# The seed What has to be inside the unit for it to reproduce itself, and what has to be stockpiled instead because no bootstrap can make it. This is the document that turns "and then we build a fab" into named steps with masses and durations. Read with `logic-element.md` and `core-memory.md`, which define the parts that actually have to be reproduced. ## Precedent The concept is not new and has been studied at programme scale. The 1980 NASA and ASEE summer study, published as NASA Conference Publication 2255, *Advanced Automation for Space Missions*, devotes a long chapter to a self-replicating lunar factory. The related systems study is von Tiesenhausen and Darbro, NASA TM-78304, 1980. The shape of the conclusion matters more than the numbers: feasible in principle with then-current technology, seed mass on the order of a hundred tonnes, replication measured in decades, and closure never total. A few percent of the mass always had to be imported, and the shortfall was precisely the complex parts. VERIFY: the hundred tonnes, the decade-scale replication time and the size of the closure shortfall are from memory of the summary literature, not from the chapter itself. Read CP-2255 chapter 5 and TM-78304 before quoting any of them. The qualitative conclusion is the load-bearing part and is not in doubt. The design lesson is the second half: do not aim for complete closure. Aim for a closure table where every imported item is named, counted and stockpiled. ## The two columns Every item in the machine belongs in exactly one column. Deciding which is the whole engineering effort. | item | column | note | | --- | --- | --- | | ferrite cores | make | iron oxide plus Mn/Zn oxide, mill, press, fire near 1300 C. Pottery plus chemistry | | copper wire | make | drawing dies and a drawbench, given copper | | diodes | make | point contact or grown junction, 1950s bench practice, low yield at first | | coarse boards | make | single and double sided, no fine lines | | mechanics | make | lathe, mill, castings, bearings | | glass, ceramics | make | kiln work | | kiln, press, mill, drawbench, winder | make | generation two, built by generation one | | ceramic permanent magnets | make | barium ferrite is within reach | | rare earth magnets | stockpile | needs the whole separation industry | | silicon ROM | stockpile | the one part with no bootstrap path | | precision voltage, frequency, length references | stockpile | small, cheap, durable | | photoresist, ultrapure acids, electronic gases | stockpile | only needed if chasing silicon | | optical sensors, good transistors | stockpile early | makeable later at poor yield | ## Why a 1958 machine is the right target The anachronism is the feature. A magnetic ternary machine needs no photolithography. Ferrite is fired oxide, wire is drawn, diodes are 1950s bench parts, and the logic is wound. The entire machine sits inside a 1970 electronics capability. Had it needed 28 nm CMOS, the seed would need a civilisation rather than a workshop. This is the strongest argument for the architecture that has nothing to do with elegance or speed. It is the only architecture in this lineage that is reproducible by the thing it reproduces. ## Calibrating the twenty years The relevant data point is the home fab work of the last decade: a determined individual, a garage, scavenged equipment, a few years, working devices at around 1 um. That is 1970s silicon at person scale. So twenty years of concerted effort buys roughly 1970, not 2026. That is enough, because the machine is a 1958 design. Any plan that quietly assumes the seed reaches modern nodes has substituted a civilisation for a workshop and the timeline is fiction. ## The two dependencies that actually break these plans Neither is machine tools. **The purity cascade.** Electronic grade chemistry requires electronic grade chemistry. It is circular. The only way through is to accept low grade early, bootstrap in steps, and eat the yield loss and the time. Budget for it explicitly, in years. **The metrology chain.** You cannot make a one percent part without measuring one percent. The seed needs measurement before it needs manufacturing: length from gauge blocks, mass from a reference, temperature from fixed points, voltage from a stockpiled standard, frequency from a stockpiled resonator. Stockpile the small precise things, manufacture the large approximate things. For this project specifically, the load bearing instrument is the BH tracer, not the winder. Cores can be made approximately. Whether a core is acceptable is defined by measurement, so the tester is the machine that gates everything. ## Generation depth The right metric is tool generations from ore to machine, not percentage of self-made parts. RepRap's "half the parts" claim was true and useless, because the missing half was the motors, electronics and bearings. Sketch to be filled in with masses, labour hours and yields: 0. ore, oxide, copper, clay 1. kiln, press, ball mill, drawbench, lathe 2. cores, wire, diodes, boards, castings 3. cells, wound assemblies, wiring 4. first machine, tested against the acceptance tests 5. tracer, winder, second generation tooling 6. second machine, built by the first generation of tools Each arrow has a yield. The plan is credible only when every arrow has a number. ## The ROM circularity The instructions live in the ROM, and the ROM needs a fab. That is a genuine logical circle and it has to be broken deliberately, not hoped away. Break it three ways, same content, hash linked: - silicon ROM, stockpiled in quantity, hermetic, machine readable - printed and etched copy, human readable, no reader required - the calibration manifest and checksums in the core array itself The rule from the storage discussion holds and is now a requirement rather than a nicety: the archive must be readable without the machine. ## The four documents that are the seed Everything else, schematics included, is downstream of these. 1. The closure table above, complete, every import named and counted. 2. The generation chain, with mass, labour and yield per step. 3. The metrology chain, with the stockpiled references listed. 4. The acceptance tests: what makes a core, a diode and a cell good enough, stated as measurements with tolerances, so a builder with no predecessor can tell whether they have succeeded. ## Energy A seed cannot make a photovoltaic panel. Silicon, ultrapure chemistry, and the encapsulation are all outside the closure. It can rebuild a water wheel, a turbine, a windmill and a generator, because those are metalwork. So the site choice is a seed parameter, not a lifestyle preference. A site with existing hydro and a grid, such as the Naryn river cascade, gives the seed power it can maintain and repair with its own output. A site with only PV gives it a power source that degrades and cannot be replaced, which is a twenty year clock on the whole project. ## What done looks like A unit that can be rebuilt by people who never met the builders, using only what is in it, plus a named and stockpiled list of imports, with tests that tell them when they are finished. ## Open questions - Exact CP-2255 figures, as above. - The lowest technology diode that still works in the cell. This decides whether the seed makes its own semiconductors or stockpiles them for a generation. - Geology at the chosen site: iron oxide, copper, clay, limestone, fuel. - Labour hours per generation. Unknown and probably the dominant term. - Whether the seed should include a machine tool set at all, or assume the site has one and start from there. ## Sources - NASA Conference Publication 2255, *Advanced Automation for Space Missions*, 1980, chapter 5. - von Tiesenhausen and Darbro, *Self-Replicating Systems: A Systems Engineering Approach*, NASA TM-78304, 1980. - Setun book chapters VII and X for the parts that must be reproduced.