Before a moonbase has one of everything, it has many of a few things: excavators, haulers, prospectors. The Foreman runs a gang of identical vehicles on one duty, on a site whose stations they share with the gang beside them — and keeps every one of them working.
Surface work is repetitive work. Regolith is dug a bucket at a time, hauled a load at a time, surveyed a cell at a time, and the way a site gets more of it done is more of the same machine. That makes the hard problem a site problem, not a vehicle problem: six excavators and three dig faces, four dumpers and one plant hopper, and every one of them needing the same two chargers.
The Foreman is the software for that tier. It sits between a single vehicle, which knows how to dig, and the Marshal, which plans a whole build with one of each kind of machine. A Foreman has one kind of vehicle and one duty, repeated, and one job: decide who goes where, in what order, so that no station has two vehicles on it and no vehicle stands idle that could be working.
Aegis Station does not intend to design its own excavators or dumpers. The machines that work a site will be whoever's are on it, so the Foreman is written to take any vehicle that can state its duty: which of its states mean idle, working, done and faulted, which kinds of station its work visits, and what it reports. That statement is one file. Nothing in the Foreman is written for a particular machine.
It is the generalisation of the Fleet Commander that runs the lunar tanker fleet. The tankers taught it what a coordinator of many identical vehicles has to do. The Foreman does that for any vehicle that can say what its duty is.
Frames below are screen captures of one run, 93 minutes into a shift in simulation: six excavators and four dumpers working one pit, at four times real time.
Three things have to be said to run a fleet, and they belong to three different owners. Keeping them apart is what lets a new machine, or a rearranged site, be a change to one file and not to the software.
An idle vehicle is given the next job at once. The fleet is limited by its stations and its vehicles, never by someone noticing that one is free.
Every station has a line. A vehicle waits its turn a few metres off, is called on when the station is free, and holds it only while it is being worked on. Two loading bays serve four dumpers because a bay is free again the moment its dumper pulls away.
A charger does not belong to the excavators or the dumpers. Both fleets wait in the one line and are served in the order they arrived, with nothing above the two Foremen to arbitrate.
When a plant takes one load every so many seconds, the hopper keeps slots. A dumper is called onto it just in time for its slot, so the hopper is held for the tip and the waiting happens in the line.
When a vehicle faults, its job goes back to the front of the queue, its station is freed, and a vehicle held in reserve takes the job. A vehicle that sits too long in one state is flagged; one that is waiting for a reason the site explains is not.
For each kind of station: how long one is held, how many are in use, how many vehicles wait. Whether the site needs a fourth dig face, a second hopper or another dumper is read off the board.
The Foreman exists and runs, as of November 2026. What it has run is of two kinds, and they should not be confused.
Real flight software. The WOK water prospector is the one surface vehicle with flight software today, and the Foreman runs it. Two copies of that software were started by a separate launcher, checked in to an empty shift with the prospector's duty manifest, formed a fleet on a survey grid, and were dispatched and tracked on their own telemetry. Nothing in the Foreman is written for the WOK: its states, its commands and the layout of its telemetry are all read from its manifest.
Generic vehicles, simulated. The excavators and dumpers on this page are nobody's design. They are generic stand-ins for the earthmovers a site would bring, each described to the Foreman by the same one-file duty statement a real machine would supply: they drive across a map at a fixed speed, take the same commands and report the same states. The run pictured is six excavators and four dumpers, one of each held in reserve, with the plant taking a load every 90 seconds. In 93 minutes of site time the excavators made 262 runs from three dig faces and the dumpers 61 hauls, delivering 265 tonnes to the hopper with 106 tonnes left on the stockpile. No vehicle faulted, none was flagged as stuck, and no station step overran.
The useful result is the shape, not the tonnage. The excavators dig faster than the plant will take it, so the pile grows; the dumpers spend their waiting at the hopper, not at the loading bays; and the board says so in one row. Change the plant's rate or add a dig face in the site plan and the same fleets rearrange themselves around it.
What it does not do is stated as plainly. The pit's distances, speeds, loads and times are the simulation's own and come from no vehicle design. The stand-ins drive in straight lines and nothing keeps two of them apart on the way; only the stations are deconflicted. The site's own equipment, such as a loader at a bay, is a timed step and not a machine. Vehicles check in through a launcher that speaks for them, on the same computer, over a link with no authentication; a vehicle that announces itself over a real radio link is not built. And the tanker fleet still runs under its own Fleet Commander: expressing a tanker's sortie, with its launch windows and booked berths, as a Foreman duty is designed for and not yet done.