A route is a fact with a half-life
Datum is route planning for contested ground. It knows how old every road on your map is, plans on what you actually know — and tells you which unknown is worth checking before you move.
Confidence
Neuhardenberg airfield, Brandenburg.
Real extract · confidence states illustrative.
Measured on the demo ground
The problem
Someone spends hours establishing that a corridor works. From the moment they drive it, that knowledge starts rotting. A crossing is destroyed. A ford floods. A farm track turns to mud after one night of rain.
The map on the screen does not change. It draws the road exactly as confidently as it drew it a week ago, because every planning tool in service treats a road as permanent infrastructure.
What that costs does not show up at a desk. A crew spends three to four hours planning a single run and another two having it reviewed, and what they carry out of that room is a photograph of a road that has already moved on. When the photograph is wrong, nobody finds out in the planning cell. They find out at the crossing, in the dark, with a loaded vehicle and no second way round.
Planning and review hours: one UGV group commander’s own account. n=1 — the weakest number on this page, and we would rather say so than round it up.
The route and the situation can change daily. Today a guided-bomb strike. Tomorrow a burned-out vehicle stuck on a narrow causeway. The day after — new mining.
In contested ground a road is not infrastructure. It’s a claim with a date on it.
See it
Nothing on this map gets destroyed. No bridge drops, no ford floods. The only thing that happens is that time passes and nobody goes out to look — and every other planning tool in service draws all of it in the same confident line at hour 96 as it did at hour zero.
62 of 102
roads you’d still bet a vehicle on
26
never checked by anyone, ever
The planner
Takes the direct corridor.
Most of the ground was seen this morning. The short way is a defensible bet.
Drag it.
Real roads, from the committed extract east of Neuhardenberg airfield. The ages on them are assigned for this demonstration — in the product they come from your own reports, and each road class decays at its own rate.
The gap
Military logistics has been digitised, and seriously. All of it answers what, how much, and by when. None of it plans a route — or says how sure it is about one.
Already solved
Fuel requisitions that took weeks now clear in hours, across hundreds of units.
Digital marketplaces move platforms to depots by the million.
NATO logistics systems are fielded and licensed down to brigade level.
Still open
Whether the road that got the pallet there yesterday still exists today.
Which of three ways in is worth the risk, given what was last seen on each.
Where to point the next drone so the answer actually changes the plan.
A ground-robot platoon flies a drone over the entire route before every single sortie — not as a formality, but as a condition of the vehicle coming back. They are already paying for reconnaissance.
We tell them where it’s worth spending.
Why now
Routing over a map you are only partly sure about has been studied since 1991, as the Canadian Traveller Problem. Three decades of papers. Nothing you can buy.
A belief about a road is worthless unless something keeps checking it. A drone over the route before every sortie, reports coming back off the vehicles, satellite revisit measured in hours — that stream is a handful of years old, not a decade. Before it there was nothing to decay.
A consumer navigation vendor ages the confidence of its map sources. The US Coast Guard has planned searches on a belief map, and ranked where to look next, since 2007. Deciding what to collect first is written doctrine. Nobody has put the three on one object — the road a crew is about to drive.
That is an argument, not a measurement — and it comes with a condition we won’t bury: if your people are not already looking at the ground, there is nothing for Datum to keep a belief about, and it will not help you.
The product
They have to happen in this order. You can’t tell someone what’s worth checking until you know what a route decision would do with the answer.
First
Every road carries the date someone last saw it, and a confidence that falls as that date recedes. A report from this morning, a report from last week and pure assumption stop looking alike on the screen.
You moved that clock yourself, further up this page. In the instrument it runs on your unit’s own reports — and it is the product working, not an animation.
Then
Routing runs on that, not on a clean map that pretends everything is still true. You get the fast way, the safe way and the balanced one — each with the number of unknowns it is resting on, in the open.
Same information in, same route out, every time. Nothing about the plan is a guess you can’t inspect.
Three ways in
Same information, same route, every time00:29
ETA, 20 Aug UTC
01:05
ETA, 20 Aug UTC
01:19
ETA, 20 Aug UTC
The last row is the one no other planner prints: how many roads on this route nobody has actually checked.
Only then
This is the output nothing else produces. Of everything you don’t know, Datum computes the short list whose answers would actually change the route — and ranks it by the time each answer is worth.
It answers the question a crew already asks before every sortie: we’re flying the drone anyway — where should it look?
Worth going to look at
Planner outputThe five questionsDriving time the answer saves
Ranked by the time each answer is expected to save. Everything else you don’t know stays off this list, because knowing it would not change where you drive.
Evidence
The demo runs on Neuhardenberg airfield and the Oderbruch around it — drained marsh, canal grid, farm tracks. Every figure here comes out of the committed dataset and can be reproduced on the stand.
1
of 64 bridges
Road bridges within 10 km of the demo site. Exactly one publishes what it can carry. The map does not know what your vehicle can cross.
2,386
of 4,300 roads
Carry something nobody has checked. Most planners draw all 4,300 with identical confidence anyway.
13
of 2,429 questions
Open questions whose answer would change where this vehicle drives. Five are worth asking first.
0
bytes uploaded
Runs on one machine, on an airfield with no signal. No cloud, no callback, nothing to leak.
Boundaries
Stated up front, because you would find out in the first meeting anyway and we’d rather spend that meeting on something else.
Who’s building it
Facundo Maidana. Machine-learning engineer with an operations-research background from the University of Bologna. Datum is value of information over a graph — the seam between exactly those two fields, which is why it looks obvious from here and unusual from almost anywhere else.
Solo and self-funded. No customer, no pilot, no contract — what exists is the software on this page, running on the airfield outside. The first two people I want are someone who has planned real ground movement under fire, and a second pair of hands on the solver.
The product is shaped by operator interviews, and where a number on this site comes from one of them, the site says how many people it came from.
Contact
Neuhardenberg, 20–23 August. The demo runs on the airfield it was built on — offline, on one laptop, on the roads outside the hangar. If you plan ground movement, or fund the people who do, we want thirty minutes and the hardest question you have.