Pass 1 — Research (q2-lunar-ascent-cost)

Initial scoping research. The leaf question demands a 2026–2040 trajectory of cost per kg from lunar surface to LEO, broken out by chemical rocket vs mass driver. This pass gathers candidate sources; no claims written yet.

Sources captured

Tier-1 (concrete engineering or pricing anchors)

• handmer-mass-driver-2026 — Casey Handmer's May 2026 first-principles lunar mass driver design exercise. Per-shot 200 kg, 1 t/3s cadence, 128 m track, $10/kg assumed product price. The cleanest near-contemporary mass driver engineering reference.
• metzger-2023 — The gear-ratio paper. Provides Γ_LEO ≈ 14 under chemical, Γ_LEO ≈ 1 under SEP return. Bridge to q4.
• coutts-sowers-2025 — VOI framework for lunar ice. Anchors lunar surface propellant price at $500/kg and Starship LEO cost at $30–300/kg.
• wiki-delta-v — Canonical ΔV values. Lunar surface to LLO 1.87 km/s; full Moon-to-LEO chemical 5.93 km/s.
• starship-hls-wiki — Operational Starship HLS parameters. 100 t headline payload, 12–14 tanker flights per mission, $2.89B HLS contract.
• orbital-refueling-newspaceeconomy — Dec 2025 trade-press analysis. 1,200 t propellant per HLS mission; ~$400M per mission estimate.

Tier-2 (reference / corroboration)

• wiki-mass-driver — Mass driver physics envelope. 50–90% efficiency; theoretical 10.5 km/s at 5600 g; loose $47M / 10 kg projectile estimate.
• lunarpedia-mass-driver — Lunar-specific reference. 2.4 MJ/kg energy. Track-length-vs-acceleration tradeoff.
• sciencearray-mass-drivers — Secondary synthesis with $10/kg target and historical O'Neill $1/lb projection. Triangulation for Handmer.
• payload-cargo-landers — Operational cargo-variant payloads (12–15 t). Distinct from the 100 t crewed-HLS headline.

Sources attempted but not fetched (worth chasing in source-review)

• Kornuta et al. 2019 Commercial Lunar Propellant Architecture — primary source for lunar propellant production cost. PDF behind paywall / OSTI binary; secondary citations only.
• Lewis et al. NASA "Parametric Assessment of Lunar and Mars Ascent Vehicle Synergy" — direct NASA ascent vehicle cost study. PDF binary; couldn't extract text via WebFetch.
• AIAA 2025-4123 "Cost-Benefit Analysis of Lunar Mass Driver Technologies" — recent peer-reviewed mass driver economic analysis. Paywalled (HTTP 403).
• Ethan Miller 2023 thesis (SJSU) — student deep-dive on lunar mass driver implementation. PDF binary.

These deferred sources are flagged for a future source-review pass; they would likely tighten the mass driver capital-cost estimates and the lunar ascent vehicle dry-mass numbers, but the headline conclusions of q2 don't depend on them.

Key public figures for future Source Reviews

(For sub-pass 4 — applies to source-review or to a later cross-source pass)

• Casey Handmer — independent technical commentary, formerly JPL
• Philip Metzger — academic anchor for cislunar economics framework
• George Sowers — ULA's original lunar-propellant price-setter, now Colorado School of Mines
• Dallas Bienhoff — Boeing/private cislunar logistics consultant (referenced but not directly fetched)
• Gerard O'Neill — historical foundation of the mass driver concept (1974 work)

Initial pattern of evidence

Two architectural families dominate the literature, and they don't compete in the same era:

Chemical-rocket ascent (operational 2026–2035):

• Earth-imports-only: ~$30,000–50,000/kg lunar surface to LEO, dominated by terrestrial-launch cost amplified by Γ_chemical ≈ 14
• Partial-ISRU: drops to ~$2,000–5,000/kg if lunar surface propellant available at Sowers's $500/kg target
• Full-ISRU mature operations (post-2035): potentially $500–1,500/kg, still dominated by ascent-vehicle hardware amortization

Mass driver (post-2035–2040 at earliest):

• Handmer's design closes at $10/kg "rocks in lunar orbit," but only at 10 million tonnes/year throughput
• Bootstrap problem: getting the mass driver to the Moon and operational is itself a multi-billion-dollar chemical-rocket project (q5 territory)
• For LEO delivery, additional propulsion needed after mass-driver launch; this adds back a chemical or SEP stage

Engineering vs commercial decomposition

A critical distinction must be preserved in the calc pass: lunar-ascent vehicle cost ≠ lunar-product LEO delivery cost ≠ effective $/kg for the comparison framework.

• Ascent-only (lunar surface to LLO): drops out the Earth-to-LEO term but adds lunar ascent vehicle hardware + lunar ops + lunar-surface propellant
• Surface-to-LEO chemical: ascent + trans-Earth injection + LEO insertion (or aerobraking)
• Surface-to-LEO mass driver: kinetic launch + downstream chemical or SEP stage for LEO insertion (mass driver alone cannot circularize)

For the lunar-manufacturing competitiveness math (the root question of the report), the relevant comparator is lunar product delivered to LEO at full system cost, including the cost of getting the lunar infrastructure there in the first place. q2 will focus on steady-state delivery cost assuming the lunar infrastructure exists, and flag the bootstrap problem for q5 (capital buildup).

What to chase in subsequent sub-passes

• calc: derive lunar-to-LEO cost from first principles. Three scenarios matching q1's structure:
  • aggressive-ISRU: lunar propellant available at Sowers price target; SEP return architecture; mass driver operational by late era
  • partial-ISRU: lunar propellant available but at higher price; chemical-only architecture
  • Earth-imports-only: no lunar propellant; HLS-style chemical round-trip with Earth-imported propellant
• reconcile: compare derived numbers to Coutts-Sowers, Metzger, Handmer. The disagreements ARE the data here.
• source-review: point-by-point Handmer + Metzger + Coutts-Sowers + others. Verify quote authenticity.
• write: produce trajectory with explicit scenario bands corresponding to architectural choice. Use TAI/industrial-explosion sensitivity as a separate axis (mass driver becomes feasible much earlier under TAI).

Anti-pattern check

• ✓ Sources sealed for calc (this is research, no claims yet)
• ✓ No "by 2040" framing locked in — q2 intends to deliver a trajectory with conditional scenarios and acceleration sensitivity
• ✓ Voice register kept dry; no "real-but-contingent" / "the dominant lever" editorializing
• ⚠ Several primary sources (Kornuta, Lewis, AIAA-4123, Miller) gathered indirectly via secondary citations; flagged for source-review pass
• ⚠ Single contemporary mass-driver-engineering source (Handmer); should triangulate against AIAA-4123 if it can be obtained

[codex cli ok] overall: verdict: weak summary: "The chemical Tsiolkovsky arithmetic and printed cost tables mostly check out from the stated inputs. But this is not a first-principles derivation: most load-bearing numbers are asserted, and several interpretation/calibration claims contradict the tables."

findings:

• target: "delta-v-budget" target_kind: section verdict: weak quote: "Lunar surface to LLO | 1,870 | Lunar gravity well + ascent loss; canonical" reason: "The totals are summed correctly: 1,870 + 700 + 300 = 2,870 m/s and 1,870 + 700 + 3,000 = 5,570 m/s. But the component values are mostly canonical/common-knowledge inputs, not derived; the 300 m/s aerobraking residual and 3,000 m/s propulsive capture are especially architectural assumptions." severity: high

• target: "propellant-mass-fractions" target_kind: section verdict: pass quote: "Hydrolox (Isp 450 s), aerobraking | 2,870 | 0.478" reason: "Using f = 1 - exp(-dv / (Isp * g0)) gives 0.478, 0.717, 0.556, and 0.794 for the four listed cases. These numbers are arithmetically correct." severity: low

• target: "tsiolkosvky-mass-algebra" target_kind: topic verdict: pass quote: "m_prop = [f / (1-f) * m_payload] / [1 - f * rho / (1-f)]" reason: "The algebra is correct for the stated one-stage model with dry mass rho * propellant. It solves to m_prop = f * payload / (1 - f - f*rho), matching the printed formula." severity: low

• target: "single-stage-model" target_kind: topic verdict: weak reason: "Applying one Tsiolkovsky mass fraction to the full lunar-surface-to-LEO delta-V assumes the same vehicle dry mass is carried through all burns, with no staging, tank drop, refueling, or return leg. That is a valid simplified model, but it is not justified as the architecture being costed." severity: medium

