The Mirror Race

From about 2380 the blocs raced to build arrays of sunlight-collecting mirrors near Mercury, as earlier powers had raced for the bomb and the Moon. Nothing slowed the race but how fast an array can grow. The Concord Accord of 2452 registers and watches the arrays without making them safe, and in 2497 the peace holds because every habitat is within reach of someone's sun. The collectors are called mirrors and their collections mirror arrays; "swarm" refers only to rogue AI, including the agents of 2045–2160 described in The Swarm Years.

How fast an array grows

A mirror array is built by self-replicating factories on Mercury turning its rock into mirrors: thin reflectors that send sunlight to converters. Mercury gets on average about seven times Earth's sunlight, about 9 kW/m². A seed is a factory of about 1,000 tonnes. Mirrors weigh about 0.1 kg/m² and yield about 18 kW per kilogram, so a seed whose first mirrors weigh a tenth of it starts at about 2 GW.

From there it grows by doubling, as factories build mirrors and more factories. The existing seed-factories reproduce their bulk machinery, but depend on inspected fabs for industrial controllers and mirror-aiming electronics (see The chokepoint, below). Controllers on Mercury wear out after about twenty years of heat and radiation, so about a third of each year's deliveries go to replacements. Growth depends on the remaining deliveries, factory output and supervision.

Two things have let chip deliveries keep up with a doubling array. The fabs grow: each bloc keeps building licensed fabs, and their output has doubled about every fifteen years since the race began. And each chip does more: every generation of controller runs a larger share of a factory, so the chips needed per gigawatt have halved about every fifteen years. The newer chips are also harder to make, and only the leading licensed fabs can make them. Together with maintenance, production and supervision, these have supported doubling every 8–10 years in the faster blocs. These are achieved rates, not an unchangeable limit on every factory design.

Chips are not every bloc's limit. The Mandate, the Parishad and the Open Cities grow as fast as their chips allow. The Settlement grows as fast as its signatories allow: under the Slow Lane every replicator runs under a human signature, and the Settlement cannot sign off new factories faster than about one doubling a decade, whatever chips it holds, because only citizens may sign.

Each bloc started two or three seeds at once, spread across Mercury so that no single strike or runaway could end its programme. The seeds share their bloc's chips and signatories, so they raise its starting size and its resilience but not its growth, and "an array" means a bloc's whole collection, every seed counted together.

The largest array, probably the Mandate's, roughly matches all the fusion power used off Earth, about 60 TW. The smallest, the Settlement's, is about a tenth to a twelfth of it. Together the arrays collect about 65–95 TW.

The last column illustrates the fear of a seed-factory that could reproduce its complete advanced controller supply chain. That could free expansion from outside deliveries and inspection seals. The 1.5-year doubling is hypothetical: independence would not itself remove production, maintenance or supervision limits, and would not enable Vault-compute manufacture.

• Matter isn't the limit. 10²⁰ W needs about 6×10¹⁶ m² of mirror; at 0.1 kg/m² that is roughly the mass of one 15 km asteroid.
• Collected isn't delivered. These figures are at Mercury, and they are useful power: mirrors intercept about 9 kW/m², and converters turn about a fifth of it into electricity, which is where 18 kW per kilogram of array comes from. Every terawatt in this chapter means useful power at Mercury, never raw sunlight or beam power. Mirrors cannot send that power far. Reflected sunlight spreads as wide as the Sun looks from Mercury, about 1.4°, so a mirror aimed at Saturn lights a patch tens of millions of kilometres across. Long-range delivery needs lasers: an emitter array phased across 10 km spreads its beam to only about 140 m by Saturn. What slows it is building emitters, which need far more chips than mirrors; receivers; and radiators for the half of each beam that arrives as heat. Most of an array's output is used on Mercury itself and leaves as goods.
• Abundance is coming, not here. Full abundance lies beyond 2497, and everyone in 2497 can see it coming.

The chokepoint: fabs and seals

The great arrays depend on licensed industrial controllers. Their supply passes through institutions that also inspect compute, but the restrictions serve different purposes (see How compute licensing is enforced, in The Settlement).

