Impact Study SpaceX Biosphere Forcing

Impact Study — The SpaceX Trajectory and the Biosphere

Page class: IMPACT STUDY. What the sealed record of Study 31 means, mapped onto one operator's measured history and filed plans. Every figure below is produced by a program in reproduce/, answers live on the court, or carries its grade inline. Grades per the ontology.

Who is writing this: people who live under this sky, with children who will live under it longer. We state that because it is the honest disclosure of our stake — the same disclosure we would demand of anyone else. What keeps this page from being an opinion is that every number in it comes from public catalogues, the operator's own regulatory filings (one of them via a press report of the filing, graded REPORTED where used), or a court you can query yourself. Where our stake could have bent a number, the harness caught us, and those corrections are published with dates.

---

The measured record — what one operator has already done

• In 2019, SpaceX accounted for approximately none of the catalogued mass reentering Earth's atmosphere. In 2025, it accounted for 45.8% of it — nearly half of all human-made reentering mass, from zero, in six years. (MEASURED — GCAT catalogue, exact rows, reproduce/operator-reentry-ledger.swift. The scope is artificial mass: against the natural meteoric flux, all human injection combined is currently 3.1–6.3%, and this page never claims otherwise.)
• About one in ten larger stratospheric aerosol particles already contains spacecraft metal (VERIFIED quote — Murphy et al. 2023, PNAS).
• About 460 tonnes of spacecraft dry mass (range 391–531) reentered the atmosphere in 2025; of the ablation products, roughly 45 to 61 tonnes of alumina reaches the stratosphere each season — just under half of the catalogued mass from this operator's replacement cycle (45.8% in 2025). (MEASURED — reproduce/mass-uncertainty-band.swift and reproduce/operator-reentry-ledger.swift; the band is honest because 75.6% of the catalogued mass carries the catalogue's own estimate flag.)
• Human metal injection currently runs at 3.1–6.3% of the natural meteoric flux — the material the stratosphere has received for the whole history of life. (MEASURED — residence time cancels, so the burden ratio is a pure flux ratio.)

None of that is a projection. It has already happened, and no law required anyone to record any of it.

What is filed — the operator's own numbers, not ours

stage	status	what it does to the accounts
19,408 satellites	authorised	metal at 13.0–26.1% of the natural flux; regional ozone 3.1%; 1.0% of all measured greenhouse forcing
100,000 satellites (Gen3)	filed	regional ozone 15.6% — Boulder's median driven below the 5th percentile of its own healthy record; surface UV-B +17% to +31%; 5.3% of all greenhouse forcing
~1,000,000 satellites (orbital data centres)	filed	a 2026-05-29 FCC filing is reported to put 40,000 units per year into disposal reentry (REPORTED — via press coverage of the filing; the filing itself was not fetched) — and reports no unit mass, so the tonnage is a range: 4,900–20,000 tonnes a season at the two sourced masses (Starlink fleet mean 492 kg; Gen3 filed 2 t). At the high end: metal at 5.4–10.9× the natural flux, 43.7% of all measured greenhouse forcing, and an ozone projection that exceeds the physical column and is disqualified

Grades travel with the numbers: fleet sizes, cadence and status are the filings'; the metal-flux ratios are the MEASURED cadence ratio scaled to each filed fleet; regional ozone and UV-B are a PROJECTION (Vliex 2026, scaled linearly) times the MEASURED 12× regional factor; the greenhouse share is that same PROJECTION over the MEASURED AGGI baseline.

The provenance, stated exactly: the authorised counts are the regulator's grant totals (FCC 21-48, FCC 22-91, DA 26-36) — the operator filed for more than was granted; the Gen3 and data-centre fleet sizes are the operator's own FCC applications, with the Gen3 count and the 40,000-per-year disposal split reaching us through press reporting of those filings (REPORTED); the five-year replacement life is the operator's SEC filing, made for broadband satellites and applied here to the data-centre fleet as a stated assumption. We did not pick worse numbers, and we could have.

