Cislunar Foundry

The question

Cislunar Foundry asks a deliberately bounded question: what has to become true, in what order, for a programme to deliver its first net tonne of lunar-derived propellant to the Earth-Moon L1 region?

The endpoint is not a mature lunar economy. It is an end-to-end systems test. The campaign must prospect a site, land equipment, establish durable power, excavate volatile-bearing regolith, recover water, purify and electrolyse it, liquefy matched hydrogen and oxygen, load a reusable tanker, fly to EML1, and transfer enough cryogen that one tonne remains after loss.

The playground is a scenario model. It is not a forecast, a mission design, or a claim that any one organization has adopted this architecture.

Why EML1 is shown where it is

The 3D scene uses the circular restricted three-body problem. Earth and Moon are fixed in a frame rotating with their mean orbit. The collinear equilibrium is found numerically from

[ x - \frac{(1-\mu)(x+\mu)}{|x+\mu|^3}

• \frac{\mu(x-1+\mu)}{|x-1+\mu|^3}=0, ]

where (\mu) is the Moon's share of the total Earth-Moon mass. With a mean separation of 384,400 km, the solution is roughly 326,000 km from Earth's center.

That calculation anchors the EML1 marker. It does not make the displayed transfer arcs trajectories. The arcs are illustrations between modeled locations; there is no orbit propagation, delta-v optimization, ephemeris, or station-keeping model. Earth and Moon radii are also exaggerated relative to their separation so both remain legible. The 100,000 km range rings retain physical labels even though body size does not.

Four decisions close the campaign

1. Prospecting

The first launch buys down uncertainty. A focused orbiter is cheap but leaves correlated blind spots. A distributed network combines orbital and surface measurements at greater cost. A fast polar scout reaches the surface earlier but accepts a thinner resource model.

This matters because a water grade is not a universal property of the lunar south pole. Illumination, thermal history, terrain, trafficability, deposit geometry, contaminant load, and recoverability all belong to a site. The default 35 kg of recoverable water per tonne of regolith is a stipulated scenario prior, not a reserve estimate. NASA's lunar propellant production assessment is one anchor for the process chain and its technology needs.

2. Surface power

Excavation and cryogenic storage need more than an average energy budget. They need useful power at the moments when equipment must operate. Solar and storage can be deployed incrementally, but illumination and storage geometry constrain uptime. Fission reduces that exposure. The hybrid option retains a fission floor and uses solar for peak industrial load.

NASA describes a 40 kW-class fission surface power target designed for long-lived surface use. The playground aggregates multiple units when it represents a fission architecture. Mass, deployment reliability, bus losses, redundancy, and qualification schedule remain priors.

3. Autonomy and redundancy

The extraction choice asks where availability comes from. Earth-supervised robots reduce software ambition but remain communication- and staffing-bound. Higher autonomy moves planning, diagnosis, and recovery to the surface. A redundant fleet buys availability with landed mass and interchangeable hardware.

The model does not claim a known reliability for lunar autonomy. Each choice acts through a transparent productive-availability multiplier and event-risk adjustment. Long-duration polar operations would be needed to replace those priors.

4. The transport accounting boundary

The transport decision changes both architecture and accounting.

• Buy transport: surface industry and depot integration are inside the programme; mature launch and cislunar delivery are purchased as services.
• Co-develop the tanker: launch remains external, while tanker, depot, transfer, and integration are shared development responsibilities.
• Fund the stack: launch, tanker, depot, transfer, and surface industry all sit inside the programme boundary.

A smaller included cost can therefore describe a larger total capability supplied by other actors. The interface always shows included programme cost separately from external enabling capability.

The conserved mass chain

The model carries mass through five visible stocks:

[ m_{\text{regolith}} \rightarrow m_{\text{water}} \rightarrow m_{\text{H2/O2}} \rightarrow m_{\text{tanker}} \rightarrow m_{\text{EML1}}. ]

Recoverable water is

[ m_{\text{water}} = m_{\text{regolith}} \left(\frac{g_{\text{ice}}}{1000}\right) \eta_{\text{recovery}}, ]

where (g_{\text{ice}}) is kilograms of water per tonne of excavated regolith. Usable propellant applies purification, electrolysis, and liquefaction yields. Delivered mass then applies boil-off and transfer loss:

[ m_{\text{delivered}} = m_{\text{propellant}} (1-f_{\text{boiloff}}) (1-f_{\text{transfer}}). ]

The productive rate is the minimum of four capacities:

[ \dot m = \min( \dot m_{\text{resource}}, \dot m_{\text{power}}, \dot m_{\text{plant}}, \dot m_{\text{transport}} ). ]

That minimum is the active bottleneck shown by the playground. Raising a nonbinding capacity does not move the endpoint.

Cryogens remain a programme risk

Producing water is not equivalent to delivering propellant. Hydrogen and oxygen have to be purified, liquefied, stored, loaded, retained in flight, and transferred. NASA's Cryogenic Fluid Management programme treats storage, transfer, gauging, and pressure control as linked technology problems. The GAO 2026 major projects assessment also identifies cryogenic transfer as a major Human Landing System risk.

The playground represents this chain with explicit yields and loss fractions. It does not represent tank thermodynamics, propellant settling, zero-gravity acquisition, chilldown, venting, or transfer hardware in physical detail.

Why the cost is a range-shaped object

The ideation conversation converged on roughly $50B to $100B, with about $70B as a central order-of-magnitude answer for an autonomous lunar plant, tanker, and pilot EML1 node. Those numbers are not observations. They are a synthesis prior.

Public programmes provide scale analogues, not direct calibration. GAO's 2026 assessment reports multi-billion-dollar baselines for Gateway Initial Capability and Human Landing System Initial Capability. NASA's Inspector General reported nearly $7B obligated to Human Landing System work and more than $18B projected through fiscal year 2030. NASA's fiscal year 2027 request also groups surface power, autonomous logistics, robotics, propellant production, propulsion, and cryogenic management as separate enabling lines.

None is the Cislunar Foundry architecture. The model therefore uses them to discipline order of magnitude and work-package structure, not to claim empirical validation.

Expected, risk, and ensemble modes

Expected mode carries event risk as an expected delay and overrun. It is a smooth baseline.

Seeded risk mode samples one reproducible campaign. Random draws are addressed by seed, event id, and attempt, so adding or retrying a later event does not rewrite earlier history.

Ensemble mode runs 300 seeded histories and reports schedule and cost quantiles plus the fraction meeting the declared deadline and included-cost cap. Event risks are conditionally independent after decision and reserve adjustments. Correlated political, supplier, common-design, and macroeconomic failures are omitted, so the ensemble should not be interpreted as a statistical confidence interval.

Frozen 2026 evidence ledger

What would change the model most

The highest-value evidence would be:

1. Ground-truth polar drilling with volatile concentration, depth, contaminants, trafficability, and processing tests.
2. Multi-year autonomous excavation and maintenance data under polar thermal and lighting conditions.
3. Flight-qualified surface-power mass, deployment, availability, and delivered bus power.
4. Repeated flight-scale cryogenic storage and transfer demonstrations.
5. A traceable integrated work breakdown structure for a named programme and accounting boundary.
6. Propagated tanker and depot mission designs replacing the illustrative arcs.

Until then, the useful output is not one dollar value. It is the visible dependency structure, the bottleneck, the mass balance, the accounting boundary, and the decisions that move them.

Visual asset credits

• Earth albedo: Three.js example Earth texture, derived from public Earth imagery and stored locally for deterministic rendering.
• Moon albedo: NASA Scientific Visualization Studio CGI Moon Kit, based on Lunar Reconnaissance Orbiter data.