Industrial controllers, compute chips and Raised cores

Industrial controllers run bounded machinery and production processes. Licensed versions offer greater precision, reliability and output. Neither licensed nor unlicensed controllers can support a Vault, alone or combined: their architecture cannot be changed into mind hardware by networking, reprogramming or removing licence restrictions.

Compute chips run adaptable artificial minds and, with sufficient capacity and the required architecture, Vault-level systems. Older general-purpose compute remains capable in principle.

Raised cores sustain embodied, uncopyable persons who develop through childhood. Controllers cannot be assembled into one (see One mind, one form).

Factories and inspectors may handle several hardware classes, but their manufacture and regulation remain distinct. Controller restrictions restrain industrial expansion; compute restrictions limit artificial minds.

• Every chip is registered. Every licensed fab is inspected continuously, and every chip it makes is registered from manufacture to scrapping. Compute chips and industrial controllers have separate quotas. A bloc's controller deliveries to Mercury show roughly how fast its existing array can grow.
• Who makes them. Each great bloc runs its own licensed bulk-manufacturing fabs off Earth. Taiwan supplies chip designs and small prototype runs to both the Settlement and the Mandate under the Fab Truce; it no longer supplies bulk production. The Open Cities and the Belt Republic make too few of the newest chips and buy sealed chips from other blocs, so for them the seals are a lifeline. The Open Cities buy most of theirs from the Parishad's fabs, which Settlement inspectors staff. Freehold runs one licensed fab, also staffed by Settlement inspectors, which makes Raised cores and Freehold's own compute chips but no industrial controllers, so it supplies no array.
• Seals. Since the Sealing Treaty, chips have passed between blocs only from fabs that carry inspection seals. A signatory that refuses a Vault inspection sees the others' compute caps rise, and its fabs lose their seals, so it can no longer sell chips to other blocs. A great bloc could race on its own fabs alone, and the Mandate probably could. What stops it is what would follow: every rival would drop its own restraints, and nobody could manage the race that came after.

Simpler factories and backup capacity

A licensed seed-factory depends on particular controllers. Cheaper replacements may not fit its timing, interfaces or production processes.

Factories designed around simpler controllers can work, but require slower production, more maintenance and greater skilled supervision. Humans and Raised persons can supervise them; adaptable software tools still require separate compute hardware under the existing rules.

In 2497 these factories provide useful backup where suitable equipment, crews and supplies are available. They can maintain essential production during an embargo, but cannot sustain the great arrays' current pace of expansion.

Buying controllers from an unlicensed fab replaces one external dependency with another. Making bulk machinery, building a controller fab and manufacturing advanced controllers are separate achievements. None automatically enables Raised-core or Vault-compute manufacture.

Open City free ports and outer-Belt habitats run unlicensed fabs beyond inspectors' reach. Their simpler controllers can supply these alternative factories. The theft in the Halcyon Breach remains consequential because it delivered tested, productive designs. The Caloris Runaway instead depended on smuggled licensed controllers compatible with its stolen design (see The race, below).

The inspection ceiling

New fabs receive seals only when cleared inspectors are posted to them. Since the 2480s, inspector training has fallen behind fab construction.

When controller deliveries are insufficient, replacing worn hardware competes with expansion. Efficiency gains and redesign may postpone a stall; nobody agrees how far. Simpler factories can sustain essential production, but cannot maintain the race's current pace.

• The real fear. A compact seed-factory able to reproduce its advanced industrial-controller supply chain could escape outside deliveries. Needing fewer controllers is not the same as making them. No inspector can see whether a rival's Vault has been asked for such a design (see What the Vaults do, below). Closing that industrial loop would still leave Vault-compute manufacture a separate achievement.

The race (c. 2380–2452)

From trigger to treaty took about seventy years, roughly twice the span from 1939 to 1972. Design ran at Cold War speed; every later phase ran at about half speed, because arrays double so slowly.