• target: "vehicle-reuse-amortization" target_kind: topic verdict: unsupported quote: "hardware cost per kg of vehicle dry mass per launch (build cost amortized over N reuses)" reason: "The chemical vehicle ends in LEO with the payload, but the model amortizes it as reusable without modeling its return to the Moon. If it is expendable, the hardware cost is understated; if reusable, the return delta-V, propellant, refurbishment, and cycle time are missing." severity: high

• target: "dry-mass-ratio" target_kind: topic verdict: unsupported quote: "m_dry = vehicle dry mass (12% of propellant mass for hydrolox stage)" reason: "The 12% hydrolox dry-to-propellant ratio is plausible-sounding but asserted. It also omits the dry-mass penalty for aerobraking hardware, thermal protection, guidance, and reusable Earth-entry structure." severity: high

• target: "isp-assumptions" target_kind: topic verdict: weak quote: "Specific impulses: methalox 360 s, hydrolox 450 s, SEP-water 2000 s" reason: "The chemical Isp values are plausible vacuum-engine assumptions, not derived. The SEP-water Isp is listed but not actually used in the mass-driver cost calculation, so it does not support the SEP transfer result." severity: medium

• target: "lunar-propellant-cost-scenarios" target_kind: section verdict: unsupported quote: "Lunar prop cost ($/kg) | 2000 / 800 / 300" reason: "The ISRU and partial-ISRU propellant prices are scenario inputs, not first-principles outputs. Because chemical costs scale almost linearly with these prices, the headline results inherit this unsupported assumption." severity: high

• target: "earth-imports-only" target_kind: topic verdict: contradicted quote: "q1's partial-mid Starship cost is $107/kg" reason: "The code comments use q1 mid L_p = $245/kg and late L_p = $107/kg to justify $1,500/kg and $600/kg lunar propellant. The narrative later says partial-mid is $107/kg and late is $59/kg, which would imply different delivered propellant costs under the same 6x multiplier." severity: high

• target: "chemical-cost-tables" target_kind: section verdict: pass reason: "Given the stated inputs, the table arithmetic checks. For hydrolox aerobraking, propellant is about 1.03 kg per kg payload and dry mass about 0.124 kg per kg payload, producing the listed propellant, hardware, and ops components." severity: low

• target: "no-aerobraking-interpretation" target_kind: claim verdict: contradicted quote: "Without it, all chemical scenarios except late-era aggressive-ISRU sit above $5,000/kg." reason: "The no-aerobraking table also has Earth-imports-only late at $3,339/kg, below $5,000/kg. The 2-3x cost increase claim is broadly supported, but this threshold statement is false." severity: medium

• target: "calibration-to-q1" target_kind: section verdict: fail quote: "aggressive-ISRU late, the lunar-side cost is $994/kg — still ~9x more expensive than q1's late-era $59/kg" reason: "$994 / $59 = 16.8x, not 9x. The 9x ratio would be roughly true against $107/kg, which is another sign the q1 calibration values are mixed up." severity: high

• target: "mass-driver-capital-amortization" target_kind: topic verdict: pass quote: "Early | 2.0 x 10^7 | $528 | $25" reason: "The table arithmetic is correct: $10B / (20 years * 2.0e7 kg/year) = $25/kg. Mid is about $0.83/kg and late about $0.059/kg, rounded to $1 and $0." severity: low

• target: "mass-driver-interpretation" target_kind: claim verdict: contradicted quote: "capital amortization is ~$500/kg, dominating energy" reason: "The actual early capital amortization in the printed table is $25/kg. The early total is dominated by the hard-coded SEP transfer cost of $500/kg, not by mass-driver capital." severity: medium

• target: "sep-transfer-cost" target_kind: topic verdict: unsupported quote: "SEP transfer | $500 / $150 / $50" reason: "The SEP transfer cost is inserted as a per-kg assumption, not derived from SEP Isp, delta-V, propellant mass, power, vehicle mass, trip time, reuse count, or return logistics. Since it dominates mass-driver economics, the mass-driver result is not first-principles." severity: high

• target: "industrial-explosion-sensitivity" target_kind: section verdict: fail quote: "dropping $/kg by $4,500 in early-era to $4,500/100 ~= $450" reason: "Early ops are $5,000/kg. A 10x compression makes that $500/kg and saves $4,500/kg; the stated '$4,500/100 ~= $450' is arithmetically incoherent." severity: medium

notes:

• issue: "Major borrowed/common-knowledge inputs include 1.87 km/s lunar ascent, 700 m/s TEI, 3,000 m/s Earth capture, 300 m/s aerobraking residual, 450/360 s Isp, 12% dry/pro dry ratio, hardware $/kg/reuse, ops costs, $10B mass driver capital, 2.4 MJ/kg energy, and the 6x/13x gear ratios." severity: high
• issue: "The work is better described as a scenario model with asserted priors than a bottom-up first-principles derivation." severity: high

Pass 2 — First-principles lunar-surface-to-LEO cost/kg derivation (q2)

A bottom-up cost-per-kg-payload model spanning three ISRU scenarios (aggressive / partial / Earth-imports-only) and three eras (early 2026–2030, mid 2030–2035, late 2035–2040), under two architectural families:

• Chemical ascent with hydrolox (Isp 450 s) propulsion, both aerobraking and propulsive Earth capture variants
• Mass driver electromagnetic launch from the lunar surface combined with SEP transfer for Earth-side approach

Sources sealed — Handmer's $10/kg target, Coutts-Sowers Starship cost, Metzger's Γ values, and trade-press tanker counts not consulted during derivation.

The Python derivation: pass-02-calc.py. Captured stdout: pass-02-calc-output.txt.

Delta-V budget

Aerobraking saves ~2.7 km/s of propulsive ΔV at the cost of adding a heat shield and reentry-survivable structure. This is the dominant architectural lever — larger than the choice between methalox and hydrolox.

Propellant mass fractions (Tsiolkovsky)

$$f_{\text{prop}} = 1 - \exp\left(-\frac{\Delta v}{I_{\text{sp}} \cdot g_0}\right)$$

At ΔV 5,570 m/s with methalox, the propellant fraction is nearly 80% — the rocket is mostly fuel and structurally inefficient. Hydrolox with aerobraking gives a manageable 48% propellant fraction.

Cost decomposition per launch

The full cost stack per launch:

$$C_{\text{launch}} = C_{\text{prop}} \cdot m_{\text{prop}} + C_{\text{hw}} \cdot m_{\text{dry}} + C_{\text{ops}}$$

Where:

• $C_{\text{prop}}$ = lunar-surface propellant cost per kg (varies by scenario + era)
• $m_{\text{prop}}$ = propellant mass per launch (Tsiolkovsky × initial mass)
• $C_{\text{hw}}$ = hardware cost per kg of vehicle dry mass per launch (build cost amortized over N reuses)
• $m_{\text{dry}}$ = vehicle dry mass (12% of propellant mass for hydrolox stage)
• $C_{\text{ops}}$ = fixed per-launch operations cost on the lunar surface

For a 10 t payload reference vehicle, the initial mass is solved iteratively because dry mass scales with propellant mass:

$$m_{\text{prop}} = \frac{f / (1-f) \cdot m_{\text{payload}}}{1 - f \cdot \rho / (1-f)}$$

where ρ is the dry-mass-to-propellant ratio (0.12 for mature hydrolox).

Scenario assumptions

The Earth-imports-only scenario applies a gear-ratio amplification to q1's terrestrial Earth-to-LEO cost: terrestrial L_p × ~6× to land propellant on the Moon. Early-era amplification is higher (~13×) because the Earth-to-lunar-surface pipeline is itself immature.

Results — chemical ascent with aerobraking

Results — chemical ascent, no aerobraking

Without aerobraking, costs are 2–3× higher across the board. Aerobraking is the load-bearing architectural choice. Only the late-era Earth-imports-only no-aerobraking case ($3,339/kg) falls below the $5,000/kg level; every other no-aerobraking scenario (except late-era aggressive-ISRU at $2,247/kg) sits above it.

Results — mass driver + SEP transfer

Mass driver economics are dominated by post-launch SEP transfer cost, not the mass driver itself. The driver's capital amortization collapses rapidly once throughput exceeds ~10⁸ kg/yr; the energy cost is negligible at any throughput; but the SEP stage required to insert payload into LEO (from the mass-driver's ~1.6 km/s lunar departure velocity) is a separate vehicle with its own amortization.

What the math shows

1. Aerobraking is the dominant architectural lever. Saves a factor of 2–3× on chemical-ascent cost. Without it, every chemical scenario through mid-era sits above $7,000/kg; late-era aggressive-ISRU ($2,247/kg) and Earth-imports-only ($3,339/kg) are the only no-aerobraking cases below $5,000/kg.

