Mars Mission Architectures

The cube

Four Decisions Generate Every Mars Architecture Ever Proposed

Name the four answers and you have named the architecture. One of the four dominates the other three.

The second decision dominates because it is the only one that recurses. A vehicle that lands must carry the propellant to leave again, and that propellant must itself be decelerated, landed, and lifted — so the penalty appears in every previous stage of the journey. The other three add or subtract; this one multiplies.

The field’s most consequential unexamined premise is not any of the four answers. It is the assumption that there is one architecture to answer them for.

The field

The Published Architectures, Placed on the Cube

Where each proposal sits, and what it is actually optimizing.

Worth noting what the Chinese row means. Tianwen-3 is not a crewed architecture, but it is the only funded program that will demonstrate Mars ascent and Mars-orbit rendezvous — the two operations on which every orbit-staged crewed plan depends. The capability arrives on a robotic mission, and it arrives first.

The split

Cargo and Crew Resolve the Dominant Decision in Opposite Directions

Land the cargo vehicle and keep it. Don’t land the crew vehicle. Both camps are half right.

Cargo and crew differ in every parameter the cube is sensitive to. Cargo has no return leg, so there is no mass ratio to pay and every objection to landing a large vehicle evaporates. More than that, the dry mass inverts from liability to asset: a landed hull is 2,800 m³ of qualified pressure volume, about 66 t of stainless steel, and a skin that cuts into a 42 m landing pad. The vehicle stops being transport and becomes infrastructure the moment it stops moving.

Crew is the opposite case. A crew vehicle that lands must lift Earth-return capability back out of the gravity well, which is the 683 t problem — and it must be human-rated on every flight of its production line, which is the Shuttle’s most expensive lesson. Leave it in orbit, and neither cost applies.

SpaceX says land it to both questions. Zubrin says don’t land it to both. Each is right about one of them — and Starboat-as-cargo-ferry is where Zubrin over-applies his own insight, because the argument against landing the big ship is an argument about return propellant, and cargo has no return.

Two classes

Two Propulsion Classes and Two Reliability Standards, Permanently

The split is not a phase the program grows out of. It is structural.

Crew transit time is a radiation dose budget, which demands high thrust, which means chemical propulsion for the foreseeable future. Cargo transit time is a scheduling inconvenience, which permits high specific impulse and low thrust — solar-electric or nuclear-electric propulsion cutting propellant by a factor of four to eight.

A second-order consequence changes the shape of the whole program. Low-thrust departures are only weakly window-locked, so cargo flow becomes quasi-continuous while crew stays hard-quantized to the 25.6-month synodic period. There are two flows, not one, and they couple at only four places: the node, the propellant standard, the prepared site, and the interface set.

Reliability splits the same way and for a reason people routinely get wrong. Certification attaches to a vehicle and its production line, not to a mission — so one vehicle serving both roles must be human-rated on every flight it ever makes. Separating the fleets confines human rating to three items: a ferry, a capsule, and a parked habitat. The cargo line then gets to be cheap, fast, iterative and lossy, which is exactly what a 50% landing attrition assumption requires it to be.

The clean sheet

Asymmetric Flow: What Falls Out of the Physics With No Legacy

Derived from six constraints, assuming permanent presence rather than a mission with an end date.

The field’s most consequential smuggled assumption is an objective: most published architectures optimize a round trip with a beginning and an end, and never say so. Assume instead that return is a standing property of the system rather than an event, and a different architecture falls out.

What the resulting architecture does not need is the more useful list: no megawatt-class Mars reactor, no nuclear propulsion for crew, no thermal protection manufacture on Mars, no orbital ISRU for the first crew, no cislunar staging or lunar dependency of any kind, no 100-tonne-class Mars ascent, and no crewed vehicle that must both enter an atmosphere and depart a planet.

It converges with Starboat on the ferry, with the orbit-first camp on where the crew lives, and with SpaceX on how cargo arrives. Two independent derivations reaching the same vehicle is the strongest available evidence that the vehicle is right.

The fork

Ferry or Land the Ship: The Crew-Delivery Numbers

The ferry wins at every crew size and by more as the crew grows — and the decisive advantage is granularity, not efficiency.