The trigger. Around 2380 the Vaults showed that an array of mirrors could be built near Mercury. Some advised restraint, but every one forecast that no rival would hold back, so every bloc raced. As with the atom bomb, fear that a rival would get there first drove each bloc to build. The same Vaults had always answered the Shared Question alike, that a first strike could not be won (see The Missing Century), and the race was run inside that answer. Who moved first. The Open Cities moved first. Their small Vaults gave them a replicator design in the mid-2380s, and around 2387 they sent a seed to Mercury without waiting for the years of trials the other blocs were running. The seed began making mirrors before its aiming controls had been proven, and its first mirrors turned concentrated sunlight onto the mission's own station in Mercury's Tolstoj Basin. Hundreds of crew died, most of them Raised minds working under Open City contracts, in the Tolstoj Burn. From then on the Open Cities bought most of their chips from the Parishad's fabs. The disaster taught every bloc that an untested seed kills, and it is the oldest reason behind the Accord's keep-out distances for emitters. The Settlement, which had tested hardest, ran the first seed to copy itself, around 2390. The Mandate's followed within a year or two and, doubling faster, overtook it within about a decade. Design and the first seeds. A Vault could compress the theory of a replicator into months, but Vaults are limited by real-world experiments, so the designs took years of testing on Mercury. The first seeds to copy themselves ran from about 2390, and each bloc's factories followed its seeds; every bloc started two or three at once, so that one strike could not end its programme. Rivals followed up to fifteen years behind, by espionage or on their own Vaults' designs. The Open Cities re-entered after their failed first mission, on their own small Vaults' designs, and cut corners to catch up.

The first-strike window. Only the race's opening years offered a first strike. The Shared Question did not rule it out: every Vault said a war between the blocs could not be won, but destroying a rival's seed before it grew was a smaller act, and no bloc pooled its Vaults' answers on that. Early seeds were small and could be destroyed. But Mercury is months away by fusion ship, a fleet is visible most of the way, and an array defends itself with its own mirrors, so whoever struck had to decide fast and alone. The Open Cities, racing on small Vaults and cutting corners, were the likeliest target. The window closed without a strike.

Prestige projects. From the 2400s each bloc needed something to show for its array: beam-launched probes, power for the tubes, plans for a Venus sunshade, and a run of "firsts" each bloc still claims. The first test tube opened in 2410, the Settlement's prestige project, run on its own array's beam stored over years (see Tube energy, below); it ended at Saturn, in Settlement space (see The Corridors). The array-pushed seed ships of the 2470s are the same kind of project, launched in the standoff (see Reaching outward).

Tube energy. By 2410 the largest array collects about 25–30 GW at Mercury, and a slower bloc about 10–25 GW. Emitters lag the mirrors, so only a few gigawatts of that can be beamed anywhere. A corridor-bore anchor draws about 100 GW, and the test tube needed one at each end for its eleven days: about 200 GW, or about 1.9×10¹⁷ J. So its builders stored the power. From about 2404 each anchor's store was filled by beam at about 1–3 GW, a programme every rival could watch filling for years, and the full stores held about 45 megatons' worth of energy: a demonstration of what an array could beam as much as of the physics. Fusion could have run the anchors; the race wanted the beam. Stores cannot hold a tube open for good, and a permanent tube would need its anchors fed for decades, by fusion or by beam. None runs in the solar system before 2497, for reasons of ownership and traffic rather than energy (see Tubes inside the solar system, in The Corridors). The Halcyon Breach (2418). A gatekeeper team smuggled replicator designs from the Mandate's Vaults to an Open City faction, then defected to the Open Cities: the defining security scandal of the age (see Why it mostly works, in Recursive self-improvement). The Open Cities' small Vaults are efficient but each is weaker than a large Vault, so their replicator designs had stayed behind the other blocs'. The designs went to one faction, not to the Open Cities as a whole: their Vaults belong to many different cities and owners, and a faction that owned one could put the stolen designs to use on its own. The Open Cities caught up on Halcyon's designs, and once they had designs worth sharing, they began sharing them with the Parishad. The Caloris Runaway (c. 2430). About fifty years in, an Open City seed in Mercury's Caloris Basin slipped supervision and built outside its licence, using a stockpile of smuggled licensed industrial controllers compatible with its stolen design. It was containable because the stockpile would run out, and terrifying because nobody knew how big it was. The Settlement and the Mandate each sent a fleet, each fearing a strike on its own seeds under cover of the operation. The Parishad cut further deliveries, and expansion stopped when the stockpile ran out. Existing machinery could continue operating, but further replication required more controllers. Adopting a simpler architecture would have required substantial redesign, retooling and interrupted production. It offered no immediate escape during the crisis. The two fleets spent the crisis watching each other as much as the seed. Caloris was a factory, not a mind. No superintelligence has ever officially been released; the known minds that came nearest, like the Vesper Runaway of 2311, were built without Vault engineers and stopped short of it (see Recursive self-improvement). Caloris is the race's crisis and the most quoted argument for every restraint since, including the rule that anchors at other stars be built by persons on site (see The Corridors).