The finding that matters is the first row. The fleet that is already licensed would move three independent planetary accounts by percent-scale amounts simultaneously, at its SEC-filed five-year replacement life (PROJECTION — the licences exist; the steady-state flux they imply has not yet been reached) — and each account alone looks survivable, which is exactly how the exposure has stayed invisible. No instrument on Earth is required to report any of the three. This answers live: nine machines, marker STUDY31_BIOSPHERE_JOINT_LEDGER_PROVEN, call on the study page.

The harm to humanity, stated by grade

What is already measured, in bodies: the 2026 humid-heat season put 41,174 people in ambulances in one month in Japan, 79 confirmed dead, while Korea recorded 42.5 °C, the highest in 122 years of observation. That harm is not attributed to satellites — it is on this wiki (Study 28) because it is the measured precedent for what an atmospheric change feels like when it arrives: as a shifted statistic, graded by operational instruments whose own documentation licenses verdict-sized error. The atmosphere is already killing people at the margin, and the instruments deciding when to warn them are already too coarse. That is the system the reentry flux perturbs.

Established mechanism, unmeasured magnitude — every chain below is textbook science; none has been measured against this forcing, because the forcing itself is not measured:

• Ozone loss raises surface UV-B → skin cancer and cataract rates shift across whole populations, over decades — a harm that arrives as statistics, never as an event anyone can point to.
• Crop yields respond to UV-B at percent scale.
• Phytoplankton — the base of the marine food web, the primary oceanic carbon sink, a large share of oxygen production — are UV-sensitive, and damage there feeds back into the greenhouse account this same ledger measures.
• The polar vortex responds — and this comes from the reassuring paper: Maloney 2025, the study that finds a weaker ozone hole, also reports ~1.5 K anomalies and a ~10% vortex-wind reduction at a 10 Gg/yr flux. A weakened vortex is mid-latitude weather: cold-air outbreaks, disrupted seasons.

The planet has run this experiment once, small: the 2019–20 Australian fires put ~1 Tg of aerosol into the stratosphere by the same mechanism class now claimed for alumina. Observed: reactive chlorine enhanced, HCl 50–60% below climatology (REPORTED — Santee et al. 2022, satellite MLS). Modelled from those observations: 3–5% southern mid-latitude ozone loss (REPORTED — Solomon et al., Nature 615, 2023). Not nothing. Not collapse. A few percent from one transient event — and the filed constellations are a tap left running, not an eruption decaying away.

NOT_KNOWN, and refused a number: the phytoplankton dose-response at these forcings, the food-web and oxygen transfer functions, and the sign of both the ozone and radiative responses, on which the published models disagree. The court answers NOT_KNOWN on these and excludes them from every total. The exclusions cut in different directions, and the page says which: the feedback couplings, if quantified, would add rather than subtract — with respect to them the totals are a floor. The contested signs could move the ozone and radiative rows either way, and no floor is claimed over them.

The narrow-study defence, and why it does not work

The strongest argument for doing nothing goes like this: the published studies disagree, some find no harm, one finds benefit — so the science is unsettled, and unsettled science is no basis for regulation.

That argument is in circulation right now, and it deserves a precise answer rather than a dismissive one, because the studies it leans on are real, peer-reviewed, and carried on this very page as evidence in the operator's favour.

The answer is that every one of those studies is an instrument pointed at one axis of a multi-axis system, and a disagreement between narrow instruments is not a verdict about the system. Look at what the record actually shows:

• Each study models one axis and is silent on the rest. The biosphere column is empty in all five. No two of them share enough axes to even be added together. A study that never modelled the ocean food web cannot find it safe — it can only fail to mention it, and silence transmitted as reassurance is the exact mechanism by which a narrow result becomes a general one. "No effect on the axis I modelled" gets quoted as "no effect" by the time it reaches a licensing decision.
• Their disagreements are demonstrably arithmetic, not atmospheric. The field's famous "28× divergence" dissolved to 3.2× when two papers' quantities were put in one set of units — no model run required. Barker's own model returns +6.47 mW/m² under one forcing convention and −6.40 under the other, from identical inputs: one instrument, two conventions, opposite signs. The global means every group reports are twelve times kinder than the regional worst case the ozone record actually produced. None of that is the atmosphere disagreeing with itself. It is the continuous-arithmetic pipeline shearing — the same floating-point, convention-dependent, average-first machinery whose vendors document, in their own manuals, that it does not reproduce across platforms.
• A probabilistic treatment cannot be conservative about an axis it marginalised out. Averaging over spatial gradients, ensemble members, and parameter priors produces a smooth curve on the modelled axis — and says exactly nothing about the coupled axes that were never in the state vector. The smoothness is a property of the method, not of the planet.

