i went quiet for a few months. not because i ran out of things to say. because i was specifying. the kind of work that doesn't photograph well. a prompt architecture named after a character who wasn't on screen yet when i picked the name.
then Project Hail Mary came out. i went twice.
what follows is an MEng in Aeronautics & Astronautics and three years of debugging aircraft network protocols, sitting in a dark cinema, reacting to the engineering decisions of a fictional spacecraft.
note
the physical numbers here are Weir's, from the book: 96.415°C, the 25.984 micron , 1.5g under thrust. i checked they're consistent, i didn't invent them. the one figure i work out myself, the centrifuge spin rate, i show in full below.
the thing about Andy Weir
Weir picks one impossible thing. one. then he engineers everything else from reality.
The Martian: no impossible thing at all. one hostile fact, no resupply, everything else derived from it. the hab, the MAV, the potatoes, the comms delay, all textbook physics. the same discipline with the liberty dialled to zero.
Project Hail Mary: the impossible thing is , a microorganism that stores stellar energy. strip astrophage out and the ship's engineering maps onto real open problems in interstellar propulsion.
Artemis: one impossible thing, a city on the Moon that pays for itself. strip it out and the rest is derived. a currency backed by the gram-cost of a launch. a smelter as the reason the city exists. the chemistry of fire and pressure in a sealed dome. same method, lower orbit.
one impossible thing, everything else earned. you define the axiom, you work out the consequences, you don't cheat. accept astrophage, and the ship stops being designed and starts being derived: a , then a , then 1,009 of them in parallel, then the whole ship.
that's the part i can't put down. not the gadget. the discipline of it. pick the single liberty the story needs, then pay full price for everything downstream. it's the same reason The Martian holds: change one fact about the world, and refuse to change a second.
back to the spacecraft.
the propulsion system
astrophage absorbs stellar radiation and stores it as mass. it maintains 96.415°C in any environment. anything hotter, even the surface of a star, gets absorbed and banked as mass. thermal regulation as a physical constant.
the engines don't burn it. they feed astrophage into a chamber, heat it past a threshold, and the organism dumps its entire stored energy as light: infrared photons at the Petrova wavelength, 25.984 microns, the same wavelength it absorbed. not heat, not chemical bonds, not kinetic motion. light, directly.
a photon rocket. an actual photon rocket.
photon rockets are the theoretical ceiling for a reaction drive, and effectively impossible: the energy-to-thrust ratio is catastrophic unless your energy density is absurd. astrophage cheats nothing: it stored that energy from Tau Ceti's star to begin with. biologically implausible, physically consistent. Weir never breaks conservation of energy.
the fuel is also the battery is also the exhaust. one store with a fixed setpoint, a mass payload, and a known emission wavelength: a biological energy source with a byte layout. the joke writes itself, it has a wire format.
an organism with known storage, known emission wavelength, known thermal behaviour, and a ship designed around those exact numbers. the engineering follows from the biology. that's how real systems work.
1,009 spin drives
a spin drive runs a three-phase cycle. attract astrophage to coat one side of a transparent triangular slide using low-intensity light. rotate the coated face to space and crank the intensity, so the astrophage pushes off and imparts momentum. rotate back, scrape off the spent organism, repeat.
one thousand and nine of them, in parallel.
i spent years writing requirements for systems where single-point failure wasn't acceptable. redundant data buses, redundant power paths, redundant everything, because the aircraft cannot land itself if one box dies.
1,009 spin drives is that philosophy at scale. fail one. fail ten. fail a hundred. the mission continues. it isn't a number picked to sound big. it's what survivability looks like when you design for it.
the maths is real
two details say Weir did the work rather than gestured at it.
the gravity. the ship fakes gravity two ways, each solving a different problem. under thrust it accelerates at 1.5g and Grace feels the floor push up. the , 1907. that's how Grace knows he's on a spacecraft before he remembers anything else: the gravity is slightly too high to be Earth. when the engines cut and the ship coasts, it splits in two and spins like a centrifuge.
