Mining Saturn's Rings

View on GitHub → Updated 2026-07

Project ICEBERG is a mission-concept study. The vehicle flies to Saturn's B ring. The vehicle captures a 40-tonne chunk of nearly pure water ice in a fabric bag. The vehicle then carries the chunk to Earth in seven years. The vehicle uses the chunk as propellant during the flight. The cargo is also the propellant supply for the return leg.

I ran a 167-round desk-study campaign against the concept. The campaign used pre-registered hypotheses and Monte Carlo sweeps over the full architecture space. The purpose was to falsify the concept. The concept remained valid, but only in a narrow set of conditions. Closure depends on three specific engineering conditions. Nobody has proven any of the three.

Each condition has one demonstrator objective. This report gives the architecture, the method, the claims that the campaign falsified, and the results that remained.

Abstract

The orbital economy needs water. Water is propellant, radiation shielding and life support. Today every kilogram of that water is launched from the surface of the Earth.

Saturn's B ring is the largest deposit of nearly pure water ice in the solar system. The B ring is 99.5 to 99.9 percent water by volume. The ice is already broken into chunks of house size and smaller. The ring is a microgravity environment. A vehicle can match orbits with a chunk at a closing speed of millimetres per second. ICEBERG examines what is necessary to deliver some of this ice to Earth orbit.

The architecture is a vehicle with a water-fed microwave electrothermal thruster and a kilowatt-class fission reactor. The architecture has one subsystem that is completely new. That subsystem is the sublimation-capture trawl bag. The trawl bag makes the captured chunk the propellant supply for the return leg.

A 167-round analysis campaign gave a conditional verdict. The campaign used pre-registered hypotheses and runnable rounds. The campaign also used a Monte Carlo mission-graph framework. That framework sweeps 10 mission phases by roughly 36 options per phase.

At a 25-tonne commercial delivery floor, 5,656 feasible paths satisfy the closure conditions across 322 unique architectures. The best case delivers 39.5 tonnes in an 11.93-year round trip. All of these paths assume three things that nobody has demonstrated:

• The vehicle must capture a multi-tonne chunk in the ring actively.
• The microwave electrothermal thruster must operate continuously for months on ring-sourced water at a specific impulse. Only laboratory pulses reached that specific impulse until now.
• A flight fission reactor must be delivered on the program schedule.

The purpose of the campaign was to identify the three conditions and to give a cost for each one. The economics are equivalent to a regulated utility. The campaign records this result.

1. The basis of the concept

Water is the bulk commodity of any orbital economy that grows larger than a small scale. Water has these functions:

• Water is a propellant. The water is electrolyzed or heated directly.
• Water is radiation shielding.
• Water is thermal mass.
• Water is drinking water for the crew.

Every kilogram of water in orbit today came up on a rocket from the surface of the Earth. The launch price is some thousands of dollars per kilogram.

The usual proposed source is lunar polar ice. For the first decade of in-space water demand, the campaign agrees that lunar in-situ resource utilization is the better architecture. The lunar deposits are near to the Earth. Known mission classes can reach them.

The lunar deposits also have a low water content. The LCROSS impactor measured 5.6 ± 2.9 percent water by mass in the best-case cold trap at Cabeus crater. The average regolith has a water content of approximately 1 percent. A lunar mine must excavate, move and thermally process 20 to 100 tonnes of abrasive rock for each tonne of water. The mine operates at cryogenic temperatures in craters that receive no sunlight.

Saturn's B ring has the opposite properties. Cassini microwave radiometry gives the B ring a content of 99.5 to 99.9 percent water ice by volume. This is a higher purity than any ore mined on Earth. The main rings overall are above 99 percent. The deposit is already broken into pieces, because the ring particles have sizes from dust to the metre class. The particles orbit in a sheet, and the relative velocities between neighbouring particles are millimetres per second.

Excavation is not necessary. The vehicle matches orbits and moves slowly into the ring material. The ore grade is 18 to 100 times better than the lunar polar grade. The extraction energy is nearly zero. There is one disadvantage. The B ring is 1.2 billion kilometres away and the round trip takes 13 years.

ICEBERG examines whether this trade is ever justified.

2. The architecture

The mission uses one vehicle, ten phases and approximately 13 years.