So the honest statement of the situation is not "the science is unsettled, relax." It is:

The instruments disagree with each other on sign, for reasons traceable to their own arithmetic; none of them models the coupled system; and the one account that holds every axis at once shows three of them moving together at the already-licensed scale. Divergence between narrow instruments is not evidence of safety. It is evidence that the instruments shear — and it is the reason to measure, not the reason to wait.

That is also why this programme computes in exact integers, on the complete account, with the unmeasured couplings named rather than averaged away. Not because exactness is elegant — because when instruments cannot be reconciled, the loudest institution wins by default, and the sky should not be governed by default.

The question before the comparison

Every bullet in the next section compares this operator favourably within an activity. Public flourishing as the first principle imposes an ordering the comparisons must not be allowed to skip:

1. Is the service needed? 2. Is it deliverable without the atmospheric flux? 3. Only for the remainder that survives both questions do the harm comparisons apply.

A cleaner propellant answers which rocket. It silently concedes whether the rocket — and a slightly safer version of an unnecessary activity is harm reduction, not engineering. The substitutability question is answerable per service, with grades:

• Orbital AI compute: fully substitutable. Ground data centres perform this workload today; the orbital filing is an energy-economics bet, not a capability the ground lacks. For this service the propellant comparison is moot — the flux itself is the choice. (MEASURED against the alternative: the verification workload runs at 3.240 J/hour on the ground, and the ownership math is published both ways.)
• Mass-market connectivity: substantially substitutable — and the stakes are not subscriptions, they are warning chains (the lives axis, measured). Fibre, fixed wireless and community-owned mesh serve the overwhelming majority of the subscriber base the constellations bill; for the verdict layer — gauges, warnings, court calls — the ground mesh delivers it with zero rent and zero reentry mass. (MEASURED for the verdict layer; the broadband scope refusal stands in capitals in the producing program.)
• The genuine remainder: maritime, polar, disaster-cut and truly unreachable users. This is the fraction of the service that is not currently deliverable from the ground at comparable capability, and this page does not pretend otherwise. (STATED AS THE REMAINDER — unquantified here; quantifying it is exactly the kind of measurement nobody is required to produce.)

So the favours below are read at their true size: they are facts about the remainder, not licences for the whole. The chlorine advantage is real for launches that survive questions 1 and 2. For the fully substitutable services — the million-satellite compute filing above all — the flourishing-first answer precedes chemistry: the cleanest reentry is the one that does not happen, and for that service, it does not have to.

Danger versus overkill — the shape of the bet

Strip everything above to its decision structure and this is what remains, with every input graded on this wiki:

side of the bet	what it is	grade
the upside	orbital compute and mass-market connectivity — services the ground already delivers or can deliver, at 3.240 J/hour for the verdict layer and in ground data centres for the rest	SUBSTITUTABLE — measured above
the downside	three planetary accounts moving at once at the licensed scale; the chain into the ocean food web graded NOT_KNOWN; the projections disqualifying themselves past the filed scale because the instruments leave their valid range	UNMEASURED, planetary, shared by everyone — including the operator

This is not "do this or die." Nothing on the necessity side compels the flux — that is what the substitutability section establishes, service by service. The honest structure is the other one: do this optional thing, and accept an unmeasured planetary-scale tail on behalf of everyone alive and everyone after them.