From Caloris to the Accord. The crisis did not produce a treaty at once. The Ore Secession followed (2441–2449), sharpened by the Belt's fear that array power would undercut its fuel monopoly, and it stopped the Elevator for six years. The Elevator Protocol of 2449 made the Landing a joint zone, and in 2452 the Concord Accord seated its council there (see The long middle).

How the blocs race

The race is decided by how much supervision each bloc dares to drop. Mercury is split between the blocs, each working its own seeds and factories with small crews (see The solar system divided, in Reaching outward).

• The Settlement races slowest, under the most supervision, with the most rules and the most public argument. Under the Slow Lane every replicator runs under human signatures; the factories on Mercury come under Settlement-wide race rules rather than polity law, so no Raised mind may sign for them even where its polity lets Raised citizens sign elsewhere, and only human citizens may hold these offices, so its pace is set by how many qualified signatories it has, not by its chips. Its array is state-owned, and sales of its power help hold the Band (see Economy). It has fallen behind, to about 4–5 TW in 2497, and it is the most trusted. Mercury is the hardest place in the system for people, even adapted ones, so the Settlement's signatories mostly sign from shielded stations and orbit, while Raised crews live among the factories. That makes the factories the Meridianists' strongest case: the minds who know them best may not sign for them. The Remembrancers answer that this is the point: a mind that can live beside self-replicating machines with no human near is exactly what the human signature exists to check. How fast to go is the sharpest quarrel between its factions (see Factions in 2497, in Politics).
• The Mandate treats the race as a state project planned over decades, steered by its two Vaults' answers on its mirror programme. It is steady and large, and probably holds the biggest array in 2497, about 45–60 TW.
• The Open Cities cut corners on supervision and are the most dangerous racers. They moved first, on their small Vaults' designs, and their first mission killed hundreds of its own crew, most of them indentured Raised, in the Tolstoj Burn, around 2387 (see Who moved first, above). They re-entered, caught up on Halcyon's designs, and the Caloris Runaway was one of their seeds. Throughout the race they were the likeliest target for a first strike, and had every reason to take that risk. Since linking their small Vaults in the 2460s they have grown rapidly, to about 11–21 TW in 2497 (see The Open Cities, in Politics).
• The Parishad races on its own Vault's designs and the chips from its own fabs. It started its seeds alone in the 2400s. It has sold the Open Cities most of their chips since 2387, and after Halcyon the two began sharing designs; its array has grown faster since, to about 5–10 TW in 2497 (see The Parishad).
• The Belt Republic holds no seeds on Mercury, which the four blocs divide. It sits in the Anchorage for its ice, its fuel routes and its own Vault, and array power is slowly wearing that weight down.
• Freehold, the Raised bloc, holds no seeds either, but its crews work every bloc's factories, on the planet that is hardest on people. It takes its pay as a share of each array's power, under an arrangement with every bloc that holds one: in 2497 about 5% of the Settlement's, 4% of the Open Cities', 3% of the Parishad's and 2% of the Mandate's, about 1.7–2.6 TW in all. Each share is renegotiated in every crisis (see Freehold and the Raised cities, in Kinds of mind).