1. Launch and injection burn. A Falcon Heavy launches a launch stack of approximately 50 tonnes to low Earth orbit. A chemical kick stage performs the 7.3 kilometres-per-second trans-Saturn injection. The vehicle then jettisons the kick stage. This is the only chemical propulsion in the whole mission.
2. Cruise and capture. The vehicle flies a 6.1-year Hohmann transfer. The vehicle then performs a multi-pass low-thrust capture at Saturn on the microwave electrothermal thruster. NASA's Dawn mission used the same low-thrust orbit-insertion approach at Vesta and Ceres.
3. Trawl. The vehicle moves into a circular orbit at the B-ring radius. The vehicle deploys a fabric trawl bag with a lining of ballistic weave and aerogel. An induced radial drift at millimetres per second moves ring material into the intake. The soft layers decelerate the ring particles inelastically. The light time to Saturn is approximately 83 minutes, therefore the trawl operation is fully autonomous. This operation decides whether the program succeeds or fails. Persons on Earth receive the result only after one hour and 30 minutes.
4. Cinch and convert. The bag closes. The sun-facing wall sublimates ice. The cold-side wall cryopumps the vapor back to frost. A heated harvest port meters vapor on demand into the propulsion feed. The cargo has three functions at the same time. It is the cargo, the propellant tank and the thermal management system.
5. The return cruise. The vehicle performs a multi-pass Saturn departure and then applies seven years of continuous chunk-fed thrust. A multi-flyby lunar-gravity-assist capture at Earth follows. The missions Hiten, WIND, Geotail and ARTEMIS proved this trajectory class. A low-thrust spiral then takes the vehicle to the depot. Aerocapture is not possible for this cargo. Tens of tonnes of ice at interplanetary speed become vapor in an atmosphere.

The propulsion is a water-fed microwave electrothermal thruster. A resonant cavity heats the water vapor directly. The thruster has no electrodes in the plume. It has no grids that erode. The design tolerates contamination.

Laboratory thrusters of this class operate in the 700-to-900-second specific-impulse band on water in short pulses. A NASA small-business program targeted more than 800 seconds. This performance is comparable to hydrogen-oxygen chemical rockets. It is approximately 2.5 times better than the best storable chemical propellants. The vehicle collects this propellant in space. A kilowatt-class fission reactor in the Kilopower lineage supplies the power.

Both of these statements have important limits. A later section gives these limits.

3. The trawl bag

Every other element of the architecture has a defined technology readiness level. These elements are:

• the water-fed microwave electrothermal thruster
• the small fission reactor
• deep-space autonomy
• the lunar gravity assists

Nobody has built a subsystem like the sublimation-capture trawl bag. The campaign treats the trawl bag as the primary engineering risk. The physics of this subsystem is also the least difficult.

The closing speed is the reason. Ring particles in neighbouring orbits move past each other at millimetres per second. The kinetic energy at capture is therefore very low. A 1-metre chunk of water ice with a mass of 500 kilograms enters the bag at 1 millimetre per second. This chunk has a kinetic energy of approximately 0.25 millijoules.

A larger chunk gives a similar result. A 10-metre chunk has a mass of 470 tonnes. At the same closing speed this chunk has a kinetic energy of only a quarter of a joule.

The energy does not damage the fabric. The problem is the geometry. A 10-metre chunk does not go through the intake. Therefore the design has a mesh pre-screen. The pre-screen size is the size of the largest admissible body. The design also has a forward-looking lidar that starts a reject maneuver when a body larger than the intake enters the approach corridor.

The two other bench questions are material aging and mass accounting. The bag laminate is a ballistic weave with an aerogel liner. The laminate must keep the vapor loss below approximately 5 percent of the cargo mass. This limit applies across the seven-year inbound coast at sun-facing wall temperatures of 200 to 280 kelvin. If the loss is more than 5 percent, the chunk-fed delta-v budget does not satisfy the rocket equation on the return leg.

A bench test answers two questions. The first question is whether current materials meet that value. The second question is whether the liner needs metallization.

As the bag fills, the center of mass of the launch stack moves. The orbit also changes. The vehicle holds a stationkeeping box of a few hundred metres through a multi-hour fill. This operation costs on the order of 10 metres per second of thruster delta-v. This value is small and the budget includes it.

The subsystem has three defined engineering questions. It has no physics obstacle. Every difficult item in the subsystem is testable on Earth at bench-test cost. This property makes the staged program structure in section 8 possible.

4. The method: 167 attempts to falsify the concept

A concept of this type causes strong motivated reasoning. It is easy to write a proposal that appears good. The important question is whether the concept survives an adversarial analysis campaign. That campaign uses rules that make undeclared self-deception difficult.

The campaign protocol comes from machine-learning experiment conventions. The repository contains the protocol at PROTOCOL.md. Each round is a directory. The directory holds a pre-registered hypothesis, a runnable run.py file and the results. The campaign states each hypothesis before the run. Each hypothesis is falsifiable and gives numeric ranges.

The campaign commits the results without changes. Each round also has a mandatory Revisit clause. The Revisit clause records whether the prediction held. If the prediction was incorrect, the clause records the cause. The campaign sorts each finding into adopt, drop or defer. The retractions stay in the record.

Later audit rounds marked several headline claims in the architecture decision matrix as RETRACTED in place. The audit trail identifies the round that falsified each claim.

The primary analysis tool is a mission-graph framework. The framework divides the full mission into 10 phases. The 10 phases are launch, assembly, outbound, capture, harvest, departure, inbound, Earth arrival and the processing variants. Each phase has approximately 36 options. The framework examines these options with a Monte Carlo sweep over the physics anchor values.

A closure predicate filters the space. The closure predicate has three conditions:

• The delivered mass is above a floor.
• The round trip is below a ceiling.
• Every phase is physically consistent.

