Archaeological survey in remote terrain.

The mission

A tell in the steppe, a terraced hillside, a necropolis under ploughland — three weeks of permit, a photogrammetry drone, a LiDAR aircraft for the ridge, and a field camp of tents and generators. The campaign’s product is data: overlapping imagery in the tens of gigabytes per flight, dense point clouds, ground-control measurements. The institute at home wants to review orthophotos while the team is still on site, because the review decides where tomorrow’s flights and trenches go. And the whole enterprise runs against two clocks that never pause: the permit and the weather. A flight day lost to a data problem is not delayed — it is gone.

What breaks

There is no network over the site — that is usually why the site survived. What the classical workflow does about it is fragile in three distinct ways. Cards are shuttled by hand from aircraft to laptop to disk, a chain of manual copies where one tired evening loses a flight. The institute sees nothing until someone drives hours to town for bandwidth, so the feedback loop that should steer the season runs on a days-long delay. And if the camp has a satellite terminal at all, it is metered — a budget one flight’s raw imagery would exhaust — so it either sits idle as an emergency phone or gets burned in an afternoon of misjudged uploading.

The quieter failure is coordination on the site itself: vehicles positioning ground-control targets across a survey box kilometres wide, out of radio contact with the camp, coordinating by driving back.

The architecture on this mission

The camp is the hub. Its field server runs a node, and every aircraft bonds to it locally the moment it is in range — a short, fast hop that costs nothing and needs no infrastructure the team did not carry in. A flight’s imagery is on the camp server, checksummed and safe, while the batteries are still warm; the manual card-shuttle chain and its one-tired-evening failure mode disappear. Encryption is always on, which matters more than it sounds on a site where the data’s location is itself sensitive.

The satellite terminal joins the same node as a deliberately different kind of link: metered, budgeted, spent only on what is classed to travel it. The evening’s flagged products — the day’s orthophoto tiles at review resolution, the anomaly crops, the flight log — go south under quota for the institute’s morning review; the raw terabytes never touch the meter and travel home with the team. The feedback loop the season needs — fly, review, redirect — runs overnight instead of over a week, and the review happens at the institute, where the senior eyes are.

The vehicles make the survey box one network. Each runs a small node and joins the mesh, relaying when the geometry helps; a target-laying crew at the far edge of the box shares position and status with the camp without driving back, and the ground-control measurements they take queue and deliver themselves, store-carry-forward, as soon as any contact allows. Every link is measured continuously throughout, so the campaign accumulates a coverage picture of the site as a by-product — which matters the day a bigger operation returns to excavate.

How the field season unfolds

1. Camp-up. The team raises the camp and powers the field server; aircraft and vehicle nodes authenticate by key and join. The season’s satellite quota is set once, in one file. There is a network over the site for the first time in three thousand years.
2. A flight day. Four photogrammetry sorties between wind windows. Each lands its take on the camp server over the local bond during battery swaps — no cards handled, no copies queued for the evening, nothing to forget.
3. The far edge. The vehicle crew lays ground-control targets at the limit of the box. Their positions and check measurements flow back through the mesh as they work; the survey plan updates at the camp in real time instead of at dinner.
4. The evening window. The field director flags the day’s products; they leave under quota as high-class traffic. By morning the institute has reviewed the tiles and moved tomorrow’s flights two fields west — a decision that used to take a week.
5. The bad stretch. Three days of wind, and a sand-jammed generator takes the satellite down with it. Nothing is lost: queues hold in priority order and drain when power returns. The flying resumes exactly where the plan — not the data situation — says it should.
6. Pack-out. The raw dataset drives home with the team, complete and already organised by the season’s flow. Beside it sits the journal: every flight’s delivery timestamped, every satellite byte accounted against quota, and a measured coverage map of the site for the excavation to come.

What each mechanism contributes

• The local bond at the camp — terabytes landing at full local speed, minutes after landing, with encryption always on. Shipped.
• Quota discipline on the metered link — the satellite spent deliberately on flagged products, never accidentally on bulk. Shipped.
• Mesh across the survey box — vehicles and aircraft relaying for each other; the far crew connected without driving back. Shipped.
• Store-carry-forward — measurements and products queue through outages and deliver in priority order; the field’s ordinary interruptions become delays, not losses. Shipped.
• Traffic classes — the flagged product and the camp’s operational traffic ahead of everything bulk on the one thin link. Shipped, opt-in.

The honest boundary: Atlas cannot widen the satellite, fly in the wind, or extend the permit — and the raw dataset still travels home as cargo, because that is the economics of metered links, not a software gap. The aircraft, the sensors, the photogrammetry pipeline and the archaeology are the institute’s.

What a pilot should prove

• Zero-loss delivery: every sortie’s imagery on the camp server the same day, checksummed, with no manual card handling — timestamped by the journal.
• The overnight loop: flagged products at the institute by morning, under quota, with the spend accounted to the byte.
• Field coordination: the target crew’s measurements delivered from the far edge of the box without a return drive.
• Outage behaviour: a real multi-day interruption with queues intact and delivery order preserved on recovery.

One site, one camp, one season. The evaluation format covers the structure.