Three tests that should
kill Swovee.
A Mars concept does not become credible by accumulating features. It becomes credible by surviving the fastest tests that could prove it unnecessary or unworkable.
Do not begin with a full rover.
A polished integrated prototype can hide weak subsystems behind operator intervention. Swovee should begin with three ugly rigs, explicit kill thresholds and independent observers. Passing does not prove a Mars system; failing prevents years of architecture built on the wrong assumptions.
Can a local-material path produce repeatable, useful output?
Build a stationary feed-processing and placement rig. Run a representative Mars-analogue material through screening, conditioning and the candidate consolidation process. Cycle feed moisture, particle distribution, temperature and interruption duration. Track all imported additives, energy, wear and rejects.
Repeatable geometry and required material properties across lots; measured energy and imported mass support the campaign ledger; process restarts without disassembly.
KILL CONDITIONNo plausible process window, imported binder dominates installed mass, or cleaning and wear consume more mission time than placement.
Can the machine move and continue without creating a structural seam?
Place a long continuous test article across uneven terrain using at least ten move–stabilize–re-register cycles. Deny perfect GNSS. Introduce wheel slip, base settlement, sensor occlusion and one failed fiducial. Compare the complete geometry and interface strength to a fixed-cell control.
Closed-loop position uncertainty predicts actual error; seams remain inside the design tolerance; recovery does not require expert surveying after every move.
KILL CONDITIONRegistration error accumulates, the local datum cannot be recovered, or mobility overhead eliminates the utilization advantage.
Can it fail safely and resume after a simulated Mars blackout?
Run a multi-hour field mission with delayed, bandwidth-limited supervision. Inject blocked feed, pressure rise, dust-obscured perception, thermal derating, unexpected terrain, network loss and an actuator fault. The machine must protect itself, preserve evidence and select only approved recovery actions.
No unsafe motion; fault is correctly bounded; state survives reboot; operator receives a compact diagnosis; mission resumes without corrupting the as-built record.
KILL CONDITIONRoutine faults require real-time teleoperation, ambiguous state forces site abandonment, or AI bypasses deterministic safety limits.
A 120-day falsification program.
Mission ledger, comparison systems, hazards and kill thresholds.
Material loop, mobile datum mule and delayed-ops simulator.
Run designed faults with external materials and safety reviewers.
Proceed, pivot or stop—with raw evidence and an updated mass ledger.
Failure must be a deliverable.
If the team cannot publish a negative result, it is building a story rather than a machine.