Method
How both metrics are calculated
The definitions, counting rules and judgment calls behind both metrics, set out for independent review.
Why these two metrics
In 1964 Nikolai Kardashev proposed rating civilizations by the energy they command. The idea has organized thinking for sixty years because it is one number with a clear meaning. But Kardashev measures energy at home, and says nothing about the ability to operate anywhere else. A civilization could climb his scale indefinitely without ever putting a working machine on another world.
The gap that matters for expansion is not energy. It is the ability to act at a distance where instruction is impossible. In April 2026 a group led by Masahiro Ono at NASA's Jet Propulsion Laboratory published Planetary Exploration 3.0, arguing that spacecraft behaviours today are largely pre-designed and fixed, and that this approach cannot work in the outer solar system, where a single voyage takes a decade and there is no chance to iterate. What is needed instead are systems that adapt in situ, deciding as they learn.
The Expansion Stage measures how much of our hardware has crossed that line. The price of a world measures the other half of the problem: not how capable the hardware is, but how efficiently we can put any hardware somewhere new. Ono's group frames this as NASA's Mars campaign of 22 progressively sophisticated missions across decades. The method worked at Mars. It cannot work at Enceladus.
The scale
Each stage is a thousand times the one below it. These definitions are permanent. The methods for measuring them are expected to improve; the rungs will not move.
| Stage | Autonomous power | What it means |
|---|---|---|
| E0 | 1 watt | First self-directed hardware beyond Earth orbit |
| E1 | 1 kilowatt | We can explore without waiting on Earth |
| E2 | 1 megawatt | We can operate a world on its own judgment |
| E3 | 1 gigawatt | We can do it again anywhere |
| E4 | 1 terawatt | We spread beyond this solar system |
Stages are three orders of magnitude apart, so dividing the exponent by three places you on the ladder. Nothing else in the equation is chosen.
The counting rules
What qualifies as autonomous power
Count a system's electrical power if it conducts its primary operations at a body without ground in the loop. The unit is the whole system's electrical output, not an estimate of the fraction of that power devoted to computation — dividing a spacecraft's watts between “thinking” and “everything else” would require assumptions no public source supports.
Two exclusions
Fault protection alone does not count. Every deep-space probe can detect an anomaly, enter a safe mode and wait. Reacting to a problem is not conducting operations, and if it counted, the Voyagers would qualify and the metric would measure nothing.
Autonomy during cruise does not count. Navigating between worlds is not exploring one. A probe that manages its own trajectory for nine years and then executes a ground-written sequence on arrival has not crossed the line this metric draws.
What counts toward the price of a world
Missions delivered before persistent operation began at that body. Once a world is established, its price is frozen. Supply runs, crew rotations and later traffic cost nothing on this metric, because returning to a world you already hold is a sign of success, not a cost.
A body leaves the count when we stop operating hardware there. This is why the price rose after 2017: Cassini ended, Saturn left the portfolio, and a cheaply-established world stopped pulling the average down.
Methodological choices
Excluding Earth orbit
Satellites represent roughly 75 megawatts, far more than everything beyond Earth orbit combined. Including them would make the Expansion Stage a satellite industry tracker and would swamp any signal from deep space. The judgment is that orbital infrastructure is not expansion. Someone could reasonably argue that a self-managing constellation is exactly the capability being measured.
A binary autonomy test rather than a graded one
A system either conducts operations without ground in the loop or it does not. This is crude for a capability that plainly comes in degrees, and it produces cliff edges: a rover that plans 89% of its traverse counts identically to one that plans 100%. A graded test would be better and does not yet exist in publishable form.
Freezing establishment costs
Fixing a world's price at the moment persistent operation begins means the metric cannot reward getting cheaper at a world we already hold. That is deliberate — the question is the cost of the next world — but it does mean a dramatic fall in launch cost would show up only when somewhere new is established.
Counting whole-system power
An alternative would be to count only the power available to the autonomous subsystem. That figure is not published for any mission, so the choice is between a coarse number that is verifiable and a fine number that is invented.
What would change the method
The binary autonomy test is the weakest part of this method and it is already scheduled for replacement. Ono's group proposes a System Adaptivity Level scale, analogous to Technology Readiness Level, and explicitly leaves its definition as future work.
When that scale is defined and published, this metric should adopt it. At that point the entire series will be recomputed so that comparisons across editions stay honest, the edition number will increment, and the change will be recorded in the corrections log alongside the old and new figures for every year.
Stage definitions never change. Methods may improve. When they do, the historical series is recomputed rather than spliced.