond and gradually takes the load. The journal — a per-site, per-link record of what the bubble carried for three weeks — goes into the after-action review and sizes the kit list for the next deployment.
What each mechanism contributes
- Self-forming mesh — sites, vehicles and drones as one encrypted network with no central server; relaying up to 8 hops; relays cannot read the traffic they carry. Shipped.
- Bonded backhaul per site — satellite plus whatever cellular survived as one measured fabric; sub-second reweighting when a path dies; the metered link spent deliberately. Shipped.
- Six priority classes — coordination and medical traffic first under scarcity; bulk queued, never competing. Shipped, opt-in.
- Continuous measurement and the journal — the record of what the bubble carried, for the after-action review and the next deployment’s planning. Shipped.
- The external parts — the Wi-Fi access points, responder radios and telephony at each site, and the spectrum coordination behind the inter-site links, are essential to this mission and are not Atlas.
The honest boundary: Atlas is the transport between the sites, not the network in anyone’s hand. The access layer people actually touch — the shelter’s Wi-Fi, the responders’ radios, the hospital’s phones — is separate equipment that hangs off the mesh and must be brought, powered and configured by the operator. Making the inter-site radio links lawful and de-conflicted in a disaster area is spectrum coordination with the national authority — procedure, not software. And a metered satellite uplink stays metered: Atlas rations and shares it honestly, it does not make it larger.
What a pilot should prove
- Time to network: three improvised sites joined into one routed, encrypted network on the first exercise morning, from vehicle arrival to hospital-to-headquarters traffic flowing.
- Coordination continuity through a deliberate kill of a site’s cellular path, with the sub-second reweighting onto satellite visible in the journal.
- Class behaviour under a congested uplink: coordination and medical traffic delivered first, bulk measurably queued behind them, from the built-in record.
- A handover artefact: the per-site, per-link journal of what the bubble carried, complete enough to brief the after-action review from.
Three sites, one exercise weekend, one journal. The evaluation format covers the structure.