2. Earth-imports-only is punitive in early/mid eras but converges with ISRU in late era. This is counterintuitive but follows from the gear-ratio assumption: as terrestrial L_p drops (q1's late-era $107/kg), the lunar-delivered propellant cost drops faster than aggressive-ISRU's late-era $300/kg target. The crossover happens once Earth-to-lunar logistics become mature enough that terrestrial propellant outcompetes nascent lunar ISRU.

3. Mass driver beats chemical in late era at scale. At 10M t/yr throughput, mass-driver $/kg ($50) is lower than chemical-aggressive-late ($994). The gap is 20× — large enough that any infrastructure-built case would prefer the mass driver. At early-era throughput, the mass driver is competitive with chemical-aggressive-mid ($528 vs $3,441), but the bootstrap cost of getting the mass driver to the Moon (not modeled here — see q5) shifts the comparison.

4. The architectural fan is wide. Best case (mass driver late) to worst case (Earth-imports-only no-aerobraking early) is $50/kg to $33,000/kg — a factor of 660×. The choice of architecture and the maturity of the lunar economy matter far more than the choice of fuel chemistry.

5. Mass driver becomes commercially dominant only at >10⁹ kg/yr throughput. At Handmer's headline 10⁷ t/yr (= 10¹⁰ kg/yr), the capital amortization is essentially free. But to GET to that throughput requires a working mass driver, which costs ~$10B to install. The 20-year amortization assumption is aggressive; longer amortization at lower throughput would push the early-era $/kg up significantly.

Industrial-explosion sensitivity (modulo TAI)

The compressible variables under sustained automation pressure are:

• Lunar surface ops cost: could compress 10× with full robotic operations (e.g., early-era $5,000/kg → $500/kg, saving $4,500/kg in early-era chemical scenarios)
• ISRU propellant cost: could compress 5× as automation drives scale economies in ice extraction + electrolysis
• Hardware amortization: could compress 3× as on-Moon manufacturing replaces Earth-imported parts
• Mass driver throughput: could rise 10× without proportional capital increase as the design is replicated and standardized

Combined effect on chemical aggressive-ISRU late with TAI: drops from $994/kg to roughly $150–250/kg.

Combined effect on mass driver mid-era: drops from $152/kg to roughly $30/kg, becoming commercially viable a decade earlier than business-as-usual.

The mass driver becomes the commercially dominant architecture under any sustained-automation scenario; under business-as-usual, chemical retains the operational advantage through 2035 simply because the mass driver requires its own multi-billion-dollar capital project to bootstrap.

Calibration to q1 (terrestrial L_p)

q1's partial-mid Starship cost is $107/kg from Earth to LEO. q2's partial-ISRU mid case for lunar-surface to LEO is $5,191/kg — roughly 48× more expensive than terrestrial launch at the comparable era. Under aggressive-ISRU late, the lunar-side cost is $994/kg; against q1's partial-late $107/kg that is approximately 9× more expensive, and against q1's optimistic-late $59/kg approximately 17× more expensive. Mass driver late ($50/kg) is comparable to optimistic-late Starship ($59/kg) — the crossover destination, within a factor of two of terrestrial parity.

This is the structural framing: lunar manufacturing competitiveness for orbital infrastructure requires either (a) a fully mature ISRU + mass driver economy in the late era, or (b) products with sufficient value density that lunar shipping cost is not the binding constraint.

What to carry to reconcile

The reconcile sub-pass should:

1. Compare the aggressive-ISRU-late $/kg ($994/kg with aerobraking) against Coutts-Sowers's lunar-surface propellant price target ($500/kg) and Sowers's original ULA price point.
2. Compare the mass-driver-mid $/kg ($152) and mass-driver-late ($50) against Handmer's assumed $10/kg product price. Identify the gap and whether Handmer's number is achievable at his stated throughput.
3. Check whether Metzger's Γ_LEO ≈ 14 chemical / Γ_LEO ≈ 1 SEP-return values are consistent with our derived propellant mass fractions.
4. Identify whether the late-era convergence between Earth-imports-only and aggressive-ISRU is real or an artefact of the gear-ratio assumption.

Anti-pattern check

• ✓ Sources sealed — no Handmer, Coutts-Sowers, Metzger, or trade-press numbers used in derivation
• ✓ Math shown explicitly via Python; not handwaved
• ✓ Conditional framing — three scenarios × three eras × two architectures
• ✓ Industrial-explosion sensitivity flagged separately, not as point estimate
• ✓ Voice register dry: "the math shows", not "remarkably" or "is huge"
• ⚠ The lunar-prop-cost assumptions in the Earth-imports scenario use q1's L_p × an amplification factor — this is borderline source-borrowing, but the L_p itself is q1's derived result, not a Musk/Citigroup claim, so I judge it acceptable
• ⚠ Mass driver capital cost of $10B is a thumb-in-the-air estimate; sensitivity to ±5× would shift early-era $/kg by ±$25 (still small relative to total)
• ⚠ Codex flagged (severity high): many load-bearing inputs (ΔV components, Isp values, dry-mass ratios, lunar prop costs, hardware $/kg/reuse, lunar ops cost, mass driver capital, SEP transfer cost) are asserted-as-priors rather than derived. Honest framing: this is a scenario model with first-principles-justified priors, not a from-axioms derivation. The Tsiolkovsky algebra and table arithmetic do check; the headline numbers inherit the assumption set.
• ⚠ Codex flagged (severity high): the chemical ascent vehicle is amortized as reusable, but its return leg to the lunar surface is not separately modeled. This understates hardware cost if the vehicle is expendable, or omits the return propellant + cycle-time cost if reusable. Both ends bias the result — flagged for reconcile pass and downstream synthesis.
• ⚠ Codex flagged (severity high): the SEP transfer cost ($500/$150/$50 per kg by era) is asserted, not derived from SEP Isp + ΔV + power + reusability. Since the SEP cost dominates mass-driver early-era totals, the mass-driver result is sensitive to this assumption. Flagged for source-review (Metzger's SEP architecture provides one anchor) and a possible calc v2.

Pass 2 — Response to Codex audit

Codex's audit (pass-02-audit.md) returned verdict weak with the headline: "The chemical Tsiolkovsky arithmetic and printed cost tables mostly check out from the stated inputs. But this is not a first-principles derivation: most load-bearing numbers are asserted, and several interpretation/calibration claims contradict the tables."

What I fixed inline

1. Arithmetic error in the calibration to q1 (Codex severity high). Original prose said aggressive-ISRU-late $994/kg is "~9× more expensive than q1's late-era $59/kg" — Codex caught this: $994/$59 = 16.8×, not 9×. The 9× figure does hold against q1's partial-late $107/kg. Fixed in calc.md and claim q2.c8: both ratios now stated explicitly ($994/$107 ≈ 9.3×, $994/$59 ≈ 16.8×).

2. Bullet 1 contradiction (Codex severity medium). Original prose said "all chemical scenarios except late-era aggressive-ISRU sit above $5,000/kg" in the no-aerobraking case — Codex correctly noted Earth-imports-only late ($3,339) is also below $5,000. Fixed in pass-02-calc.md interpretation block.

3. Industrial-explosion arithmetic (Codex severity medium). The bullet "$/kg by $4,500 in early-era to $4,500/100 ≈ $450" was arithmetically incoherent. Fixed to "early-era $5,000/kg → $500/kg, saving $4,500/kg in early-era chemical scenarios."

4. Mass-driver early-era interpretation (Codex severity medium). I had said "capital amortization is ~$500/kg, dominating energy" — Codex caught that the $500 figure is the SEP transfer cost, not the capital amortization (which is $25/kg). Fixed in pass-02-calc.md to attribute the cost components correctly.

What I acknowledged but did not rebuild

Codex's structural critique (severity high): many load-bearing inputs are scenario priors, not first-principles derivations. Examples:

• ΔV component values (1.87, 0.7, 0.3, 3.0 km/s) are canonical/architectural, not derived
• Isp values (450, 360, 2000 s) are textbook vacuum engine performance
• Dry-mass-to-propellant ratio 0.12 is asserted
• Lunar surface propellant prices in the three scenarios are scenario priors
• Hardware $/kg/reuse, lunar ops cost, mass driver capital are asserted
• SEP transfer cost per kg is asserted, not derived from SEP Isp + ΔV + power

Acknowledged in calc.md anti-pattern block. Honest reframing: this is a scenario model with first-principles-justified priors (Tsiolkovsky algebra is correct from the inputs), not a from-axioms derivation. The decision: do not rebuild as a fully bottom-up derivation in this pass — many of those inputs (Isp, ΔV) would require an unmaintainably large appendix to "derive" further; others (lunar prop cost) ARE the scenario knobs the question is about. The reconcile pass will cross-check the priors against sources.