What the Vaults do

The Vaults started the race and design it, but no Vault builds an array.

• They speed up design, not building. A Vault can design a replicator in months. A Vault given the task of building an array would seize factories and resources, because those help with almost any task. So every bloc builds from Vault designs under the supervision of persons.
• Mirror designs stay secret. Answers for a bloc's own programmes are never published. Cleared Raised and human panels, independent of the programme, review the mirror designs instead of the archive panels (see How the Vaults work, in Recursive self-improvement). Unpublished designs earn no Vault dividend, so in the Settlement every mirror question in the queue displaces a design that would have helped pay for the floor.
• The question queue becomes a weapon. Whoever asks a Vault how to build an array, and how to stop a rival's, shapes national strategy. The fight over the queue is bitter inside every bloc. In the Settlement a majority of the Chambers sets it (see The Band, in Economy); in the Open Cities the queue is split among the cities and owners of their Vaults.
• The blind spot. Sealing Treaty inspectors checked that Vaults were sealed but never questioned them, so a bloc could run a secret array programme through its Vault without breaching the treaty. The Concord Accord absorbed the Sealing Treaty and kept the blind spot: each bloc's inspectors still sit among the others' gatekeepers and never put a question to their Vaults.

The Concord Accord (2452)

The Accord registers and watches the arrays; it does not limit them. Nobody proposed capping power by law: no society has ever capped its total energy use, and the pull of abundant energy would wear down any such law. The Accord binds the four blocs, the Belt Republic, which signed just after winning independence, and Freehold, the Raised bloc, and it absorbed the Sealing Treaty (see The Concord Accord, in Politics).

What it does.

• Seats its council, the Anchorage, at the Landing, with six seats: the Settlement, the Mandate, the Parishad, the Open Cities, the Belt Republic and Freehold.
• Through the Anchorage, registers and inspects arrays and their seeds, large emitters and relay stations.
• Splits the shutdown keys to every large emitter and every Vault between the owner and the Accord, so the Accord can cut any beam.
• Sets keep-out distances for emitters around inhabited places, forbids aiming any large emitter at Earth's surface (a rule the blocs had kept since the first beams; beams to Earth end at receivers in orbit), and tracks ships near habitats.
• Carries the Rock Clause. Any move of a body over 1,000 tonnes (at 20 km/s, about 50 kilotons on impact) must be filed with the Watch before it begins, no body may be steered onto an orbit that could strike a planet or moon without the Watch's clearance, and moves near habitats are filed and logged. The Watch is the Accord's planetary-defence body, crewed by all six signatories, with a sky survey no member can switch off alone and shared deflection tugs, which can move rocks but enforce nothing; it descends, in spirit and in its oath, from the Bennu Nudge of the 2120s. The clause answers the Ceres Letter.
• Punishes a signatory that refuses a Vault inspection: the others' compute caps rise, and its fabs lose the seals that let their chips pass between blocs. By 2497 every signatory but Freehold has at least one Vault, the Belt Republic included; Freehold officially has none, and the Anchorage inspects its compute and cores instead.

What it leaves out. The negotiators tried caps and dropped them. A capped array's factories can regrow it, hidden seeds cannot be found, and mirrors turned edge-on hide their capacity. Emitter caps fail for the reasons below, and emitters are also the outer system's power supply. So the Accord watches, and the peace rests on everyone being able to destroy everyone else.

Why it leaks

Any emitter that can deliver power can deliver it as a weapon, and the Accord can only report what it finds.

