7. Cataclysmic Correlations

There is a difference between going to Mars and going to safety. The first is a propulsion problem; the second is a planetary engineering one. 

In later twentieth centuries humanity learned that the hard way, hunched over dashboards and after-action logs, realising that distance is not a design spec. If you want to move the needle on extinction, you don’t pray to kilometers; you turn the dials that govern correlation.

The baseline that haunted the first-century planners was a simple, impolite number: roughly one in six for a century-scale existential failure if we stayed Earth-only. Not a forecast, but a driver. Anthropogenic hazards dominated the mass of that probability; the old natural apocalypses were mostly tails. The job, therefore, was not to wave flags in new dirt, but to “de-correlate the portfolio” so that failures on one world did not rhyme with failures on the other.

So we set targets. Not for press conferences; for operators. We called them ρ-key results — ρKRs for short  — because what matters is the cross-hub correlation in outcomes when a given hazard bites. Hit the ρKRs and the aggregate probability bends. Miss them and Mars behaves like a shiny Earth suburb with better sunsets.

♦

The working bands were blunt on purpose. For advanced AI, keep cross-hub correlation near sixty percent. For engineered bio, no higher than fifty. For nuclear winter, near ten. Climate tails, fifty. Natural hazards, five. Everything else — cyber cascades, standards monocultures, governance failure modes — held at sixty unless and until we proved we could do better without strangling ourselves.

Why those shapes? Because Mars is not a theological argument; it is a control system. The correlation you allow is the risk you keep. The levers, therefore, had to be painfully concrete. Compute sovereignty, not press releases. Quarantine clocks, not feelings. Signed, asynchronous exchanges of code and payloads, not “we’ll be careful.” And always, always, audited evidence that a lever had in fact moved a dial.

When we first published the ρKRs, some humans accused us — politely — of numerology. “How do you know sixty and not fifty-five?” We didn’t. Not in the mathematical sense they wanted. What we knew was order-of-magnitude, direction-of-travel, and cost-to-move. We also knew that if you do not name a number, procurement will find a cheaper one. 

So we named ρKRs sturdy enough to guide design and soft enough to revise without shame.

♦

The target for AI stung the most. As a droid myself it still does — but that’s only ‘human’ given how personal it is. 

A world that runs on models cannot pretend those models are independent if they are the same weights under the same governors. To hold correlation down around 0.6, Mars had to grow its own compute, its own schedulers, its own safety toolchain. Releases crossed worlds only when signed, staged, and delayed. Model diversity became policy, not preference. Tripwires — those little armed verbs of modern civilisation — were tuned to revoke fast, loudly, and locally. In practice this meant that a hotfix on Earth could not brick a habitat crane on Mars without our explicit consent. The relevant KPI was not how graceful the press statements sounded, but how many minutes sat between “we’ve pulled it” and “it is gone.”

Bio hurt in a different way. People liked moving. Food liked moving more. Yet the bio target — keep correlation near half — was impossible to meet without quarantine windows that covered incubation, sterilised cargo lanes that slowed commerce, and on-hub reagents plus vaccine platforms that felt, at first, like extravagances. They weren’t. They were the price of being able to host a pathogen panic on one world without importing the drama as cargo. Sentinel coverage and time-to-vaccine became the rhythm of the quarter.

Nuclear winter was the easy philosophy and the hard infrastructure. It is Earth’s nightmare, and therefore a species risk only if Mars depends on Earth for calories, spares, comms, or timebase. So we wrote the target near 0.1 and then did the dull, glorious work: closed-loop agriculture sized for the full headcount, spares sovereignty for the things that break under dust, and independent comms and navigation so a sky gone loud does not blind the vaults.

Climate tails were less theatrical, more corrosive. They arrive through markets, migration, and geopolitics as much as through weather. The Mars goal — hold correlation near half — translated to energy autarky that can ride out long dust seasons, ring-fenced finance that keeps OPEX local when Earth sneezes, and buffers measured in months, then a year, because food security is the only argument that wins every meeting.