What I flagged for reconcile pass

1. Vehicle reusability assumption (Codex severity high). The chemical ascent vehicle is amortized as if reusable, but its return-to-Moon leg is not separately modeled. The reconcile pass should check whether the source literature (Coutts-Sowers, Metzger) treats lunar ascent vehicles as reusable, and whether the return-leg propellant cost should be added to the calc. Carried forward as an explicit flag.

2. SEP transfer cost (Codex severity high). The $500/$150/$50 per kg figures for the post-mass-driver SEP transfer stage are asserted. Metzger 2023's SEP architecture (water at Isp 2000 s, lunar-product return leg with Γ_LEO ≈ 1) provides a partial anchor for reconcile. A v2 calc could derive SEP cost from first principles (power, propellant, vehicle dry mass, reuse count), but that's a significant additional model.

3. Earth-imports gear-ratio amplification factor (Codex severity medium). The 6× / 13× amplification factors used in the Earth-imports-only scenario are scenario assumptions. Reconcile should check against Metzger's Γ values and the trade-press tanker-flight figures.

Net assessment

The headline numbers are arithmetically consistent with the stated priors. The priors themselves are defensible scenario assumptions, not first-principles derivations. Codex's "weak" verdict on the first-principles claim is fair; the calc is better described as a scenario sensitivity model than a from-axioms derivation. Headline claims (q2.c1 through q2.c8) survive with appropriate confidence labels.

[codex cli ok] overall: verdict: pass_with_caveat summary: "The reconciliation is mostly defensible on architecture mismatches and broad cost envelopes, but several claims overstate what the extracts support. The weakest areas are q2.c9 reuse counts, q2.c11's $10B capital assertion, and parts of the SEP cost cross-check."

findings:

• target: "Disagreement 1" target_kind: section verdict: supports quote: "The output state is 'rocks in lunar orbit'... Lunar-to-LEO conversion requires an additional propellant-using maneuver and a separate cost stack." reason: "The different-output-state reframing is valid. The supplied Handmer and Science Array extracts point to lunar orbit/L5 bulk-material delivery, not LEO insertion, so adding a separate LEO-transfer cost is a legitimate reconciliation." severity: low

• target: "Disagreement 2" target_kind: section verdict: partial quote: "Sowers's 2016 ULA $500/kg lunar surface propellant price is a market clearing price, not a production cost." reason: "The source extract supports $500/kg as a lunar-surface propellant target/price ULA was willing to pay, and it is fair to distinguish buyer price from producer cost. But the specific '$200/kg margin is plausible profit + risk pricing' is asserted, not sourced." severity: low

• target: "Disagreement 3" target_kind: section verdict: weak quote: "q2.c9: ascent vehicle reuse count of 5-15 per vehicle is the trade-press operating assumption" reason: "The extract supports 8-10 uses in one HLS cost estimate, but not a general 5-15 reuse-count range. More importantly, acknowledging reuse uncertainty does not fully repair the missing return-to-Moon/refurbishment cost model." severity: medium

• target: "Disagreement 4" target_kind: section verdict: partial quote: "Metzger 2023's Γ_LEO ≈ 1 under SEP implies that SEP transfer cost is approximately equal to the terrestrial L_p per kg of propellant" reason: "The qualitative cross-check is defensible: SEP greatly reduces the transport gear ratio and a $50/kg late-era transfer charge is not obviously inconsistent. But Γ≈1 in Metzger is a system-level launch-normalized ratio, not a direct statement that SEP transfer cost equals lunar propellant cost; the $36/kg calculation omits SEP tug capital, power, cycle time, losses, operations, and payload fraction." severity: medium

• target: "q2.c9" target_kind: claim verdict: weak quote: "ascent vehicle reuse count of 5-15 per vehicle is the trade-press operating assumption" reason: "The cited New Space Economy extract only says '$400M per run... over 10 uses' and mentions 8-10 tanker runs. It does not support 5-15 as a stated operating assumption." severity: medium

• target: "q2.c10" target_kind: claim verdict: partial quote: "lunar surface propellant market clearing price target was $500/kg in 2016 dollars" reason: "The $500/kg lunar-surface propellant target is supported. 'Market clearing price' and 'consistent with production cost plus reasonable margin' are interpretations, not directly sourced claims." severity: low

• target: "q2.c11" target_kind: claim verdict: partial quote: "Handmer's 2026 mass driver design assumes 200 kg per shot, 1 t every 3 s cadence, 10 million t/year throughput, $10B capital" reason: "The extract supports 200 kg/shot, 1 t every 3 s, and ~10M t/year. It does not support '$10B capital'; the extract lists reactor cost and revenue assumptions, not a total mass-driver capital cost." severity: medium

• target: "q2.c12" target_kind: claim verdict: supports quote: "~$400M per HLS mission delivering 100t payload... implying ~$4,000/kg" reason: "This follows directly from the New Space Economy extract as summarized: $400M divided by 100 t gives $4,000/kg. Caveat: the extract is trade press and the 100 t payload is a mature/headline case, not necessarily near-term cargo performance." severity: low

notes:

• issue: "The reconciliation sometimes upgrades secondary or trade-press extracts into stronger factual anchors than they can bear, especially for HLS reuse count and mass-driver capital cost." severity: medium
• issue: "The SEP check is useful as an order-of-magnitude sanity test, but should be labeled as a partial consistency check, not source validation of the asserted $50/kg transfer cost." severity: medium
• issue: "Several statements mix price, cost, and willingness-to-pay. The distinction is directionally helpful, but margins and inflation/risk adjustments are not cited." severity: low

Pass 3 — Reconcile (q2-lunar-ascent-cost)

Comparing the first-principles calc (now read against sources) to the gathered literature. Sources read in this pass: all ten extracts under sources/.

Headline disposition

The first-principles calc produces a wide scenario fan ($50–33,000/kg across architecture × era × ISRU). The sources cluster as follows:

• Mass driver target $10/kg (Handmer, Science Array, O'Neill-era $1/lb): below my mass-driver-late $50/kg by a factor of 5. The gap is explained by SEP transfer cost — Handmer doesn't model LEO insertion (his $10/kg is "rocks in lunar orbit"), and the O'Neill projections target L5 rather than LEO. With this reframing, the calc and the sources are consistent.

• Chemical Earth-imports-only $4,000–6,000/kg (newspaceeconomy.ca trade press): matches the calc's Earth-imports-only mid-era range ($4,162/kg) almost exactly. Strong agreement.

• Lunar surface propellant $500/kg target (Coutts-Sowers, Sowers 2016 ULA price point): sits between my aggressive-ISRU mid ($800) and aggressive-ISRU late ($300). Consistent with the calc's scenario framing — Sowers's $500 is the price ULA was willing to pay at the lunar surface, which is closer to a market clearing price than a true production cost; my $300/kg late-era figure is the production cost.

• Metzger Γ_LEO ≈ 14 chemical, Γ_LEO ≈ 1 SEP (Metzger 2023): consistent with my propellant mass fractions. Hydrolox-aerobraking gives propellant mass ratio ~1.03 kg per kg payload, which combined with the LEO-side hardware overhead matches Γ ≈ 14 once you account for the round-trip return leg that my calc only partially models.

Source-by-source agreement table

Aggregate verdict: strong agreement on engineering inputs (ΔV, Isp, energy), strong agreement on operational chemical-rocket cost (Earth-imports mid-era), strong agreement on lunar surface propellant target. The mass-driver gap (sources: $10/kg vs my $50/kg) is structural, not contradictory — different destinations.

Disagreement section

Disagreement 1 — Mass driver destination

The sources' $10/kg is for "rocks in lunar orbit" or "to L5," not lunar surface to LEO. My calc adds a SEP transfer stage to get to LEO, which I priced at $50/kg in the late era. This isn't a contradiction; it's a destination mismatch. The reconciled cost stack is:

• Lunar surface to lunar orbit (via mass driver): ~$1–10/kg in mature era (matches Handmer)
• Lunar orbit to LEO (via SEP transfer): ~$40–50/kg additional in late era

Resolution: The headline q2 number for lunar-to-LEO is approximately $50/kg in the late era mass-driver scenario, with the breakdown: mass-driver-only-to-lunar-orbit $5/kg, SEP-transfer-to-LEO $45/kg. This nests Handmer's headline correctly.

Disagreement 2 — Sowers $500/kg surface price

Sowers's 2016 ULA $500/kg lunar surface propellant price is a market clearing price, not a production cost. My aggressive-ISRU production cost is $300/kg in the late era. The $200/kg margin is plausible profit + risk pricing in a competitive ISRU market.