That asymmetry is the whole argument, and it needs no catastrophe proof to bind:

• A bounded, replaceable upside cannot price an unbounded, unmeasured downside. Not because the downside is proven large — this page has refused every such proof — but because it is proven unmeasured, at the exact links where the biosphere's food and oxygen begin. NOT_KNOWN is not zero. Betting as if it were zero is not engineering; it is exposure without an instrument.
• The bet is placed on parties who never sat at the table. The upside accrues to an operator and its subscribers. The downside is drawn on the one atmosphere, against every person under it — which is why this page keeps saying we live here, and why the licences' skipped environmental review is a procedural fact that matters.
• Overkill is the precise word for the upside. A million-satellite compute fleet is not the only way to run the workload — it is not even the cheap way past the first hardware replacement cycle, on the operator's own filed five-year life. An optional activity, deliverable from the ground, carrying an unmeasured planetary tail, is not a trade-off. It is an unnecessary bet with everyone's stake on the table.

The refusal this page asks for is therefore the mildest one available: do not place the unmeasured bet while the measurement is this cheap. Measure the flux, publish the regional minima, run the dose–response — three years, near-zero cost — and then let the licences argue from data. Until then, the flourishing-first answer stands: we do not need what is being sold, and the price on the tag is not the price of the bet.

Is it worth it at any odds — and who decides?

Put the question at its sharpest: suppose the chance that this cascades into the food chain were one in a million. Is it worth it?

First honesty: one-in-a-million is not a measurement. No instrument on Earth currently produces that number — the dose–response has never been run, the transfer functions do not exist, and this page's own chain stops at NOT_KNOWN exactly there. A stated tiny probability would at least be an instrument reading; what actually exists is no reading at all. That is worse for the bet, not better: you cannot price a tail you have refused to measure.

Second: even granting the number, the arithmetic does not close. A small probability prices a bounded loss — that is what insurance is. It cannot price a loss that includes the ability to try again. A wager that consumes the table on failure has no acceptable odds when the upside is substitutable at no comparable risk — and the substitution is measured, service by service, above. Small annual probabilities also compound over a permanent program: a flux that runs forever is not one draw from the urn, it is a draw every year, indefinitely, and the tail accumulates while the measurement stays unfunded.

Third — who decides? The record already answers, and the answer is the scandal: as of today, nobody decided. The licences issued under a categorical exclusion mean the question was never formally asked — not answered wrongly, never asked — while the parties bearing the stake are everyone under the atmosphere and everyone after them, none of whom hold a seat. The upside has an owner; the downside has no representative.

The constructive answer to who decides is the one this whole substrate exists to make possible: a decision of this class belongs to the people staked, and they can only decide what they can check. That is what measurement buys — it converts an unconsented gamble into a decidable public question, with verdicts the bound parties re-derive instead of trust. Three years of inventory, one output field of regional minima, one standard dose–response experiment. Until those exist, the honest status of the wager is: placed without odds, without consent, and without an instrument — on everyone.

What cuts in SpaceX's favour, with the same weight

What a favours edge is, and is not — the observer-independence rule. Each entry records evidence, not endorsement, and each must state its axis and its comparator — because a relative claim quoted without them changes meaning with whoever is selling it, while a claim with both pinned is the same fact for every observer, which is the only kind this closed system's ledger admits. The recurring sales structure across every vertical is exactly that omission: a benefit stated relative to a chosen worst comparator while the absolute, measured axis stays silent. "Cleaner than solid propellant" on the chlorine axis. "Safer than the average human driver" on a fleet-statistics axis. "The studies disagree" on a single-model axis. Each is real on its stated axis, and each goes silent on the question that binds: the absolute flux, and who bears its unmeasured tail. The edges below carry their axes so they cannot be quoted past them.

And the capacity of that shear is not hypothetical — this programme has measured it at astronomical scale. The same continuous arithmetic these verticals settle their claims in has, on this wiki's own sealed record, manufactured structure that exact counting dissolves: a black-hole "ring" that lives in the normalised pipeline and not in the raw visibilities (Study 07, 8/8 raw WIN, 8/8 normalised MISS), float triangulations that shear where integer counts hold (Studies 11–13), and — chartered, not concluded — the open question of whether a stacked continuous analysis can conjure the appearance of missing mass at galactic scale (Study 10, which explicitly does not disprove dark matter and says so). The inference runs one way and it is the plain logic this page runs on: an arithmetic medium capable of hiding galaxies' worth of apparent mass in its shear is trivially capable of hiding a company's inconvenient facts in the same shear. That is why every claim on this page is an integer a stranger can re-derive — the one medium where hiding is not available to either side, including ours.