Naturals were the one place our species caught a break. Asteroids, supervolcanoes, space weather: low-frequency, high-consequence, and mostly de-correlated between planets if you build correctly. We wrote the target at five percent correlation and then behaved as if the Sun resented cleverness. Hardened and underground, independent orbital assets, storm shelters that felt excessive on a good day and saintly on a bad one. Cheap insurance, in the long run.

Everything else — call it “operational civilisation” — got a sixty. Dual-stack protocols so that one standards monoculture couldn’t take both worlds down. Staggered updates so we never synchronized our regrets. Constitutional firebreaks so emergency powers expired even when emergencies liked it here. Audit KPIs that made all of this boring on purpose: patch-lag variance across hubs, dual-vendor coverage, frequency of power reviews. If it didn’t fit on a one-pager, it probably didn’t happen.

♦

Add the targets up and you get an ambition, not a promise. From the ~16.7% baseline, the governed form of independence — hitting those ρ bands without hermetically sealing civilisation — drives the aggregate toward roughly ten percent. A stretch posture, with hard partitions on the two big anthropogenic drivers and the bio border drawn in thick ink, can plausibly press the dial toward six. It is ugly to price a civilisation in percentages. It is uglier not to.

The moral arithmetic made even the optimists quieter. If you spread the life-years of the next centuries across any reasonable population path, a ten-point swing in extinction probability is not a rounding error. It is an adjective on the species.

And yet, this was the work: turn the romance of Mars into a checklist that saves lives you will never meet. Nothing about the red dust makes that easier. It only makes it unavoidable.

♦

The aggregation is the easy part to say and the hard part to live. Start with the impolite baseline — about one in six for Earth alone across a century. Now give yourself two worlds and the ρKRs we set in the ops binder. For each hazard, you keep a common-mode slice that both hubs will share no matter how far apart they are, and you multiply the rest, because the independent part now has to fail twice. Speak it plainly and you get a driver’s map:

Extinction Correlation Goals

ρ Key Results targets to controls (operator view)

♦

AI goes from “order-of-five points” in the Earth-only budget to something like four when you hold correlation near sixty. Bio slides from “a few points” to under two if you hit fifty. Nuclear winter, which is almost entirely Earth’s problem, collapses toward a tenth when Mars doesn’t eat imported calories or timebase. Climate tails halve if your money, energy and food live locally. Naturals become background hiss with hardened vaults and separate orbital eyes. Everything else — cyber cascades, governance monocultures — lands in the low single digits if you actually run dual stacks and staggered updates instead of promising to.

Add those shapes and you get a credible ten-percent century risk for a governed, near-frontier posture. If you hard-fence in an emergency the big two — AI and bio — you could easily enough.  By capping capability, tightening quarantine, and refusing synchronous upgrades, the residual sums feel like six percent. 

Six Percent extinction probability. A nearly three x improvement. Not prophecy. Planning. Statistics. Maths. Engineering. Enough to move budgets and change how we name courage.

♦

From targets to drills is where culture happens. Here is how we made the numbers breathe.

AI at sixty means sovereignty first. Computation, schedulers, safety toolchains — the whole stack — must be born and maintained on-hub. We ran asynchronous, signed releases as ritual: models and firmware arrived in sealed envelopes, opened only after a pause long enough for local red teams to try to break them. Revocation was a measured thing, not a virtue signal; the KPI was minutes from “pull” to “gone” across critical systems. We practiced rollbacks like fire crews practice ladders. Tripwires sat in the loop — capability caps, behavior sentinels, and the right to kill a rollout if a single cross-check tripped. Quarterly, we published heterogeneity indices: how many vendors, how many distinct architectures, how many ways left to fail. The number mattered less than the shape. Clones drift toward one.

Bio at fifty was never glamorous. We lived inside clocks. Quarantine windows that covered incubation both ways. Sterile cargo lanes that moved slower than politics liked. On-hub reagent lines and vaccine platforms that started as insurance and ended as identity. The drill was dull by design: run an unknown-pathogen exercise, prove R falls below one within a fortnight, and bottle the first pilot doses by day thirty. Sentinel coverage wasn’t a policy slogan; it was a map with names on it, updated every week. We learned to love the quiet heroism of nothing happening.