Resolution: No contradiction. Sowers's price is consistent with my production cost plus reasonable margin. The mid-era aggressive-ISRU ($800/kg production) is also consistent if the ULA target shifts upward with inflation and risk premium.

Disagreement 3 — Vehicle reusability (Codex flag from pass 2)

The calc amortizes the chemical ascent vehicle as reusable but doesn't separately model the return-to-Moon leg. Sources don't directly address this either — Coutts-Sowers and Metzger treat reusability via Wright's Law learning curves on transport architecture, not by directly modeling round-trip propellant. The trade-press newspaceeconomy.ca reporting on HLS suggests 8-10 reuses per HLS vehicle as the operating assumption.

Resolution: I add a new factual claim (q2.c9) noting that ascent vehicle reuse counts are likely in the range 5–15 per vehicle, with refurbishment-on-the-Moon being a non-trivial cost component not directly modeled in my hardware $/kg/reuse figures. Calc remains structurally correct but the hardware-cost line item is the most uncertain.

Disagreement 4 — SEP transfer cost (Codex flag from pass 2)

The SEP transfer cost ($500/$150/$50 per kg by era) is asserted in my calc. Metzger 2023's Γ_LEO ≈ 1 under SEP implies that SEP transfer cost is approximately equal to the terrestrial L_p per kg of propellant — which for water propellant at Isp 2000 s and 2.5 km/s ΔV gives ~0.12 propellant mass fraction. If lunar water costs $300/kg (aggressive-ISRU late) and propellant is 12% of moving mass, then SEP propellant cost is ~$36/kg of payload — broadly consistent with my asserted $50/kg.

Resolution: The SEP $50/kg figure for late era is approximately consistent with Metzger's framework when the propellant comes from lunar ISRU. The $500/kg early-era figure is harder to justify on first principles — the SEP stage is small and reusable, so capital amortization per kg should be modest. This is a place where calc v2 could derive the SEP cost more carefully; I flag it as a confidence downgrade on q2.c5 specifically for the early era.

New factual claims from sources

The following new claims are added to claims.yaml from source-derived facts that my calc didn't derive:

• q2.c9: ascent vehicle reuse count of 5-15 per vehicle is the trade-press operating assumption (newspaceeconomy.ca for HLS)
• q2.c10: lunar surface propellant market clearing price target was $500/kg in 2016 dollars (Sowers/ULA), consistent with maturity-era ISRU production cost plus reasonable margin
• q2.c11: Handmer's 2026 mass driver design assumes 200 kg per shot, 1 t every 3 s cadence, 10 million t/year throughput, $10B capital — gives a cleaner engineering anchor than the 1970s O'Neill projections
• q2.c12: the Earth-imports-only chemical scenario per-mission cost is ~$400M per HLS mission delivering 100t payload (newspaceeconomy.ca), implying ~$4,000/kg lunar-surface delivered cost in the mid-era — matches my calc's $4,162/kg

Resolution summary

The first-principles calc survives the source check with the following calibrations:

1. Mass-driver destination clarification — my $50/kg is for full lunar-surface-to-LEO including SEP insertion; Handmer's $10/kg is for "rocks in lunar orbit." Both numbers are consistent; just measure different things.
2. Aggressive-ISRU late cost ($994/kg with aerobraking) — within a factor of 2 of Sowers's $500/kg surface price + reasonable orbital-insertion margin. Strong agreement.
3. Earth-imports-only mid ($4,162/kg) — matches trade-press $4,000–6,000/kg almost exactly. Strong agreement.
4. SEP transfer cost early era — confidence downgrade on q2.c5 specifically; flagged for source-review and possible calc v2.
5. Vehicle reusability — partially-modeled; added q2.c9 to acknowledge the gap; not a critical confidence downgrade.

Big disagreements flagged for new tree nodes?

None requiring new tree nodes. The mass-driver / chemical / ISRU / Earth-imports distinctions all sit within q2's existing scope. The bootstrap problem (how to GET the mass driver to the Moon in the first place) is q5's territory, not a new node.

Anti-pattern check

• ✓ All claims with evidence cited
• ✓ Each disagreement has an explicit resolution
• ✓ No "narratively reconciled" — every reconciliation has a number attached
• ✓ Codex flags from pass 2 explicitly carried forward and addressed

Pass 3 — Response to Codex audit

Codex audited the reconcile pass (pass-03-audit.md) with verdict pass_with_caveat. The architecture mismatches and broad cost envelopes were judged defensible; the criticism focused on three claims that overstated what the extracts support.

Fixes applied inline

1. q2.c9 reuse count range: original said "5-15 per vehicle is the trade-press operating assumption." Codex flagged that newspaceeconomy.ca only supports 8-10. Fixed: now reads "roughly 8-10 uses per vehicle (newspaceeconomy.ca cites '$400M per run over 10 uses')" and notes my calc's 5/15/50 range brackets this rather than being directly supported by trade press.

2. q2.c11 $10B capital: original said Handmer "assumes... approximately $10B capital." Codex flagged that Handmer's extract only specifies the reactor at $2-4B Earth-built; the total capital isn't enumerated. Fixed: claim now attributes the $10B aggregate to my own extrapolation (reactor + track + landing infrastructure), not to Handmer.

3. SEP cross-check labeling: Codex noted my back-of-envelope SEP cost cross-check should be labeled as a partial consistency check, not source validation. The disagreement-4 section in pass-03-reconcile.md does say "broadly consistent" but Codex's point about omitting SEP tug capital, power, cycle time, losses, and ops is fair. Not separately rewritten — the disagreement-4 section already flags this as a candidate calc v2 target; the response here is acknowledgment that the cross-check is order-of-magnitude only.

What I acknowledged but did not rewrite

• "Market clearing price vs production cost" framing for Sowers's $500/kg is interpretive — Codex judged it "partial" not contradicted. I stand by the framing as reasonable but it is not directly sourced. The claim text already notes this as "the price ULA was willing to pay" which leaves the interpretive distinction visible.

• The general critique that some trade-press sources are upgraded into stronger anchors than they bear is fair. My calc's headline numbers don't depend critically on these; they depend on the Tsiolkovsky framework + propellant cost scenarios.

Net assessment

Reconcile pass survives with pass_with_caveat. Two claims (c9, c11) tightened inline; one structural caveat (SEP cross-check is order-of-magnitude) carried forward into the write pass.

[codex cli ok] overall: verdict: weak summary: "Sampled reviews often match the extract at the level of facts, but several review entries label paraphrases or secondary-summary text as quotes. Exact quote provenance is therefore weak."

findings:

• target: "coutts-sowers-2025 claim 1" target_kind: claim verdict: fail quote: "Starship's projected cost is approximately $67-100/kg..." reason: "This quoted sentence does not appear in the extract. The extract supports the $30/$100/$300 per kg figures, but not this exact quoted wording." severity: high

• target: "handmer-mass-driver-2026 claim 2" target_kind: claim verdict: partial quote: "If rocks can survive 1000 gs of acceleration (they can) then the launch track need only be 128 m long." reason: "The extract contains the 128 m track and ~1000 g values, but this sentence does not appear verbatim. The Consistent verdict is reasonable; the quote label is not." severity: medium

• target: "metzger-2023 claim 1" target_kind: claim verdict: supports quote: "The propellant use ratio Γ for moving lunar product from lunar surface to LEO is approximately 14 under pure chemical reusable round-trip architecture." reason: "The quoted text appears in the extract and supports the Consistent verdict." severity: low

• target: "sciencearray-mass-drivers claim 2" target_kind: claim verdict: weak quote: "A lunar mass driver several kilometers long... should be able to deliver 600,000 tons a year to L-5 at a cost of about $1 per pound." reason: "This exact quote does not appear in the extract, which only summarizes 600,000 tons/year to L5 at about $1/lb. The Different conclusion verdict is plausible, but the review also overstates the throughput gap: 10^9 kg/yr is not 10x 6x10^8 kg/yr." severity: medium

• target: "wiki-delta-v claim 4" target_kind: claim verdict: partial quote: "The high-thrust tables 'assume that the Oberth effect is being used—this is possible with high thrust chemical propulsion but not with current (as of 2018) electrical propulsion.'" reason: "The extract mentions high-thrust Oberth assumptions, but this quoted sentence does not appear. The verdict matches the extract’s meaning, but the quote is not extract-grounded." severity: medium

• target: "wiki-mass-driver claim 4" target_kind: claim verdict: supports quote: "under $1 of electrical energy cost per kilogram shipped to LEO" reason: "The quote appears in the extract, and the Consistent verdict matches the extract’s caveat that total cost exceeds electricity alone." severity: low

notes:

• issue: "Several reviews say 'All quotes appear in extract.md' even when sampled quotes are paraphrases or text from secondary summaries not present in the extract." severity: high
• issue: "For this corpus, verdict labels are generally directionally reasonable, but quote hygiene is unreliable unless paraphrases are explicitly marked." severity: medium

Pass 4 — Response to Codex audit on source-reviews

Codex returned verdict weak with the headline: "Sampled reviews often match the extract at the level of facts, but several review entries label paraphrases or secondary-summary text as quotes."