The canonical sweep evaluates tens of thousands of paths per run.

Artificial intelligence research agents did the work. The research agents are Claude. The campaign used parallel sessions. Each session worked its assigned rounds against the shared protocol. The project owner set the direction. The owner locked the audited findings as ground truth and issued kill directives when an analysis went further than the evidence.

The round documents record a session codename for each round. The codenames titan, rhea, phoebe and hyperion come from the moons of Saturn.

I consider this a new method for desk studies. A 167-round pre-registered campaign is approximately a person-year of analyst work. The campaign took approximately ten weeks of evening work. The failure modes of this method are also new. Section 9 states these failure modes.

5. Claims that the campaign killed

The most useful results of the campaign are the claims that it falsified. The table gives examples from the retraction record of the architecture decision matrix.

The original proposal document in the repository contains reader notes. The notes identify the tables that later integration falsified. The notes stay in the document as an audit trail. Nobody rewrote the document without a record.

The campaign made this choice early. It is the most valuable convention of the campaign. A reader can trace each headline value to the round that set it. A reader can also trace each retired value to the round that falsified it.

6. Results that stayed valid

The canonical sweep gives the results that stayed valid. After the corrections, the sweep finds 5,656 feasible paths across 322 unique architectures at a 25-tonne commercial delivery floor. The best single delivery is 39.5 tonnes in an 11.93-year round trip. The most frequent pattern that satisfies the closure conditions is simple. That pattern has these steps:

• a single launch
• an autonomous assembly
• a low-thrust spiral out
• a chunk-fed spiral home
• a direct propulsive arrival or a lunar-assisted arrival

The framework also encoded less usual options. These options are Venus-Earth gravity assists, lunar-orbit processing waypoints and Saturn-moon flyby captures. These options satisfy no closure condition that the simple path does not satisfy better. One programmatic value has more effect than the physical lower bound. The table below gives the closure rate for each delivery floor.

The delivery floor has more effect on the closure verdict than any single physics axis. The customer sets that floor. A demonstrator-class mission satisfies the closure conditions easily. At the current anchor values, no architecture satisfies the closure conditions at a 50-tonne commercial delivery floor. The commercial case is possible only in the 20 to 40 tonne band. The requirement document lists the delivery floor as its most important open decision.

The requirement document sets the delivery floor provisionally at 25 tonnes. This value is 5 tonnes below the engineering ceiling. The value stays provisional until a full financial model is available.

The matrix above shows the audit sweep as a dashboard view of the closure conditions. The matrix has 48 cells. The chart outlines each cell that satisfies the closure conditions. Each cell that satisfies the closure conditions is in the corner of the matrix that needs both flight-unproven conditions together. An interactive version is available at robotrocketscience.com/projects/iceberg-matrix. The reader selects a delivery floor and reads the margin of any cell.

Two other results stayed valid. Small vehicles give better results than large vehicles. The campaign started with a 50 to 200 tonne vehicle grid. That grid came from launcher assumptions and not from analysis. Vehicles of 10 to 50 tonnes satisfy the closure conditions as well as the large vehicles or better.

The trawl-bag physics is the second result. The trawl bag is the one subsystem that is completely new. The trawl-bag physics gives three bench-testable engineering questions. The campaign found no physics obstacle.

The kinetic energy of a ring particle at ring closing speeds is in millijoules. The problem is the geometry and the material aging. Bench tests answer questions of geometry and material aging.

7. The three unproven conditions

The result of the whole campaign has three parts. Closure needs all three of the following conditions together.

Condition 1: active chunk capture at large scale. The desk-study anchor value assumes 85 percent single-pass capture efficiency. A decomposition of the capture sequence gives approximately 46 percent. The sequence is rendezvous, deployment, catch, containment and survival. The sweep shows that almost no cell satisfies the closure conditions below approximately 75 percent of the anchor value. At one half of the anchor efficiency, the best delivery is below 20 tonnes. The mission then does not reach the commercial delivery floor.

Cassini flew through the ring material and sampled the material passively. Nobody has actively captured a multi-tonne chunk in a ring environment. This condition has no flight precedent at any scale.

Condition 2: continuous water electrothermal operation at flight scale on ring water. The ground test evidence is valid. Laboratory microwave electrothermal thrusters reached the 700-to-900-second specific-impulse band on water vapor in short pulses. Momentus has flown water-fed microwave electrothermal propulsion on its Vigoride vehicles. The difference is the operating profile. Laboratory results come in 50-second pulses on clean water.

The mission needs continuous operation for months. The propellant is B ring water with approximately 0.3 percent non-icy contaminants. That water passes through a resonant cavity.

The sweep gives the sensitivity of the closure rate to the specific impulse. The 800-second anchor value is at the point of maximum sensitivity. A value 100 seconds lower reduces the closure rate by one half. A value 200 seconds lower reduces the closure rate to almost zero.