• Their engines burn kerosene, which emits no chlorine. (CITED, NOT MEASURED — Barker 2026, carried as published.) Axis and comparator, pinned: a relative claim, on the chlorine axis only, against the dirtiest comparator — solid propellant. True there, and silent everywhere the ledger measures: on the absolute mass axis every kerosene launch still adds flux, this operator is the measured largest single contributor of catalogued reentering mass (45.8%), and "better than solid fuel" is not "making the sky cleaner" — it is adding to the stratosphere more cleanly than the worst available way of adding to it. Conditional per the ordering above: a fact about launches that survive the necessity and substitutability questions, not a licence for the ones that do not.
• One peer-reviewed model finds the opposite sign (a weaker ozone hole); another finds a small ozone increase. (CITED, NOT MEASURED — Maloney 2025; Revell 2025, whose alumina is launch alumina.)
• The radiative term today is negligible: 0.1% of all greenhouse forcing — one part in 860. It becomes large only at the filed, unbuilt scales. (PROJECTION over a MEASURED baseline — reproduce/radiative-baseline.swift.)
• The most recent year is not continuing the climb. 2026 to date annualises to 187.3 t/yr against 2025's 211.6 t — 0.89×, flat to slightly down — and the 2026 mass share reads 44.5% against 2025's 45.8%. (MEASURED — the same program as the headline figure, reproduce/operator-reentry-ledger.swift, which states this itself rather than glossing it.)
• On cost at the moment of purchase, they win: $1,594,900 against our $1,825,000 at like-for-like aggregation. (MEASURED — reproduce/cost-matrix.swift.) Completed over time on their own SEC-filed five-year hardware life, the ownership path overtakes at the first replacement cycle and wins every year after ($4,784,700 vs $3,193,750 through year fifteen at our worst maintenance band) — both readings published, because a t=0 price and a lifetime cost are different quantities and quoting only the favourable one, in either direction, is the units error this programme keeps a memory about. (MEASURED — reproduce/cost-ownership-horizon.swift.)
• The catastrophic models we were offered against them failed our own control arms and are refused on their own arithmetic, on the study page, in public. (MEASURED — both refutations are programs anyone can run.)

A page that hid any of this would be advocacy. This one is a ledger.

The impact statement

One operator has moved, in six years, from zero to nearly half of all catalogued human-made mass reentering Earth's atmosphere, and holds licences for a fleet whose replacement cycle projects human metal injection into the stratosphere to between an eighth and a quarter of the natural flux — with filings on record that project it to five to eleven times nature and to nearly half the radiative forcing of every greenhouse gas humanity has ever emitted. The biological consequence of that trajectory is not known, because the transfer functions have never been measured, and no law requires the forcing itself to be recorded. The harm to humanity is therefore not a prediction — it is an exposure: three planetary accounts moving at percent scale under an environmental review skipped by standing rule — a rule the government's own auditor has questioned at this scale (grounded here) — graded NOT_KNOWN at the exact link where the ocean food web begins. We live here. So do they.

What would change this assessment

Stated so it can be falsified, not so it can be believed:

1. A reentry emissions inventory — mass, composition, ablation altitude, per object. Three years of it settles the sign disputes that five models currently cannot.
2. A UV-B dose-response measurement for phytoplankton at the projected regional doses. Standard marine biology; unfunded.
3. Regional minima published alongside global means by the existing models — one output field.
4. Any demonstration that the ablated fraction, the 12× regional factor, or the linear scaling breaks in the kind direction — each is flagged in the study as an assumption that could fail either way.

If those come back benign, this page gets a dated correction saying so, the way our other wrong numbers did. Until someone measures, the exposure stands, and it is not ours to accept on anyone's behalf — including on behalf of the operator's own children, who live under the same sky.

---

Related: Study 31 — the joint ledger, live · The forcing nobody measures · Ask someone you trust to check this · Study 28 — the wet-bulb court · The ontology