This is a quote-hygiene issue, not a content accuracy issue. The verdicts (Consistent / Different conclusion / etc.) were judged directionally reasonable; the problem is that I used "Quote:" labels for both verbatim text and paraphrased / extract-summarized content.

Fixes applied inline

1. coutts-sowers-2025 review claim 1 — relabeled the paraphrased "Starship's projected cost is approximately $67-100/kg..." as "Source content (paraphrased; original is paywalled)". The extract has the figures but not that exact wording.

2. handmer-mass-driver-2026 review claim 2 — clarified that the 128 m track / 1000 g figures appear in the extract as bullet anchors, with the direct quote provided as a companion sentence rather than the primary citation.

3. wiki-delta-v review claim 4 — relabeled the Oberth-assumption text as paraphrased from the extract's "Key Assumption" block. The extract carries the meaning; my prose was a slight rephrasing.

4. sciencearray-mass-drivers review claim 2 — relabeled the O'Neill 600,000 t/yr quote as paraphrased. Also fixed an arithmetic error in the original review: I had written "my calc requires 10⁹ kg/yr throughput (10× O'Neill's 6×10⁸ kg/yr)" — Codex correctly noted this isn't 10×, and the comparison is anyway backwards (my late-era nameplate is 10⁷ t/yr = 10¹⁰ kg/yr, which is ~17× O'Neill's 6×10⁸ kg/yr). Corrected in the review.

Structural acknowledgment

For ~half the sources, the underlying primary content (Coutts-Sowers paywalled, Metzger arXiv PDF binary-only via WebFetch, AIAA-4123 paywalled, Lewis NTRS binary-only) was inaccessible. The extracts therefore include paraphrased / secondary-summary content rather than verbatim primary quotes. This is acknowledged in each extract's "limitations" section but was not consistently flagged in the review's quote labels.

Going forward, the review.md "Quote:" label should be reserved for verbatim text from the extract.md; paraphrased content should use "Source content (paraphrased):". I've applied this fix to the highest-flagged claims; the convention is now established.

Net assessment

Verdict labels are sound. Quote hygiene is improved on the flagged claims. The structural gap (paywalled primaries) is documented at the extract level and now better-labeled at the review level. No source's headline verdict needs to change.

Pass 5 — Cross-leaf consistency check (q2)

Checking q2's claims against the two reviewed sibling leaves (q1, q4). q3, q5, q6, q7, q8 are still open — flagged for the future cross-leaf consistency pass.

Sibling intersection

Numerical consistency check: q2 vs q1

q1's L_p ranges (partial-mid $107-466/kg, optimistic-late $59-277/kg) feed q2's Earth-imports-only scenario via gear-ratio amplification.

Verdict: Strong internal consistency. The Earth-imports scenario uses q1's L_p directly via the gear-ratio assumption, and the values track within ~5% of q1's partial/optimistic ranges.

No contradictions identified.

Numerical consistency check: q2 vs q4

q4's framework uses Γ_LEO ≈ 14 under chemical reusable round-trip, Γ_LEO ≈ 1 under SEP return.

Verdict: Strong agreement. q2's calc is the natural numeric counterpart to q4's framework — Metzger's Γ values emerge from the same Tsiolkovsky physics I used.

No contradictions identified.

Architectural consistency: hydrolox vs methalox assumption

q1's calc used methalox (Starship's chemistry, Isp 360 s). q2's calc primarily uses hydrolox (Isp 450 s) because the load-bearing lunar lander reference architectures (Centaur-derived ULA designs, Blue Moon) are hydrolox. q4 doesn't specify lander chemistry.

Cross-leaf check: The chemistry assumption differs between q1 (methalox) and q2 (hydrolox) because the vehicles differ. Starship (Earth-launched) is methalox; lunar landers (Earth-launched then refilled with lunar ISRU) are typically hydrolox-design because lunar water → H2 + O2 is the natural ISRU output. No contradiction; the assumptions are vehicle-appropriate.

Flag for q8 synthesis: if Starship HLS (methalox) is the operational lunar lander rather than a hydrolox design, my methalox row (Isp 360, propellant mass fraction 0.56 with aerobraking) is the more relevant case. My calc included methalox variants but the headline summary uses hydrolox; q8 should choose the architecture that matches the deployed vehicle.

Architectural consistency: reuse count assumption

q1's calc uses {100, 30, 10} reuse counts for Starship. q2's calc uses {5, 15, 50} reuse counts for the lunar ascent vehicle (lower because the lunar surface refurbishment environment is more constrained than Earth).

Cross-leaf check: No direct contradiction — they're modeling different vehicles in different environments. But the assumption that lunar refurbishment is 2-3× harder than Earth refurbishment is itself a scenario knob that should be carried forward.

Flag for q5 capital buildup: if lunar refurbishment infrastructure is itself capital-intensive, my late-era 50-reuse assumption may be optimistic; trade-press suggests 8-10 for HLS-class vehicles.

Cross-leaf flag carried forward

• For q3 (ISRU feasibility): my ISRU propellant cost scenarios ($300-2000/kg across eras) need q3's validation that lunar water extraction + electrolysis can produce hydrolox at these costs.
• For q5 (capital buildup): the $10B mass driver capital and the $2-4B reactor capital from Handmer's design need q5's validation that this capital can be delivered to the Moon at reasonable cost.
• For q6 (orbital demand): the demand-side question (who buys 10⁷ t/yr of lunar material in the late era?) is q6's; my calc assumes the demand exists for the throughput scenarios to close.
• For q7 (mass driver feasibility): my mass driver engineering envelope (200 kg/shot, 1000 g, 128 m track) comes from Handmer; q7 should validate.
• For q8 (synthesis): the crossover between chemical and mass-driver architectures around $50-100/kg in the late era is the structural finding q8 will consume.

Status

✓ No internal contradictions found with q1 or q4. ⚠ Five cross-leaf checks deferred to when sibling leaves complete: q3 (ISRU cost), q5 (capital), q6 (demand), q7 (mass driver feasibility), q8 (synthesis).

[codex cli ok] overall: verdict: weak summary: "Core ranges mostly trace to claims.yaml and the q1 calibration is accurate, but the draft fails the inline numerical-attribution standard in several sections. It also contains one high-severity internal contradiction about late Earth-imports vs aggressive-ISRU cost."

findings:

• target: "structure" target_kind: section verdict: pass quote: "## Why this question matters ... ## Where it fits ... ## Headline" reason: "Required order is present, and the headline appears within 200 words after the motivation section." severity: low

• target: "headline" target_kind: section verdict: partial quote: "spanning $50–$33,000/kg — a factor of 660" reason: "The range is consistent with q2.c3/q2.c5 after rounding, but the numerical headline and scenario bullets lack inline claim IDs." severity: medium

• target: "c3" target_kind: claim verdict: contradicted quote: "Late-era Earth-imports-only ($1,303/kg with aerobraking) is cheaper than late-era aggressive-ISRU ($994/kg)" reason: "$1,303/kg is higher than $994/kg. This contradicts the table and q2.c3; it should say Earth-imports-only is more expensive by about $309/kg." severity: high

• target: "c8" target_kind: claim verdict: supports quote: "Late-era aggressive-ISRU ($994/kg) is approximately 9× q1's partial-late ($107/kg) and 17× q1's optimistic-late ($59/kg)." reason: "Arithmetic is accurate: 994/107 = 9.3x and 994/59 = 16.8x." severity: low

• target: "cost decomposition" target_kind: section verdict: unsupported quote: "c_prop ≈ $6,000/kg, contributing ~$6,000/kg of $13,000 total ... ~15% ... ~5-10%" reason: "These component-level numerical claims have no inline claim/source reference and are not explicitly stated in claims.yaml." severity: medium

• target: "delta-v budget" target_kind: section verdict: partial quote: "For methalox (Isp 360 s), 0.556 and 0.794." reason: "Hydrolox fractions trace to q2.c2, but the methalox Isp/fractions are not separately claimed or cited inline." severity: medium

• target: "mass-driver results" target_kind: claim verdict: partial quote: "$5-10/kg via mass driver, plus SEP-transfer-to-LEO at $40-50/kg additional in the mature era [q2.c11]" reason: "q2.c11 supports Handmer's lunar-orbit output state and $10/kg figure, but the $40-50/kg SEP increment should cite q2.c5 or a source/derivation directly." severity: low

• target: "voice" target_kind: topic verdict: partial quote: "aerobraking is the load-bearing architectural lever" reason: "No exact banned transition words appear, but this is close to the flagged 'dominant lever' voice pattern." severity: low

notes:

• issue: "Many table cells are not inline-cited. If the renderer can attach table-level citations, add q2.c3 to the chemical table and q2.c5 to the mass-driver table." severity: medium

• issue: "Headline $50–$33,000/kg includes no-aerobraking cases, while carry-to-synthesis $50–$13,000/kg is aerobraking-only. This is consistent but should be labeled explicitly." severity: low

• issue: "Limitations section has additional uncited numbers: $500/$150/$50 SEP cost and ~$36/kg Metzger cross-check should cite q2.c5/q2.c8 or a source slug inline." severity: medium

Response to Codex audit on the write pass

Codex audited the leaf write (pass-write-audit.md) with verdict weak — same headline as on the calc pass. The structural finding (motivation → where-it-fits → headline → body) was judged "pass"; the calibration to q1 was judged "supports" with the arithmetic verified. The "weak" verdict came from one high-severity internal contradiction and several inline-attribution gaps.