0.307 rad/s. twenty and a half seconds a turn. the number that says it works is easy. the number nobody quotes is 2.93 rpm, just over the line where your inner ear starts arguing with your eyes. the spec is real. it's also slightly nauseating. both things are true, which is the whole point.
held together by cables (polybenzoxazole, one of the strongest fibres per unit mass, the same material used in Mars-probe parachutes). someone did that maths and put it in a film. i tracked satellite orbits in MATLAB at 3am once, on StackOverflow and stubbornness. you can tell when someone has sat through that. Weir has.
the xenonite. the alien ship is built from solid xenon. xenon is a noble gas. full electron shells, doesn't want to bond, doesn't want to be a solid. under extreme conditions it forms rare compounds like XeF₂; extrapolates that into a structural material.
and it holds together. Rocky's species use xenonite for everything: ships, tools, the tunnel between the craft, Rocky's environment ball. it's their steel. and it's transparent to echolocation, which is how Rocky sees through it. Rocky has no eyes. he perceives through sound. his entire material science is built around a substance that is acoustically transparent to his senses.
the Eridians didn't find xenonite despite their biology. they found it because of it. the engineering and the organism are inseparable. same lesson as the astrophage, told from the other side.
things the post-credits drive didn't fix
the thermostat. astrophage maintains 96.415°C, always. if it cools, it spends its own mass-energy to climb back. a biological thermostat with a fixed setpoint. i spent time validating propeller control on a turboprop, blade pitch adjusting continuously to hold optimal RPM across every flight phase, and here is a fuel source that is its own thermal management system. i don't know whether that makes the engineering easier or harder.
the protocol with no shared anything. Grace works out that Eridians communicate by echolocation and bootstraps a translation system for Rocky's musical speech. two species, no shared language, no shared senses, no shared atmosphere, one breathes oxygen, one ammonia, and they build a working comms protocol from numbers, then shapes, then units, then science: an OSI handshake where the endpoints don't share layer 0. i once spent months on how two software systems exchange data reliably over an AFDX network. shared spec, shared tooling, and we still found edge cases that took three rounds of IRF revisions. Grace and Rocky did it with a whiteboard.
the requirement nobody wrote. the taumoebae, the organism meant to solve the problem, evolve the ability to pass through their xenonite containment and start eating the ship's astrophage fuel. the solution evolved past the spec.
a distributed-systems failure mode and a biological one and an engineering one at once. the organism passed a requirement that was never written down. i wrote a lot of requirements in my aerospace career. i never wrote that one, because no one thought they had to.
why it held
the science is correct and the orbital mechanics check out, but that's not the part i was still turning over on the drive home. it was the method.
Grace wakes with no memory, no context, no support. he doesn't panic. he debugs. drops a test tube, times the fall, measures the height, calculates local gravity. first reading 0.37 seconds, then 0.33, on down the line. twenty times, for an average.
twenty measurements. amnesiac. on a ship he doesn't recognise. because one data point isn't a measurement. you average. you reduce error. you never trust a single reading.
that's how people who do this work think. i spent three years writing Input Requirement Files for network-protocol verification. every packet accounted for, every test repeatable, every result traceable to a requirement.
Grace would have written good IRFs.
and then there's Eva Stratt
i'd already written about her, in the piece on why the name keeps both Ts (stratt.blue/journal/why-two-ts). i named STRATT after her before i watched the film.
Stratt runs Project Hail Mary from the ground. the administrator, not the scientist. one motivation: save the species from astrophage, let nothing stop her. she does things during the mission she fully expects to be jailed for. she's right. afterward they imprison her as a scapegoat. she breaks out and keeps working, a fugitive still fighting for the species that branded her a criminal.
no deviation. no apology. keep working whether or not anyone is watching.
that's why the system carries her name. execute, verify, continue. it doesn't care if you're watching.
coming back
i went quiet for a few months. i came back because i finished specifying something worth coming back for.
STRATT is a typed, versioned, cryptographically fingerprinted prompt architecture. every unit has a canonical address. every published unit has a fingerprint. every chain has gates that can't be bypassed. every change ships with a blast-radius report before anything merges.
the gate can't be bypassed, and what fails it doesn't error out. it gets pointed at Null0, the black-hole interface where chaos goes to die. the routing table stays honest because there's always somewhere to drop what won't verify.
named after a character who won't deviate when the whole world is watching, and won't deviate when no one is.
not to win. to keep building things that deserve to exist.
one impossible thing. everything else earned.