Condition 3: a flight fission reactor on schedule. The power section of the campaign contains the most negative results in the repository. The United States put exactly one fission reactor in orbit. That reactor is SNAP-10A, in 1965. Every flight program after SNAP-10A ended before it reached orbit. These programs are SP-100, Timberwind, Prometheus, and most recently DRACO, cancelled in 2025.

There were six programs and a cost of approximately 1.7 billion dollars. No program reached orbit. The KRUSTY ground test proved the physics at 2.4 watts per kilogram at system level. A flight program is not funded. NASA had not awarded Phase 2 of the Fission Surface Power effort at the end of the campaign.

The campaign treats the availability of the fission reactor as a Bayesian prior of program failure. The campaign applies this prior in addition to the physics. The campaign makes the commercial case conditional. A flight reactor must be under contract before the program commits fleet capital.

Each condition has a different failure mode:

• Condition 1 fails on the mechanism.
• Condition 2 fails on the endurance.
• Condition 3 fails on the programmatics.

Each condition has a demonstrator objective. Each demonstrator objective reduces the risk at two orders of magnitude less cost than the mission that it protects.

8. The economics in plain terms

The original proposal document described venture-scale returns from sovereign-scale cash flows. The financial rounds falsified that description. The campaign replaced it with a smaller and more defensible description.

The campaign examined one upfront commitment for the whole program. In that case the mean expected net present value is negative. The loss cases dominate the probability mass. The positive result occurs only in the top decile of the water-price outcomes.

The campaign then examined the program as a set of staged options. That structure has these parts:

• a demonstrator gate approximately every two years
• kill criteria at each gate
• fleet capital committed only after proof of Saturn capture

The distribution of the values changes, but the mean net present value stays negative. The gate 0 commitment decreases to approximately 11 percent of the program capital. The option to abandon has a value of approximately 2.3 billion dollars against the fleet decision. The decision gate stops the fleet in 65 percent of the Monte Carlo draws. Staged commitment protects against the downside. The campaign states this result in these words.

The infrastructure logic stays valid. Assume that the demonstrator missions satisfy the three unproven conditions. In that case the delivered product has the price behaviour of a regulated utility with a physical barrier to competitors. Orbital mechanics puts the second competitor 13 years behind. More capital does not remove this delay.

I state exactly what the campaign claims and what it does not claim. The campaign does not claim that ICEBERG is a good investment. At the current anchor values the mean expected net present value rounds give the opposite result. The campaign makes these claims:

• The option structure has a low cost in comparison with what it purchases.
• The first decision gate limits the downside.
• The cost to obtain the answer is three demonstrator missions of normal scope.

9. The value and the limits of the method

This project is also an experiment. The experiment measures how much an artificial intelligence agent analysis campaign can do for a mission concept. This section gives the advantages and the limits.

The parts of the method that worked are as follows. Pre-registration with mandatory revisits found errors in the campaign at a higher rate than I reached alone. Later rounds audited earlier rounds under standing rules. Those later rounds gave these results:

• the delta-v anchor value corrections
• the retracted closure headline claims
• the falsified tables of the original proposal document

The mission-graph framework made the question about unusual architectures a database query and not a discussion. The campaign also used parallel sessions. Therefore the audit rounds did not share context with the rounds that they audited.

The method also has limits. The method cannot make hardware. The method has the quality of its anchor values and no more. Every closure value in this report is desk-study grade. Public literature is the source. The campaign performed a sensitivity sweep on each value.

Each value is still a paper value. A bench test of the bag laminate permeability has more value than every Monte Carlo sweep in the repository. The campaign concludes that the next expenditure must go to hardware questions and not to more sweeps. The method has one more failure mode. An artificial intelligence agent campaign produces analytical confidence faster than it produces evidence. A large quantity of internally consistent documentation can appear more mature than the evidence permits.

The owner monitored for this failure mode. The kill directives in the audit trail are the result.

10. Conclusion

Saturn's B ring contains chunks of water ice of house size. The purity is higher than the purity of any mine on Earth. The physics of the mission is mostly established physics. The 167-round campaign did not falsify the concept completely. At a 25-tonne delivery floor, thousands of mission paths satisfy the closure conditions with the corrected anchor values. The campaign reduced every open question to three conditions.

The three conditions are as follows:

• The vehicle must capture the chunk.
• The vehicle must use the chunk as propellant.
• The vehicle must have sufficient electrical power.

Nobody has demonstrated any of the three conditions. All three are demonstrable at demonstrator scale for a small fraction of the program cost. The staged structure means the program does not have to commit fleet capital to obtain the answer.

The concept stays in the record as a low-probability option. The method is the pre-registered adversarial desk study at campaign scale. I expect to use this method again in the next month.

Update 2026-07-20 -- the de-spin propellant question

Rounds 168 and 169 ran after the first publication of this article. These rounds answered a question from the project owner. The de-spin of a captured chunk needs propellant. That propellant is delivered water. The rounds measured the quantity.

The answer has two parts. The de-spin operation itself is not a problem. Ring chunks rotate slowly. The Cassini-anchored prior of the campaign puts the median near 0.005 revolutions per minute. Collisions re-equilibrate the ring-particle spin every orbit. At those rates the removal of the angular momentum of a 200-tonne porous chunk needs grams to single kilograms (R_chunk_despin_budget).