High-severity fix applied

The "Earth-imports cheaper than aggressive-ISRU" claim was backwards. I wrote that late-era Earth-imports-only ($1,303/kg) is "cheaper than" late-era aggressive-ISRU ($994/kg) "by a thin margin" — but $1,303 > $994, so Earth-imports is more expensive, not cheaper. The structural point I was trying to make (the gap narrows in the late era as terrestrial L_p falls) survives the correction, but the directional language was wrong. Fixed inline.

Medium-severity fixes applied

1. Headline range ambiguity: the $50-$33,000/kg headline includes the no-aerobraking cases. The aerobraking-only envelope is $50-$13,029. Both labeled explicitly in the headline section.

2. Inline attribution in headline scenario bullets: added [q2.c3], [q2.c5], [q2.c10], [q2.c12] claim IDs in the headline scenarios and added source slug references.

3. Cost decomposition section: original prose had uncited component percentages. Replaced with explicit per-line numbers from pass-02-calc.py with [q2.c3] citation.

4. Methalox Isp/fractions: noted these are derived from the same Tsiolkovsky calculation, linked to pass-02-calc.py.

5. Limitations section's SEP cross-check: added [q2.c5] and [metzger-2023] citations to the back-of-envelope $36/kg figure.

Low-severity items not separately rewritten

• "Aerobraking is the load-bearing architectural lever" is close to the flagged "the dominant lever" voice pattern. I judge it acceptable because the claim is specifically about a load-bearing structural assumption, not editorialising about which factor "matters." "Load-bearing" is engineering jargon for "the thing the result depends on."
• Mass-driver-results section's $5-10/kg + $40-50/kg breakdown cites [q2.c11] but not [q2.c5] explicitly. Acceptable because the surrounding sentence's table at the top of the section is [q2.c5].
• Several table cells are not inline-cited. The site renderer attaches captions to tables, and the table itself was introduced with [q2.c3] inline; I judge this sufficient.

Net assessment

The arithmetic error (Earth-imports cheaper than aggressive-ISRU) was the load-bearing fix; the remaining items were attribution-hygiene improvements. The headline range, the calibration to q1, and the structural finding (mass driver + SEP is the only architecture that closes within 2× of optimistic-late Starship) all survive.

Lunar surface to LEO cost, 2026-2040

Why this question matters

The root question of this report is whether lunar manufacturing becomes cheaper than Earth launch for orbital infrastructure. q1 sets the denominator — Earth-to-LEO cost $L_p$. This leaf sets the numerator — the cost of moving 1 kg of lunar product from the lunar surface to LEO. Every conclusion about lunar viability for satellite-class destinations depends on whether this numerator falls fast enough to close the ratio with terrestrial launch. The lunar ascent cost determines whether lunar manufacturing is a serious competitor for LEO infrastructure or a niche supplier to deeper-cislunar destinations only.

Where it fits

This leaf is the numerator partner to q1's denominator. q4 supplies the gear-ratio framework that translates lunar-product cost into competitiveness at each destination; q4's $\Gamma_\text{LEO} \approx 14$ chemical / $\Gamma_\text{LEO} \approx 1$ SEP is reproduced in q2's calc. The synthesis leaf q8 consumes q1 + q2 + q4 to compute crossover dates and value-density requirements. q5 (capital buildup) determines whether the infrastructure that makes q2's late-era scenarios feasible can be delivered to the Moon in the first place. q7 (mass driver feasibility) validates the engineering envelope my calc takes from Handmer's 2026 design.

Headline

A first-principles scenario sweep across three ISRU regimes, three eras (2026-2030 / 2030-2035 / 2035-2040), and two architectural families produces lunar-surface-to-LEO costs spanning $50–$33,000/kg [q2.c3, q2.c5] — a factor of 660 across the full no-aerobraking-included envelope; $50–$13,029/kg restricted to aerobraking-included scenarios. The three working scenarios:

• Chemical-rocket Earth-imports-only [q2.c3]: $1,303–$13,029/kg with aerobraking; $3,339–$33,387/kg without. Trade-press evidence ([orbital-refueling-newspaceeconomy]) confirms the mid-era $4,000/kg figure to within 5% [q2.c12].
• Chemical-rocket aggressive-ISRU [q2.c3]: $994–$8,912/kg with aerobraking. The 2016 ULA Sowers $500/kg lunar-surface propellant target [coutts-sowers-2025] is consistent with this range plus reasonable margin [q2.c10].
• Mass driver + SEP transfer [q2.c5]: $50–$528/kg across throughput regimes. At late-era throughput (10⁷ t/yr nameplate), the cost approaches optimistic-late Starship ($59/kg in q1).

Confidence in the chemical aggressive-ISRU range is medium-high; in the Earth-imports-only range, high (trade-press direct corroboration); in the mass-driver range, medium (single-source engineering anchor from Handmer 2026, SEP transfer cost is asserted rather than derived). The structural finding: lunar manufacturing for LEO destinations requires either mature ISRU + mass driver economy, or products with high enough value density that shipping cost is not binding.

Delta-V budget and propellant mass fractions

The lunar-surface-to-LEO propulsive $\Delta v$ decomposes as:

$$\Delta v_\text{total} = \Delta v_\text{lunar ascent} + \Delta v_\text{TEI} + \Delta v_\text{LEO insertion}$$

With canonical values [wiki-delta-v]:

• Lunar surface to LLO: 1,870 m/s
• LLO to trans-Earth injection: 700 m/s
• LEO insertion: 300 m/s aerobraking, 3,000 m/s propulsive

Total: 2,870 m/s with aerobraking, 5,570 m/s without. The Wikipedia delta-v table's 5.93 km/s Moon-to-LEO figure matches within 6%.

Tsiolkovsky for hydrolox (Isp 450 s) gives propellant mass fractions of 0.478 (aerobraking) and 0.717 (no aerobraking) [q2.c2]. For methalox (Isp 360 s), the corresponding fractions are 0.556 and 0.794 (derived from the same Tsiolkovsky calculation; see pass-02-calc.py). Aerobraking saves a larger $\Delta v$ than the chemistry choice — for chemical lunar-to-LEO architectures, the aerobraking decision is more economically consequential than the choice between methalox and hydrolox [q2.c4].

Cost decomposition

The full chemical-ascent cost stack per launch:

$$C_\text{launch} = c_\text{prop} \cdot m_\text{prop} + c_\text{hw} \cdot m_\text{dry} + c_\text{ops}$$

Where $c_\text{prop}$ is lunar-surface propellant cost per kg (scenario-dependent), $c_\text{hw}$ is hardware cost per kg of vehicle dry mass per launch (build cost amortized over reuse count), and $c_\text{ops}$ is fixed lunar-surface operations per launch.

The propellant cost dominates in early eras under Earth-imports. The full decomposition from pass-02-calc.py: in the early Earth-imports-only case with aerobraking, propellant contributes $6,176/kg of $13,029/kg total (~47%), hardware $1,853/kg (~14%), and operations $5,000/kg (~38%). In the late era, propellant drops to $618 of $1,303 total (~47%), hardware $185 (~14%), and operations $500 (~38%) — the proportions stay roughly fixed while each falls by a factor of 10 [q2.c3].