The ratio reaches 4.1 to 1 at 50 seats, and it widens rather than narrows with scale, because a ferry never carries Earth-return capability up out of the gravity well. The common objection — that a ferry only carries four or five people — is answered by scaling the ferry rather than by landing the ship: cost per seat falls monotonically as fixed dry mass amortizes.

Two ceilings apply, and both are easy to miss. Aeroshell diameter caps the seat count before propellant does: entry mass at 50 seats needs roughly a 14 m decelerator, and the largest rigid aeroshell that ships inside a Starship is about 9 m, which caps a ferry near 10 seats at a ballistic coefficient of 250 and about 20 at Starship-class terminal velocity. And the crew ceiling is a staircase, not a line: a 20-seat ferry burns 69.3 t carrying 20 people and 60.1 t carrying 10, because the burn is dry-mass dominated. A 100 t tank farm funds one sortie and evacuates 20 people, not the 29 a linear reading gives.

That staircase is the real argument. Landing the ship makes return a step function — 50 people cannot leave until the full 683 t exists. A ferry makes return capability a continuous function of the tank farm, which is the only form in which a no-rescue architecture can honestly claim evacuation capability at all. It also converts population from a political question into an engineering one: crew count is stored propellant divided by propellant per seat, roughly 4.2 t per resident at a Phobos node with a 20-seat ferry, so the base grows by making propellant rather than by landing people. The optimum is two mid-size ferries, which is also the redundancy answer.

Phases

Five Phases, and Divisibility as the Manifest Rule

The reactor’s arrival, not the base’s construction, is what gates crew.

Prepositioning and power are conventionally merged and should not be, because what actually gates crew is the reactor arriving intact. The manifest discriminator is divisibility, not value and not mass: send divisible cargo early and accept losses, hold indivisible cargo until the attrition rate has been measured.

Which leads to the reactor decision, and it has to be made before the unit is designed because a monolith cannot be split later. At 50% landing reliability, one 100 kW unit gives a 50% chance of zero power; two 50 kW units give 25% zero and 75% at half or better; four 25 kW units give 6.3% and 68.8%. Two units minimum, four if it divides cleanly — and the requirement is not a bigger reactor but several of the one that already exists, which is a position Lockheed has argued publicly on program-risk grounds.

One correction to the naive version: that arithmetic assumes independent landings. At moderate common-mode correlation the four-unit benefit falls by a factor of eight, to about 1.8×. Split the units across windows, not across landings in the same window.

The master gate

The Descent and Ascent Vehicle Nobody Is Funding

Every orbit-staged architecture depends on a vehicle that exists in no program at any agency.

A ferry carrying crew between the Martian surface and an orbital node is the single piece of hardware the entire orbit-staged case rests on, and it is unfunded everywhere. It is also asymmetric in a way that has never been analyzed: based in vacuum, it descends aerodynamically and ascends propulsively, so it carries a heat shield through an ascent that does not use it — a 9 to 20% ascent propellant penalty depending on shield mass fraction.

Blue Moon MK2 is substantially the lunar equivalent of this vehicle, and it does not transfer: a lunar lander faces no atmosphere, no aerodynamic entry, and no thermal protection problem. The Martian version shares almost nothing with it but the concept.

The vehicle sits at the center of a substitution triangle with orbital propellant transfer and the surface ISRU plant. Investment in any one relieves the other two, and the trade has never been made deliberately. Choosing a lander by default is choosing an architecture by default — picking a lunar-derived lander commits a program to tanker campaigns and no ISRU customer; picking direct Starship landing on Mars commits it to a 683 t propellant plant. It is the highest-leverage vehicle decision in the entire program and it is currently being made by omission.

One more piece of context belongs here. Mars entry heritage is thinner than the confidence around it suggests: the cumulative record of successful Mars landings is equivalent to roughly 200 to 270 years of Falcon 9-class operational history, not the thousands of flights that make terrestrial launch look routine. The 50% attrition assumption is not pessimism. It is the honest read of a short record, and measuring it is the cheapest high-value thing a pathfinder campaign could do.

What would overturn this

The Findings That Would Change the Architecture

Each is a specific result, and three of the five arrive before 2032 if anyone funds them.

Record

Provenance and Data Sources