Three of the five pre-registered bounds of the round still fell. All three fell because I used an approximate calculation at the solid-ice corner of a grid. The porous corner of that grid carries 1.5 times the moment of inertia. The values were incorrect, but the conclusion remained valid. The campaign records the lesson. Set the pre-registered bounds at the worst corner of the grid.

The passive method costs less. At any plausible cinch tension the friction of the bag fabric removes the cargo spin in minutes. The friction also couples the cargo to the launch stack.

The rounds also found one software defect. The first closure reprocess read a results field that did not exist. The reprocess scored every cell zero. The result recorded a pass over an empty denominator. A hypothesis that holds over zero cases is not tested.

The second part is the mechanism that the project owner identified (R_com_offset_thrust_alignment). The inbound burn consumes approximately 40 percent of the cargo through a harvest port. The harvest port draws from one side. The center of mass of the cargo then moves 1.2 to 1.7 metres away from the thrust line. The offset is the usual condition for this architecture. The offset is an unusual condition only if the design makes the draw symmetrical.

The corner reaction control thrusters can oppose that torque. This method needs 50 to 500 tonnes of propellant at any power level. This mass is equal to several deliveries. The delta-v fixes the angular impulse. Therefore a larger fission reactor does not reduce this cost.

The vehicle can instead steer the thrust vector through the center of mass as the center of mass moves. The cost then decreases to a cosine loss with a limit of 2 percent of the inbound propellant. On the canonical 40-tonne chunk this loss is 31 to 492 kilograms. In the audit sweep the two strategies give completely different results. The steer-through strategy changes none of the five cells that satisfy the closure conditions. The reaction control strategy removes all five cells.

The propellant cost of the spin is as follows. The de-spin needs grams of propellant. The steer-through strategy needs tens to hundreds of kilograms of propellant.

The result also gives a new hard requirement for the vehicle. The vehicle needs approximately 12 degrees of thrust-vector authority. This authority must track a plus-or-minus 1.7 metre center-of-mass offset. An actively symmetrical harvest draw is the alternative requirement. The vehicle specification contains neither requirement at present.

A third round tested the basis of that requirement (R_harvest_draw_symmetrization). The one-sided draw is a design choice and not a physical necessity. The round gives the following result. The design puts the harvest port on the thrust axis. The design then rolls the launch stack about that axis one time each day. The Apollo missions used the same method.

The center-of-mass offset then decreases to centimetres. The 12-degree tracking requirement becomes a fallback for the condition with no roll. A 3-degree trim and a port-centering tolerance replace it. Three of the four pre-registered bounds of that round also fell. One of the incorrect bounds gave the most useful result of the round. The offset amplification scales with the draw fraction of the cargo.

Therefore small chunks are the difficult case for this problem. Small chunks are the easy case everywhere else in the campaign. At 25 tonnes the design must center the port to approximately 7 centimetres. The decision record now holds the low-cost requirement. The high-cost requirement is the fallback.

Update 2026-07-21 -- two audits of the campaign and a fourth unproven condition

The count is now 191 rounds. Approximately twenty rounds came after the de-spin update. These rounds are the strongest evidence for the method. The campaign used most of these rounds to audit its own earlier work in this same set of rounds. The campaign also reversed part of that earlier work.

The sequence started with a question from the project owner. I expected a simple answer. Beyond approximately 3 astronomical units the solar array cannot supply the microwave electrothermal thruster. The question was whether the vehicle can electrolyze the cargo water into hydrogen and oxygen. A fuel cell energy store that uses water then supplies the return leg. I created a sequence of rounds on this question.

The first results appeared good. The rounds gave a non-fission return-trip advantage. That advantage increased from 1.7 times to 3.3 times across six rounds. Then one round (R_departure_anchor_reconciliation) performed the check that the protocol requires. That round re-derived the one value that every round in the sequence had inherited without a check. That value is the Saturn departure delta-v, carried at 1.5 km/s.

The 1.5 km/s value came from an early concept of operations estimate. An audit in May identified that estimate as insufficiently itemized. From a circular orbit in the B ring, escape alone needs 7.5 km/s. The honest impulsive minimum is 8.45 km/s. The 1.5 km/s value was incorrect by a factor of five. The whole sequence of rounds depended on that value.

At the correct value the non-fission variant needs 464 tonnes of hydrogen and oxygen staged at Saturn to depart. That is a 39-to-58-times penalty and not a 3.3-times advantage. The next round of the campaign falsified the headline result of that week. The retraction stays in the record, as the earlier retractions do. The mechanisms from that sequence stay valid as resilience components on the mission legs that remain valid. The round-trip claim reverts, and a vis-viva proof now supports the reversion.

This is the most important result of the project. A pre-registered campaign that publicly falsifies its own best week shows that the method operates correctly.

