A pool for the pit that runs through an outage

A mine runs where connectivity is poorest and a stopped truck costs most. Haulage, drilling, and loading come from different vendors, each needing models the vehicle cannot carry, and the link out of the site is the least reliable part of the whole operation. Anything that depends on that link stops when it does, and the trucks cannot. The pool has to be on the site, on the site's own network, and it has to keep running when the link does not.

What is on the site

Autonomous haul trucks run a circuit between the loading face and the crusher or the dump, on roads that are regraded, rerouted, and extended as the pit deepens. Drills work the benches ahead of the blast. Loaders and excavators fill the trucks. Conveyors and crushers run under cameras that watch for oversize, belt damage, and spillage. Fixed cameras along the pit watch the roads, the dumps, and the edges, and light vehicles move through all of it. The haulage system came from one vendor, the drills from another, the loading equipment from a third, and each has its own perception and its own expectation of a connection out.

The site changes shape every day, which is why the models matter.

Why the cloud and a GPU on every machine both fail

The link out is intermittent, and the vehicles cannot be. A vendor cloud is only as available as the link that reaches it, and on a mine that link is a long-haul connection shared by everything on the site. When it drops, every vendor's AI system drops at the same moment, because each is a separate uplink over the same failing path. Pit imagery and production data also travel that path to places the operator does not hold.

The cloud fails on time as well as on availability. A truck that finds a haul road blocked by a fallen rock, a stopped light vehicle, or a loader where the map had clear road needs an answer inside the time the truck has, not after a round trip to a distant region.

The alternative, a GPU on every machine sized for its worst moment, is compute that rides in dust, vibration, and heat, is maintained at the site with whatever staff are there, and sits idle through most of every haul cycle.

What the pool serves

Czar Grid is several Czar nodes joined into one inference pool for the site. Autonomous haulage, drills, loaders, and conveyors are served from that pool over the site network, and the pool keeps working through any outage of the link out, because nothing it does passes through that link. Whatever the operator chooses to send to a corporate system waits for the link to return. The machines do not wait.

The pool shares perception across the pit. One set of models for trucks, loaders, and fixed cameras means the site is learned once and served to every machine that asks. When the road to the dump is rerouted, the model that learns the new road on the site's own captures is the model every truck draws on the next shift, regardless of which vendor built the truck.

The pool resolves the blocked haul road. The truck carries a small, fast model for the ordinary circuit. When it meets something that model cannot classify, it asks the pool, and a larger model on the pool resolves the situation the vehicle cannot, inside the time budget the vehicle has, because the pool is on the same network. Safety, stops, and motion limits stay on the robot as the vendor built them; the pool answers questions, it does not drive the machine.

What stays on site and who holds the record

Everything the pool reads and learns stays in the operator's rack at the site. Substrate runs on the Grid: Inference serves the models, Retrieval and Memory hold what the site knows about its own roads, benches, and equipment, Adaptation learns the pit as it changes, Evaluation checks each adapted model before it is served, Watchman verifies the integrity of what is running on every node, and Local Bridge permits only the routes the operator declares, such as the haulage system, the fleet-management software, and the site's own production systems. Egress is denied by default, and a site that chooses to be air-gapped has no route out at all.

Tasks that cross vendors, a bench inspection before a shift begins, a conveyor walk after a belt alarm, a sweep of a dump edge after rain, are authored in Agency Studio on the Grid, bounded to what they may touch, approved by a named role at the site, and recorded. A run can be replayed when a question is asked later. Pit imagery and production data stay the operator's, in a record the operator holds and reads.

How it is placed and bought

The Czar nodes go in the site's own rack, in the operations building or the plant's server room, in a conditioned and secured space the operator already controls. The site network, not the link out, carries inference to the trucks, the drills, and the cameras along the pit. The operator holds one subscription for the configured Grid, hardware, platform license, and sustainment in one line, fixed for the term, and adding trucks or a second vendor's fleet to the pool does not move it. Where a haulage vendor or integrator prefers to ship the pool under its own name with its fleet, it does so as Czar Chassis. It is the same Grid, held by the vendor.

A mine already builds its own power, its own roads, and its own water because it cannot depend on anyone else's. The pool belongs on the same list: the perception for every machine on the site, run on the site, with nothing that stops when the link does.