Methodology v0 · 2026-10-31
Every number on this platform comes from the same three layers. This page is the whole model — the formulas, the sources, the dates, and the places where it can be wrong.
notional ≠ realizable
We never present mass × spot price as a value. That number assumes infinite demand: sell the whole asteroid and the platinum market collapses to the cost of production. Realizable value is discounted, market-capacity-constrained, net of mission cost — usually 10³–10⁶× smaller than the headline figure everyone quotes. Where we do quote the undiscounted figure at all, it is labelled notional ceiling, carries that definition on the same line, and is never what a chart or a ranking is sorted by.
Some bodies get no ceiling at all. Water and ice have no terrestrial market at asteroid tonnages — the spot price would be a fiction — so their pages say none and point at the only honest question instead: a realizable in-space propellant scenario. A missing number is a statement, not an omission.
一句话:我们不给"质量 × 现价"的天文数字,只算把货卖出去、扣掉任务成本、按时间折现之后真正能落袋的钱。
1 · Resource layer — what is physically there
Input is what a telescope actually gives you; output is always a range, never a point.
| Step 0 · Diameter | D = 1329 / √p_V × 10^(−0.2H) (Pravec & Harris 2007, arXiv:0706.1140). WISE/NEOWISE measured diameters win; an assumed albedo is flagged, because mass ∝ albedo−1.5. |
|---|---|
| Step 1 · Spectral type | Type → meteorite analog → composition family. Bare X spectra cannot separate M/E/P — a factor-of-many difference in value — so metal/PGM claims require albedo or radar confirmation. |
| Step 2 · Bulk density | Rubble-pile bulk density (Carry 2012, Table 3), not mineral grain density: 40–70% porosity otherwise inflates mass 2–5×. |
| Step 3 · Mass | m = ρ_bulk × (4/3)π(D/2)³, with D and ρ propagated as ranges. |
| Step 4 · Grade | Water 7–15 wt% (CI/CM), Fe-Ni 3.6–17.8 wt%, PGM 0.17–75 ppm with 6.8 ppm at the 50th percentile (Elvis 2013, Table 1). |
| Step 5 · In-situ tonnes | T_insitu = m × g — the optimistic physical inventory. |
| Step 6 · Recoverable tonnes | T_rec = T_insitu × f_extract × f_recovery, f_extract 0.20–0.30, f_recovery 0.80–0.90. These two are engineering assumptions, not sourced values — the UI labels them as such. |
| Step 7 · Uncertainty | Measured D + albedo: ~0.3×–3×. Assumed albedo + vague class: ~0.1×–10×. |
2 · Market layer — what could actually be sold
Three scenarios, mutually exclusive by design. Adding them together would double-count two incompatible business models.
| Earth return · PGM | Earth spot prices, capped by absorbable share. Pt, Pd, Rh, Ir, Au. Water and volatiles excluded. |
|---|---|
| In-space propellant | Priced at avoided launch cost by delivery node (Calla, Fries & Welch 2018): LEO ≈ $3,000/kg, GEO ≈ $21,500/kg, cislunar/NRHO ≈ $35,000/kg. Water/ice only. |
| In-space construction | Avoided launch cost for structural mass. Fe, Ni, Al, Si, regolith — never Earth metal spot prices. |
Path A — absorbable share (default)
q_cap = f × Q0, Rev_yr = q_cap × P0 × (1 − retracement), years to exhaustion N = T_rec / q_cap. f = 10% for PGMs, 2–5% for bulk metals. This is the conservative, easy-to-defend case: no matter the size of the deposit, the market only swallows a slice per year.
Path B — iso-elastic demand (opt-in)
P(q) = P0 × ((Q0 + q)/Q0)^(−1/E). With inelastic demand, revenue peaks at q* = Q0·E/(1−E) and falls beyond it — selling more can earn less. Rhodium fell ~86% year-on-year in 2008–09 on a demand shock of a few percent, which is what that curve looks like in real life.
3 · Finance layer — what it is worth today
| Cash flow | CF_t = Revenue_t − OPEX_t − CAPEX_t, discounted at r. NPV, IRR (bisection on r ∈ [−0.99, 5], null when there is no sign change), discounted payback ('never' when it never turns). |
|---|---|
| Discount rate | Method selector, never a hidden default: Earth mining 8–10% (marked not applicable to space), space-resource hurdle ~25% (McKeown et al. 2024), VC 30–70%, or custom. |
| Mission cost anchors | OSIRIS-REx $1.16B · Psyche ≈ $0.96B · DART ≈ $324M · Hayabusa2 ≈ $150M · KISS ≈ $2.6B. OPEX ≈ 3–8% of CAPEX per year. Each anchor links to the agency's own mission page so the assumption can be checked rather than taken on trust. |
| Success probability | A Bernoulli gate on revenue — failure means zero revenue and fully sunk CAPEX — not a haircut on cost. Default 0.6, slider 0.3–0.9. This is what puts the fat left tail in the distribution. |
| Monte-Carlo | 10,000 draws over CAPEX overrun, OPEX, price, grade, duration and success. Output is P10/P50/P90, P(NPV>0) and a tornado of what actually moves the answer. |
4 · Data sources
Every model input below has a public primary source. Follow the links to verify the numbers yourself — the page does not ask you to take them on trust.
| Orbits & physical data | NASA Science · NSSDCA (JPL Small-Body Database and JPL Horizons ephemerides). Refreshed monthly by an automated workflow. |
|---|---|
| Lunar surface | NASA LRO / LOLA colour and elevation mosaics; polar water-ice estimates from LCROSS (Colaprete 2010). |
| Market volumes | USGS Mineral Commodity Summaries 2025 — annual production weights, reviewed by hand each year. |
| Prices | Johnson Matthey base prices, Kitco and Technic dealer quotes for the PGM basket. A daily pipeline stores the blended basket and the frontend applies it at boot; the packaged snapshot is the floor when no fresh quote exists. |
| Body-specific deposits | Ceres (Dawn/GRaND, Prettyman 2017), Mercury (MESSENGER, Lawrence 2013), Mars (SWIM/USGS), Enceladus (Cassini) — mission data archived at NSSDCA and NASA Science. |
The PGM basket changes daily. The current blended price, each metal's quote date and the coverage check are published on the daily market brief (also JSON / RSS); this page intentionally does not freeze a number that would be stale tomorrow. Weights are USGS 2025 production (Pt 170 t, Pd 190 t, Rh 24 t, Ir 7 t, Ru 30 t → 421 t/yr). Pt and Pd refresh daily; Rh, Ir and Ru ride the dated packaged snapshot.
5 · Where this model can be wrong
- Albedo and taxonomy. An assumed albedo is the single largest error source: a factor of two in albedo is ~2.8× in mass.
- Double counting. One deposit feeds one scenario. Notional and realizable never get added together.
- Local ISRU is not export revenue. Making propellant on Mars to use on Mars is a cost offset, not income — only mass that replaces something launched from Earth to a customer counts.
- Transfer time. Seven years to Saturn kills any NPV at a sane discount rate; time is a first-order term, not a footnote.
- He-3. No commercial fusion reactor, grades in the parts-per-billion. Not counted. Anywhere.
- Price staleness. Three of five PGM prices can be days old. The basket carries its date for exactly this reason.
6 · How the numbers are kept honest
Valuation code is covered by gates that run before every deploy: 59 security assertions, 34 engine edge-case assertions (degenerate inputs, node precedence, Path B fallback), 19 auth assertions, 30 price-math assertions including a lockstep check that the packaged basket equals the pipeline fallback, plus browser-level regression gates for the panels and the workbench. A number that changes without a test changing fails the build.