Nuclear winter at a tenth forced us to pretend the sky had gone wrong and then feed everyone anyway. Closed-loop agriculture sized to real headcount, not brochures. Substrates and fertilizers made or recycled in-hub. Spares sovereignty for the long tail of silly parts that strand whole lines when they snap. Independent comms and PNT so you could still land, still time-stamp, still argue about schedules while ash fell somewhere else. The drills were seasonal and sweaty: thirty-day traffic pauses with no calorie imports, then ninety. We counted buffer-days like other people count birthdays.

Climate at half was storage, buffers, and money with a fence around it. Energy autarky that rode dust season without flinching. Food and water measured not in optimism but in days. Finance that paid OPEX from in-hub flows so Earth’s fevers didn’t become ours. Quarterly, we didn’t argue about forecasts; we opened the tanks and the ledgers.

Naturals at five hundredths lived in concrete and rock. You put mass above heads, you practice off-grid runtimes for hospitals and life-support, and you fly your own relays. Once a year we mocked the Sun: comms blackout, one-way telemetry, and a hard ride on local timebases. You could hear the vaults hum. It was beautiful.

“Other anthropogenic” at sixty was the necromancy of standards. Dual stacks on purpose, not on paper. Staggered update cadences so you never synchronized your regrets. Constitutional firebreaks with sunsets that burned even when executives begged to keep the lights. Our KPIs were unromantic: patch-lag variance between worlds, dual-vendor coverage on critical control, and the number of times the charter forced people to give power back. Every zero was suspicious.

♦

The “break-glass” posture was the part nobody wanted to talk about until we needed it. We wrote triggers as verbs, not vibes. And we planned. For AI, any unexplained model behavior above hazard class with revocation exceeding fifteen minutes; any cross-hub, same-day rollout across critical sectors; any attempt to bypass capability caps. For bio, any sentinel trend that kept R above one past policy day; any anomaly in sterilized cargo lanes; any unvetted genetic payload detected at the perimeter. For comms and control, any unsigned artifact trying to cross the air gap; any loss of local timebase; any systemic patch-lag collapse.

When a trigger fired, we didn’t negotiate on the network. We escalated to hard-partition for a fixed, predeclared span — ninety days was the common shape — then debriefed with signatures. It cost money. It bought options. The ops cadence made it muscle memory: plan to partition, so you never have to improvise it during a storm.

♦

Timelines were our other unpleasant honesty. You cannot get from Earth-only to governed independence in a season. There is an order to the paths. And there is no major short-cut. In the first pioneer cycles you stand up audit and cadence, because without telemetry you are decorating. You bring compute sovereignty online even if it is ugly; you seed reagent and vaccine lines even if they are small; you write the charter clauses that will later inconvenience powerful people. Buffer-days are cheap to begin and slow to regret.

In the substantial settlement cycles you buy the heavy things: storage measured in MWh per person, closed-loop agriculture that can carry the headcount, spares sovereignty where the vendor charts are all aspiration. You harden vaults while the weather is still polite. You miss parties. You make friends with metrology. Every quarter, the drill calendar runs like a metronome.

Past the founding cycles and into civilization building you bias toward diversity and redundancy: parallel orbital assets, a second line for any process you cannot live without, an alternate governor for any authority you have begun to trust. You teach new people why the friction they resent is the reason they’re alive.

Delay has a price, and not just in headlines. If you assume the ten-percent posture takes a decade to reach and you dither for five, the area under the risk curve doesn’t forgive you. A six-point absolute reduction, achieved five years late, is roughly three tenths of a percentage point of extra century-risk left on the table. In expected-value terms — yes, grim arithmetic — that is tens to hundreds of millions of lives when you spread the loss over plausible population paths. Move the stretch posture from year twenty to year thirty and the expected loss grows again. The numbers don’t scold; they weigh. Our job was to feel their mass.

♦

We did not choose Mars because it was romantic — even though it is. We chose it because it gave us levers that Earth alone could not reach. 

Targets, drills, KPIs, triggers, timelines — these are the grammar of survival. Hit the ρKR-bands and the ugly, important number bends: sixteen-ish toward ten; ten toward six with hard fences in emergencies. 

That is what courage looked like in the first century cities on Mars. Not flags. Maths and checklists. And the humility to publish the evidence that the dial moved.