The correction made the fission reactor condition more severe. The campaign re-derived the power requirement from the departure burn alone. This derivation made no reference to the older chunk-movement argument. The result is 155 to 175 kilowatts electric for a two-year departure. That value is six to seventeen times the ten kilowatts of the Kilopower flight design. The value is deep inside the band that the credibility audit of the campaign already rated at under two percent deliverable.

I used one round to look for an alternative. The alternative was a Cassini-style gravity-assist tour at Titan to increase the departure velocity without propellant. This idea comes from a sister project on gravity-assist trajectories. The ephemeris check falsified this option (R_titan_tour_ephemeris). A flyby conserves the velocity relative to Titan. Every orbit between two flybys must stay bound for the vehicle to return.

These two constraints limit the flyby chain. A pure flyby chain limits the departure excess velocity to approximately one half of the necessary value. Titan can drop a spent stage into Saturn at almost no propellant cost during a descent. The gravity well assists a descent and not an ascent. Titan cannot supply the departure. Condition 3 became more difficult. The physics of the mission needs a fission reactor that the credibility audit of the campaign rates as not credible.

The campaign examined a relay architecture and then rejected it. A reader suggested this alternative. A small dive shuttle moves chunks up the gravity well to a relay mothership in a higher orbit. The cost of the fission reactor is then divided over many missions. The campaign gave this alternative four rounds and four separate audit passes (R_chunk_relay_staging through R_relay_ledger_reconciliation).

The four passes examined these items in this order:

• the lifetime
• the net present value
• the operations
• the departure

Every pass gave the same marginal result, exactly at the utility cost of capital. The fourth pass made the result negative. That pass priced the departure of the relay mothership correctly. The correction is the same 1.5 km/s against 8.45 km/s error in a different architecture. The advantage of the fleet then falls below the discount rate at every setting.

The relay architecture is valid and it is better than the baseline. It is still not sufficient. The simple single-vehicle design with a fission reactor is again the baseline of record.

The campaign found a fourth condition below the three engineering conditions. The financial rounds used the price of delivered water as a free parameter. One round (R_clearing_price_time_path) connected the price to the values that determine it. Water delivered to low Earth orbit cannot sell for much more than its cheapest substitute there. The two substitutes are a launch of water from the Earth and water moved down from lunar ice. The cost of both substitutes decreases with time.

The Earth-launch anchor value is roughly $3,900 per kilogram in 2025. That value moves toward a few hundred dollars per kilogram by 2040. The price decline is about 13 percent a year. That rate is faster than either discount rate that the campaign uses. This is the important result. If the price of a delivered product falls faster than the cost of capital, the product is not profitable before the mission arrives.

Over the 13-to-23-year timeline of ICEBERG the substitute becomes several times cheaper before the first delivery arrives. The exact rate is arguable. If lunar in-situ resource utilization is the binding substitute instead of Earth launch, the price declines on a different curve with a slower decline. The direction of the change is not arguable. At every Earth-accessible destination the substitute is low in cost and the cost continues to decrease.

The only alternative is to sell water at a destination far from the Earth. At such a destination a launch from the Earth is prohibitive in cost. That demand is unproven in the 2032-to-2050 window of the mission. That demand does not appear until launch cost is low. The same low launch cost reduces the price near the Earth.

The mission therefore sells its water in one of two ways. It sells at a low price into a market where the substitute price falls faster than the discount rate. Or it sells at a high price into a deep-space market with no demonstrated demand.

This is a real fourth condition. It is an economics condition below the three engineering conditions. It sets the price of the market that the engineering depends on. The mission cannot influence this condition, because it cannot change the launch cost curve.

The last three rounds improved the risk register and did not change the verdict. Every value in these rounds is a desk anchor value with no flight precedent. The ranges are the result. The single values are not the result.

The chunk handoff of the relay architecture has a low-probability failure case. In that case an incorrect berthing operation destroys the relay mothership that carries the fission reactor. A single handoff has a probability under 2 percent. A full mission has several handoffs in sequence.

At the pessimistic end of these estimates the cumulative probability of the loss of the relay mothership reaches about 8 percent. This event ends the program. No per-chunk reliability value shows this risk.

The disposal of the spent fission reactor is low in cost for a fleet architecture. A fleet leaves the spent reactor cores at Saturn. The disposal is a controlled entry into the planet. Cassini performed this maneuver in 2017. The single-vehicle baseline is the more difficult case. The Cassini precedent does not apply to the single-vehicle baseline.

The reactor of the single-vehicle baseline supplies power for the return leg and therefore returns to the Earth. An activated reactor core arrives in cislunar space. This condition creates a separability requirement. It also creates a launch and reentry licensing problem. Nobody has given a cost for that problem.

The trawl bag is the one subsystem that is completely new. It is also the one item that this desk campaign cannot resolve on a desk. A 40-tonne non-rigid load with internal fluid motion has two to ten times the mass of the vehicle that berths it. The stability of the coupled dynamics needs a full six-degree-of-freedom simulation and not a spreadsheet calculation. That simulation is now the item of highest value in the backlog.

