what did the bootloader load?

the day after the burial we opened the corpse — the full study lives at what the bootloader loaded. 2,949,732 cyberlinks over 3,143,650 particles — the complete graph, rebuilt from raw block events, bit-exact to the chain's own counters. this is the autopsy: what did five years of bostrom actually teach?

the shape of the effort

year	links
2021	8,251
2022	99,742
2023	396,049
2024	2,433,452
2025	12,170
2026	68

82.5% of the graph arrived in one year. and most of it from one place: of 1,240 linking neurons, a single one wrote 2,289,526 links — 77.6% of everything. the runners-up are wasm contracts: programmatic linkers, not humans

a purist reads this as centralization. look closer at what that neuron was doing before judging it

the degree structure

total degree	particles
1	2,524,571
2–9	586,101
10–99	31,370
100–999	1,454
1,000–9,999	137
10,000+	17

a textbook power law: eighty percent of particles are leaves touched exactly once, seventeen hubs hold ten thousand connections or more. scale-free before any focus weighting is applied — the topology perron-frobenius-theorem needs for a unique attention distribution, delivered by the data itself

where focus converges

we ran the ranking over the full graph — the same pagerank-family iteration cyberank generalizes — and then read the winning particles straight out of the cybernode blockstore. the top of five years of collective attention, verbatim:

description · wiki · class · image · glossary ·
Gregorian calendar · asteroid · dog · spell ·
species of plant · species of bird · human settlement ·
german politician · company · film description · map of medicine

these are not memes and not spam. they are category words — the skeleton of a knowledge ontology. the giant neuron was an archivist: it poured the semantic backbone of wikidata into the graph, entity by entity, each linked to its type. random particles confirm the genre: fragments about Warhol's studio, the Art Institute of Chicago, "british army general", "mountaineer"

the bootloader loaded an encyclopedia — as a graph

how heavy is a mind's foundation

we weighed every available particle in the cybernode blockstore, file by file:

measure	value
particle blocks on disk	1.74 GB
full content with child blocks (sampled)	≈ 6 GB
median particle	30 bytes
90th percentile	749 bytes
largest particles	≈ 1 MB (images)
content types (2,500-particle sample)	94.9% text · 0.5% images · traces of json, html

the median particle is thirty bytes — a word, a name, a phrase. this graph does not store documents; it stores atoms of meaning and lets cyberlinks carry the structure. that is why five years of collective knowledge weighs less than a single 4K movie: the knowledge is in the links, and links are nearly free. 2.9 million edges over 6 GB of substance is the inversion the web never managed — structure heavier than content

what the humans linked

subtract the archivist and the contracts, and a second graph appears: 400,776 links, hand-made by 1,239 neurons, growing every single year of the chain's life — 7K, 49K, 162K, 170K — no decay until the halt itself. read its top particles:

noun · verb · adjective · author · artificial intelligence ·
cyber · Gregorian calendar · 2013 · 2014 · 2016 ·
Table of nuclides · Universal Time — Planetary Chart

the bot built the encyclopedia; the humans built the grammar. parts of speech, authorship, time itself, the periodic structure of matter — and artificial intelligence and cyber near the top, the graph quietly linking toward its own purpose. this hand-made layer is the curriculum: what 1,239 minds thought a new mind should learn first

what to do with five years of collective effort

the corpus is small enough to fit in RAM and structured enough to mean something. the research agenda it opens:

initialize the soft3 genesis with the ontology: the category skeleton becomes the prior for focus — the new chain starts knowing what kinds of things exist
treat the 2.3M entity-to-category links as labelled data: a free supervision set for grounding soma and the compiled-transformer line in graph-native embeddings
run the full tri-kernel on it — diffusion, springs, heat — as the first real-scale calibration of the collective focus theorem against a graph with known provenance
diff against wikidata-2026: what the archivist did not load is a map of the frontier — the knowledge the graph still lacks
make it the robot's memory: 6 GB of addressed meaning is a local-first retrieval substrate for cyb — the robot answers from particles it can prove
study the human layer alone: 400K curated links is a dataset of attention itself — what people link when linking costs money is a revealed-preference record of what matters
recover the missing 74,513 and close the archive at 100% — then seal it: the first complete, audited, portable knowledge graph a chain ever produced

five years, one chain, one relentless archivist, and the verdict of the eigenvector: what the bootloader loaded was the structure of human knowledge itself. now we teach it to think

focus is not popularity

the particle with the most incoming links is wiki (30,029). the particle with the highest converged rank is image (17,608 links). rank flows through structure: a hub whose neighbours are themselves central outranks a bigger hub on the periphery. this gap between degree and focus is the whole reason eigenvector attention exists — counting links is not understanding them, which is what cyberank and the collective focus theorem formalize

what this means for genesis

the cyberlinks port into the soft3 genesis carrying this structure with them. the new mind wakes up owning:

a scale-free semantic backbone — categories, types, glossaries — the hardest part of an ontology to bootstrap
47,837 moon-passport names bound to keys
and the empirical proof, audited, that the substrate holds: 97.63% of those particles still servable from the cybernode

bootloader · bostrom · cybergraph · focus · soft3