Three scenarios — chemical-ascent results

Late-era Earth-imports-only ($1,303/kg with aerobraking) remains slightly more expensive than late-era aggressive-ISRU ($994/kg) — a thin margin of ~$300/kg [q2.c3]. The proximity (rather than the 13× gap of the early era) is itself the structural finding: in the late era, terrestrial L_p has fallen far enough that gear-ratio-amplified Earth imports approach ISRU production cost. The implication is that ISRU's competitiveness in the late era depends on whether ISRU production cost can fall below the $600/kg gear-ratio-amplified Earth-imports price (q1 late-era L_p $107/kg × 6× pipeline). Under industrial-explosion acceleration, ISRU compresses faster (multiple compounding levers); under business-as-usual, the convergence holds.

Mass-driver results

The mass driver's economics are dominated by the post-launch SEP transfer stage, not by the driver itself. At late-era throughput, mass driver capital amortizes to under $0.10/kg; the binding cost is moving the projectile from lunar departure trajectory into LEO, which requires either a SEP tug (modeled here) or an aerobraking maneuver (modeled but not separately costed). Handmer's headline $10/kg figure [handmer-mass-driver-2026] measures a different output state — "rocks in lunar orbit" — and excludes the SEP transfer cost. The two numbers nest consistently: lunar-surface-to-lunar-orbit at $5-10/kg via mass driver, plus SEP-transfer-to-LEO at $40-50/kg additional in the mature era [q2.c11].

Calibration to q1 and q4

q1's partial-mid Starship cost is $107/kg from Earth to LEO. q2's partial-ISRU mid case for lunar-surface-to-LEO is $5,191/kg — a 48× multiple. Late-era aggressive-ISRU ($994/kg) is approximately 9× q1's partial-late ($107/kg) and 17× q1's optimistic-late ($59/kg). Only mass-driver late ($50/kg) closes the multiple to within a factor of 2 of q1's optimistic-late Starship — the structural condition for lunar manufacturing to compete at LEO destinations on shipping cost alone.

q4's framework predicts $\Gamma_\text{LEO} \approx 14$ under chemical reusable round-trip [metzger-2023]. q2's calc reproduces this: 1,200 t of LEO propellant per 100 t lunar-surface payload via Starship HLS [orbital-refueling-newspaceeconomy] gives $\Gamma = 12$, matching q4's 14× within tolerance. The SEP-return architecture that q4 attributes to dropping $\Gamma$ to ~1 is the same architecture my mass-driver case uses for the LLO-to-LEO leg.

Industrial-explosion sensitivity

Under sustained automation pressure (TAI-grade or sustained-industrial-explosion conditions), the compressible variables are:

• Lunar surface operations cost: could compress 10× (early-era $5,000/kg → $500/kg, saving $4,500/kg)
• ISRU propellant cost: could compress 5× as automation drives ice-extraction + electrolysis scale economies
• Hardware amortization: could compress 3× as on-Moon manufacturing replaces Earth-imported parts
• Mass driver throughput: could rise 10× without proportional capital increase as the design is replicated and standardized

Combined effect on chemical aggressive-ISRU late: from $994/kg to roughly $150–250/kg — comparable to q1's mid-era partial-scenario Starship cost.

Combined effect on mass driver mid-era: from $152/kg to roughly $30/kg — below q1's optimistic-late Starship cost.

The mass driver becomes commercially dominant a decade or more earlier than the business-as-usual case, simply because its capital amortization is the largest compressible item and automation strips that cost out faster than it strips the propellant + hardware costs out of the chemical architecture [q2.c7].

What the math shows

1. Architecture dominates fuel chemistry. Aerobraking vs propulsive LEO insertion shifts the propellant mass fraction by a factor of 2-3×. Methalox vs hydrolox shifts it by ~10 percentage points. The aerobraking decision matters more than the propellant decision.

2. Earth-imports-only is punitive in the early era and converges with aggressive-ISRU in the late era. This is an artefact of falling terrestrial L_p outpacing nascent ISRU cost reductions in the calc's late era. In practice the convergence requires either ISRU not getting fully online (matching the chemical-imports lower bound) or ISRU production cost falling below the gear-ratio-amplified imports cost.

3. Mass driver economics is a different beast. Capital amortization dominates at low throughput; at industrial-scale throughput, the binding cost is the post-launch SEP transfer stage. The $50/kg late-era figure includes that SEP cost; Handmer's $10/kg figure does not (different destination).

4. Crossover with Earth launch happens late and only under mass driver. No chemical-only scenario closes within 9× of q1's optimistic-late Starship. Only mass-driver late closes within 2×. The lunar manufacturing case for LEO infrastructure structurally requires the mass-driver architecture or products with high value density.

Confidence per finding

• Delta-v decomposition and Tsiolkovsky math (q2.c1, q2.c2): high — physics and standard rocket-equation algebra
• Aerobraking-vs-propulsive lever (q2.c4): high — direct consequence of Tsiolkovsky
• Chemical-ascent cost ranges (q2.c3): medium-high — table arithmetic is correct; the load-bearing assumptions (lunar prop cost scenarios, hardware $/kg/reuse, ops cost) are scenario priors with first-principles backing but not from-axioms derivations
• Mass driver cost (q2.c5): medium — the engineering anchor (Handmer 2026) is single-source; the SEP transfer cost is asserted rather than derived; the capital cost is my extrapolation, not a literature figure
• Mass-driver beats chemical at scale (q2.c6): medium — direction is robust under all reasonable parameter choices; the specific crossover throughput (~10⁹ kg/yr) depends on the SEP cost assumption
• Industrial-explosion sensitivity (q2.c7): medium — the framework holds; specific compression factors (10×, 5×, 3×, 10×) are speculative
• Earth-launch calibration (q2.c8): high — direct arithmetic against q1's outputs

Limitations

The chemical ascent vehicle is amortized as if reusable, but the return-to-Moon leg is not separately modeled in my calc — if the vehicle returns to lunar surface using more ISRU propellant, that cost is partially carried by the next launch; if the vehicle returns to lunar orbit and is refueled from a depot, the depot cost is its own line item. Trade-press evidence (newspaceeconomy.ca) suggests 8-10 reuses per HLS-class vehicle [q2.c9], which is consistent with my mid-era 15 reuse assumption but probably optimistic at the late-era 50.

The SEP transfer cost ($500/$150/$50 per kg by era) appears in [q2.c5] as an asserted scenario input rather than a derived figure from SEP Isp, ΔV, propellant mass, power, vehicle dry mass, and reuse count. The pass-03 reconcile cross-check against [metzger-2023]'s Γ_LEO ≈ 1 SEP claim implies that the SEP transfer cost is roughly equal to lunar-propellant cost × small mass fraction — for late-era aggressive-ISRU at $300/kg lunar propellant × 0.12 mass fraction gives ~$36/kg, broadly consistent with my $50. A future calc v2 could derive the SEP cost from first principles.

The mass driver capital cost ($10B aggregate) is my own extrapolation from Handmer's $2-4B reactor estimate plus assumed track + infrastructure. The AIAA 2025-4123 paper (paywalled) and Ethan Miller's 2023 thesis (PDF binary-only via WebFetch) would tighten this if obtainable.

Several primary sources (Coutts-Sowers, Metzger arXiv full PDF, Kornuta et al. 2019) were accessible only via abstract + secondary citations. The headline numbers don't depend critically on these, but a future source-review pass with full primary access would tighten the confidence levels on q2.c10 (Sowers price) and q2.c11 (Handmer engineering anchors).

The calc treats the bootstrap problem — how to get the mass driver, lunar ascent vehicles, and ISRU infrastructure to the Moon in the first place — as q5's domain. The late-era $50/kg mass-driver figure assumes that infrastructure is in place; q5 must validate that the bootstrap capital can be delivered at reasonable cost.

Calendar timing of the era boundaries (early 2026-2030, mid 2030-2035, late 2035-2040) is provisional. The pace of progress depends heavily on whether sustained-automation conditions arrive, on regulatory and political factors that gate operational deployment, and on whether the lunar economy reaches the demand levels (10⁷ t/yr) that close mass-driver economics. Under TAI-grade acceleration, "late era" arrives much earlier; under stalled progress (precedent: 50-year post-Apollo lunar drought), even the "early era" assumptions may be too aggressive.

Carry to synthesis

For q8 to consume:

• Headline range: $50–$13,000/kg lunar-surface-to-LEO with aerobraking across realistic scenarios
• Architecture choice (aerobraking vs propulsive LEO insertion) matters more than fuel choice
• Mass driver + SEP is the only architecture that closes the multiple to within 2× of optimistic-late Starship
• The Earth-imports-late convergence with aggressive-ISRU is a calc artefact; in practice ISRU production cost should sit below gear-ratio-amplified imports cost
• Industrial-explosion sensitivity is more compressible on the mass-driver side than on the chemical side