These rounds do not change the headline verdict. The verdict now has four conditions:

• The vehicle must capture the chunk.
• The vehicle must use the chunk as propellant.
• The vehicle must have sufficient electrical power.
• The mission must sell the water into a market with a falling price.

This update made each of the three engineering conditions more specific. It also added the fourth condition. This is the purpose of the method.

Update 2026-07-27 -- the closed-loop mass accounting, the last corridor, and the price of water

The count is now 215 rounds. In this update the campaign completes its task. The directive from the project owner was direct. That directive had four parts:

• Audit everything.
• Try to satisfy the closure conditions realistically.
• Try to satisfy the closure conditions economically.
• If closure is not possible, state exactly what closure needs.

The campaign did all four. The commercial verdict is negative. A reader can check the reasons for that verdict. I consider the 24 rounds that produced this result the best work in the repository.

First, the mass accounting became correct. Every closure value earlier on this page came from sweeps with the dry mass of the vehicle as a free parameter. The analyst selected a vehicle and a chunk, then tested whether the delta-v budget balanced. The framework now uses closed-loop mass accounting. The framework derives the mass of the vehicle from these items:

• The fission reactor.
• The radiators.
• The tanks.
• The capture equipment.

The framework iterates all four items to a fixed point. A conserved mass ledger then passes through every burn and every jettison. The power available in each phase must match the fission reactor on board. No propellant enters the ledger without a source.

That accounting found errors in the earlier values. In one case a vehicle had less mass than its own fission reactor. In another case the model took the chemical kick stage from the hardware of the vehicle itself. In a third case a closure revival was a simulation defect (R_closing_path_self_consistency). The campaign reversed that result.

Under the closed-loop mass accounting, the record-technology sweep enumerates 1,744 end-to-end mission paths. Zero of them complete the round trip. This result holds under all of these conditions:

• Any delivery floor from half a tonne to 25 tonnes.
• Any mission ceiling out to 35 years.
• Any flyable power class.

A relaxation of the ground rules of the mission also gives no closure. In that round, zero of 1,744 paths even reach the low Earth orbit depot (R_l0_relaxation_envelope).

This stretch produced a standing protocol rule. Any round that claims a new architecture that satisfies the closure conditions must give a mass ledger that a person can check by hand. That ledger must follow the launch mass through every burn to the delivered payload. No person can cite the claim before that ledger exists. That rule found errors. It also found my own errors.

The campaign then reversed the question. No architecture satisfies the closure conditions. The campaign therefore asked which conditions closure would need. The inversion round (R_closure_boundary_inversion) changed each performance axis to an extreme value, one axis at a time:

• The specific impulse went to 10,000 seconds.
• The fission reactors went to half a megawatt.
• The dry mass went toward zero.
• The launchers went to 32 times a Falcon Heavy.

No single parameter satisfies the closure conditions at any value.

The reason for that result is important. One sentence states the complete result for electric propulsion at Saturn. An increase of the specific impulse reduces the propellant mass. It also increases the burn time that the available power level must sustain.

Closure is therefore possible only in a narrow window of specific impulse. That window is roughly 2,000-4,000 seconds. The best value is not the maximum value. Inside the window the other limits still apply. Closure needs at least three parameters to change at the same time.

One of those limits then removed one half of the candidate architectures. Two corridors appeared to satisfy the closure conditions. In every earlier round the campaign priced the electric-return family at a 9.0 km/s Saturn departure. That value belongs to an impulsive burn, which gets the Oberth effect. The vehicle uses its propellant deep in the gravity well and gets more energy for the same propellant.

A low-thrust spiral is far from the bottom of the gravity well for most of the maneuver. It does not get that benefit. A direct integration of the spiral gives the correct cost at flyable electric thrust (R_departure_anchor_resolution). That cost is 22.5 to 24.2 km/s. This value is two and a half times the anchor value. A thrust acceleration of about 1 m/s² is necessary to approach the old value.

That acceleration is a hundred times the flyable band. The rocket equation increases the effect of a 2.5 times delta-v error by a large factor. The complete electric-return family fails at every launch scale that the campaign permits.

This result gave two protocol lessons:

• A delta-v value is valid only together with its thrust regime.
• The conserved quantity of a repeated maneuver must be carried as state, and not recomputed.

The second lesson has a cause. A lunar gravity assist braking claim from May violated energy conservation. A real lunar gravity assist removes at most 2.1 km/s of an Earth arrival. The old round booked 5.8 km/s.

One corridor remains. It is an unusual architecture. The vehicle harvests 500 to 1,000 tonnes of ring ice. The vehicle electrolyzes part of that ice at Saturn into hydrogen and oxygen. The vehicle burns those products in a nuclear-thermal-class engine at 750-900 seconds. That engine is the only propulsion class with a specific impulse and a thrust regime that both fit the departure window (R_saturn_water_v2_electrolysis_departure).

Each mission needs two to three chunk captures, because no single admissible chunk carries sufficient water (R_multi_chunk_per_mission). The multi-chunk physics has a very low cost. A re-rendezvous between chunks costs tens of metres per second in a 30-to-300-km spacing band. The de-spin of a 400-tonne chunk costs about a kilogram of propellant.