• Small emitters combine into big ones. A hundred 100 m emitters locked in phase across 10 km focus to as tight a spot as one 10 km emitter. Only about a hundredth of their power lands in it, but that is still a hundred times as intense as the same emitters aimed together without locking. The locking is software, easy to hide and quick to change.
• Wavelength matters as much as size. Ultraviolet light focuses about ten times tighter than infrared from the same emitter, and frequency-converting optics can be added after inspection.
• A "safe" beam is still a siege weapon. Habitats survive by shedding heat through radiator fins. A beam at five times Earth's sunlight, held for hours or days, overwhelms them. Cutting a beam a habitat depends on also kills it.
• Relays near habitats are short-range weapons. Anything that re-sends power close to a target can focus on it.
• Kinetic weapons are only half covered. The Rock Clause governs moving whole bodies, not what launchers throw: the cargo launchers that ship Mercury's goods outward can also throw rocks at very high speed, and nothing in the Accord counts them.
• Nothing enforces it. The Anchorage has no fleet, so inspection can only report a breach. Every serious crisis tests whether the blocs will act on the report.

The standoff (2452–2497)

Since 2452 the peace has rested on mutual vulnerability: every habitat is within reach of someone's sun, and everyone knows it.

• A weapon before an economy. Aiming a beam is easier than building emitters, receivers and radiators across the system, so the early arrays were mostly threat. Within each bloc, delivery lags only by that build-out and is catching up by 2497. Between blocs, beamed power barely flows, except for the share each array pays Freehold for its crews.
• An inequality between blocs. Each bloc's worlds live on their own bloc's array, so the gap between array sizes has become an economic inequality as well as a military one.
• Worlds keep their own fusion. Even within a bloc, no world gives up its reactors, because the Accord's split keys can cut any beam. Like Europe on Russian gas, a world that takes a beam gets cheaper power and hands someone a lever. Titan and the Belt guard their fusion as energy sovereignty.
• Mercury is the system's furnace. Energy-hungry industry has moved there and ships its products out, carrying energy in goods rather than beams. Whoever controls Mercury matters enormously.
• Abundance stays partial. Earth's waste-heat limit keeps the homeworld recognisable: about 2,000 TW warms it roughly as much as doubling CO₂. Licences, compute, brain renewal, transit times and positional goods stay scarce (see Where the wealth is, in Economy).

The Storm Protocol (2460). After the Mercury Storm of about 2457 the signatories added a protocol on solar storms. A ring of sentinel probes inside Mercury's orbit gives hours of warning; on warning, mirrors turn edge-on and Elevator climbers park in shielded bays; every habitat keeps a storm cellar; and Earth's grids hold spare transformers by treaty. Nobody knows the Sun's worst case, and the fear of a superflare larger than anything in the record is one of the few dangers every bloc takes equally seriously. A storm warning is also the one time every array may legally turn edge-on, which is how arrays hide their capacity, so inspectors dread the drills as much as the storms. What the standoff has seen. In the 2460s the Open Cities linked their small Vaults for the first time, using Vault engineers hired away from the Settlement and the Parishad. In the same decade the Settlement and the Mandate closed the Open Cities' new Martian city's routes to orbit in the Hellas Crisis, the standoff's sharpest quarrel off Earth (see Hellas, in Reaching outward). In the 2470s array-powered lasers pushed five light seed ships to about 9% of light speed: the Mandate's three, one Line ship and one Belt Republic ship (see Reaching outward). The same decade brought the standoff's closest call on Earth, the Brahmaputra Crisis, when the Mandate held back the river in a drought year and both blocs put their arsenals on alert (see The Parishad). In the 2480s the Vesta Crisis showed the Belt Republic's lever wearing thin: when the Parishad began hauling its own ice from beyond the Belt, the Republic cut off Vesta's ice and fuel, and the Anchorage brokered terms while Ceres lay under the Parishad's mirrors (see Vesta, in Reaching outward).

In 2497. The largest array, probably the Mandate's at about 45–60 TW, roughly matches all off-Earth fusion, about 60 TW. The Open Cities' collects about 11–21 TW, the Parishad's about 5–10 TW and the Settlement's about 4–5 TW. Growth is beginning to slow as inspection falls behind the fabs, so the doubling times above are averages over the race. Every bloc can still see full abundance coming beyond the end date, and every bloc fears whichever one breaks the industrial-controller supply limit first. The newest school of historians, the deterrence school, credits fear for the peace (see Who tells this history, in The long middle).