The reliability calculation gives severe results. At the optimized capture point of 96.1 percent per capture, a three-chunk mission holds an 89 percent joint success rate. At the un-optimized engineering prior of 54.9 percent, the joint success rate falls to 30 percent by the second chunk. Capture reliability is no longer a margin item. It determines the mission.

The corridor needs roughly 875 to 1,140 tonnes of launch mass in low Earth orbit. That mass is six to eleven fully-refueled-Starship-class flights. The corridor also needs a 25-year mission waiver. It delivers 32 to 88 tonnes of water.

I must report a correction inside that launch-mass value. I claimed that the launch requirement of the corridor was "largely an accounting artifact" of a frozen propellant-fraction convention. A pre-registered round tested that claim and falsified it (R_per_phase_propellant_fidelity). The requirement did decrease from a coarse factor of 16 to a bisected factor of 8.75 to 11.4. The cause was granularity, and not the cause that I claimed. The retraction is in the round document, per protocol.

The financial model then ran and it answered the commercial question. The financial model was an open dependency in the decision record for fourteen months. The model prices the corridor with the most favorable available inputs:

• A mature Starship launch price of $100 per kilogram.
• A $0.5 billion marginal vehicle.
• Operations at $40 million per year.
• A discount rate of zero.

The breakeven price on delivered water is $15,513 per kilogram. The top defensible bulk price band is $5,025. That band assumes that orbital demand grows to 100 tonnes a year. At a sovereign 3 percent discount rate the breakeven price rises to $26,176. At a corporate discount rate the breakeven price increases by approximately a factor of three again.

Those price bands also change with time. Delivered water cannot sell for much more than its cheapest substitute. The substitutes follow the launch cost curve down at about 13 percent a year. The mission takes 19 to 21 years to deliver. At arrival, every band has decayed to between $27 and $1,849 per kilogram. That price is one to two orders of magnitude below the breakeven price.

The total anchored demand of that market is 40 to 60 tonnes a year. One delivery of 88 tonnes is equal to approximately two years of the complete addressable market (R_corridor_financial_model). Every cost that the model excludes increases the breakeven price. The excluded costs are the program development, the fission reactor, the mechanism qualification and the insurance. The campaign has answered the fourth condition from the previous update.

The commercial case for ICEBERG on the Saturn round trip is negative. The financial model that the decision record required gives this result. The result does not come from an absence of a model.

The project now has three results that stay valid on their own merits.

The demonstrator bench program became more specific and not less specific. Porous-ice harpoon penetration and capture-cinch reliability were already the two bench items. The multi-chunk work added a third measurement that nobody has made. That measurement is the number density of hundred-tonne-class chunks in the B ring. The measurement sets whether the multi-capture spacing exists at all. It is a remote-sensing question and not a flight program.

The sovereign and science framings survive by construction, because they use criteria that the breakeven table does not measure.

The near-Earth-asteroid successor campaign satisfies the closed-loop vehicle mass condition with solar power at every point where the Saturn mission cannot. It delivers 8 to 11 tonnes, which is below the commercial level. This is a small value that the analysis supports. It is not a large value based on hope.

The corridor stays in the record as an existence proof of the requirements for closure. The record prices the corridor in launches, fission reactors, reliability and dollars. The corridor is not a commercial business.

One year of this method falsified the mission that it examined. The campaign published the result and the supporting evidence. Readers ask me whether this result is a disappointment. It is the best possible outcome short of closure. The concept started as an unquantified idea. It ends as a defined boundary, and each part of that boundary has a number:

• The vehicle must capture the chunk.
• The vehicle must use the chunk as propellant.
• The vehicle must have sufficient electrical power.
• The mission must sell the water into a market with a falling price.

The method that produced this boundary is reusable on the next concept. Saturn's rings remain available for a future study.

Technical details

• Documents: the campaign record contains the concept of operations, the L0 and L1 requirements with a variance log, and the risk register. It also contains 22 architecture-decision-record axes. It contains 215 analysis rounds with pre-registered hypotheses, and the rounds are runnable. It contains a closed-loop mission-graph framework with 10 phases and ~36 options per phase. That framework uses a conserved mass ledger and a derived vehicle mass, and it has 179 tests.
• Software: Python 3.13, numpy, scipy, Cantera for nozzle thermochemistry, and matplotlib. uv manages the environment. Unit tests cover the physics models.
• Method: the campaign registers each hypothesis before the run, per PROTOCOL.md. Artificial intelligence research agents work in parallel orchestrated sessions. Those research agents are Claude. The project owner locks the ground-truth findings. The record keeps the retractions.
• Verification: every headline value in this article traces to a round document in the repository. The text cites the sensitivity sweeps inline. The corrected anchor values come from primary sources. Those sources are the Edelbaum bounds, vis-viva, Cassini radiometry and LCROSS.
• Public availability: the repository contains the complete record of the campaign, including the retractions.

github.com/robotrocketscience/iceberg →