{"path":"research/major-transitions-heredity-and-evolvability.md","content":"# Major Transitions, Heredity and Evolvability — the Szathmáry episode read against the corpus\n\n**Date**: 2026-09-29 · **Type**: source intake plus synthesis, founder-commissioned · **Status**:\nresearch. Every design proposal below is **unruled** and filed as ledger Arc S\n([session26-open-rulings.md](session26-open-rulings.md)). Nothing is stored in the graph: the\nfounder's 2026-08-30 hold on self-hosted fractal priors applies.\n\n**Prompted by** (founder, verbatim): *\"This podcast about major evolutionary transitions and AGI (get\na transcript and read it carefully) connects abstractly to different questions both inside Deliberus\nargumentation graph and questions about why Deliberus should exist and why it can only be built now\nthat the combinatorial elements to put it together are in place (and I'm in a place in my life where\nI'm better equipped and able than ever to build/co-create it with other people) etc - think deeply\nabout how this relates to all that the docs corpus has discussed previously about Assembly Theory and\nthe questions of fractal priors, fractal priors for success etc. Also look at the analogy daemon\nidea/s and relate that to all this.\"*\n\n**Source**: *The Trajectory*, \"Stewarding the Flame\" episode 7, Daniel Faggella with **Eörs\nSzathmáry** (co-author with John Maynard Smith of *The Major Transitions in Evolution*, 1995),\npublished 2026-09-11, 80 minutes. Transcript from YouTube's automatic captions, held privately;\nquotes below are checked against it and lightly cleaned of caption errors, with timestamps. The\ncompanion paper is Müller, Steels & Szathmáry, \"Evolvable AI: Threats of a new major transition in\nevolution\", *PNAS* 2026.\n\n**Read depth**: the transcript in full. The PNAS abstract, its major-transitions section and its\nconclusion (via PMC), not the full text. Szathmáry's 2015 \"Toward major evolutionary transitions\ntheory 2.0\" (*PNAS*): abstract, revised table, and the egalitarian–fraternal section. Abstracts and\nkey passages only for Lewis & Laland 2012, Szathmáry & Demeter 1987, Szathmáry 2000 and 2006, Herron\n2021, the Stockholm Paradigm, the social-protocell papers, and the germinal-centre reviews. Corpus\ndocs named below were read in full this session unless marked.\n\n---\n\n## 0. The answer first\n\nThe corpus already used the major-transitions frame on one axis: transitions change how information\nis stored and passed on, and Deliberus targets that variable for reasoning\n([complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md) § 1,\n[convictions-across-scales.md](convictions-across-scales.md) § 2.3). It also had biology's four\nanswers to defection and the three adversary classes\n([fractal-scales-and-temporal-frame.md](fractal-scales-and-temporal-frame.md)). The episode adds\nfive things the corpus had not reached:\n\n1. **What heredity is.** Maynard Smith's flame burns and spreads but cannot tell you how it was lit.\n   An argument passed on as prose is a flame; a claim in the graph is a lineage. In Szathmáry's own\n   replicator vocabulary, Deliberus moves reasoning from *phenotypic* to *genotypic* inheritance\n   (§ 2).\n2. **Why long arguments die in retelling.** The length of reasoning a culture can keep intact is\n   capped by how faithfully each retelling copies it: Eigen's error threshold, with a measured\n   cultural twin. Language models raised the error rate of public reasoning and made a proofreading\n   substrate cheap in the same years, which is the mechanism under the corpus's \"the need arrived on\n   the same clock as the capability\" (§ 3).\n3. **Which kind of union Deliberus is.** Transitions come by two routes: like partners staying\n   together, or unlike partners joining. Deliberus is the unlike-partner kind twice over, and for\n   that kind the conflict control is **fairness in reproduction, not kinship**, which is the family\n   of mechanisms the corpus already built (§ 4).\n4. **The half to decline.** The frame has a second axis, a new level of individuality, bought with\n   the lower level's autonomy. Szathmáry himself declines to force human culture into the frame, and\n   the founder declined the \"hive mind\" in his own words. One corpus line overstated this and is\n   narrowed (§ 5). ⚠ **Corrected 2026-10-10: he had not declined it.** What he declined was one\n   platform taking the place of the bonds people already have; he is open to a hive mind in a\n   positive sense, and § 5b says which independence that keeps and which it gives up (ruled 2026-10-10, ledger S8: *\"Yes!\"*).\n5. **Evolvability, as an adversary and as a design question.** The episode's central claim, that\n   evolvability rather than intelligence is the danger, names an adversary kind the three-class\n   taxonomy lacks: adaptation without a mind (§ 6). The PNAS paper's split between *breeder* and\n   *ecosystem* regimes gives the machine passes one design question, and reading the code shows the\n   answer is only partly in place (§ 7).\n\n§ 8 relates all of it to the analogy daemon; § 9 answers the three founder questions (why exist, why\nnow, why this shape); § 10 grades the transfers and says what does not transfer.\n\n---\n\n## 1. What the episode says, in his words\n\n**Two things happen together in a major transition.** *\"A higher level unit of evolution is being\nformed from lower level ones … [and] new ways of storing, using and transmitting hereditary\ninformation arise, and usually they go hand in hand.\"* [00:03:15]\n\n**Two routes.** Following David Queller: *egalitarian* transitions, where *\"the things that are\ncoming together are unrelated, like in the case of the eukaryotic cell\"*, and *fraternal* ones,\nwhere *\"a cell divides and they just stick together\"*. Human societies with language were *\"our last\ntransition, which arguably has a genetic background as well; that's kind of a mix.\"* [00:05:43]\n\n**The units.** Multiplication, inheritance (*\"like begets like\"*) and variability (*\"heredity is not\nexact\"*); and *\"if you satisfy these conditions, then be it the universe, be it cultural items, be it\nmolecules or cells, the Darwinian type of evolution can unfold.\"* [00:08:08–00:10:38]\n\n**Life, operationally.** *\"What is life\"* in block letters is philosophy; the scientific question\nis what constitutes a living system, and the minimum is Tibor Gánti's **chemoton**: three\nautocatalytic subsystems (metabolism, rudimentary heredity, a containing membrane that must grow as\nthe others grow) which, *\"once you couple these three elements together in the right way\"*, make a\nsystem that is autocatalytic *\"at the level of the whole system now\"*. Gánti's chemoton, Kauffman's\nautocatalytic sets and Eigen's hypercycle all date from 1971: *\"there was something in the tide.\"*\n[00:17:58–00:21:13]\n\n**The flame.** Maynard Smith's lecture: *\"a flame does have metabolism, a flame can reproduce … the\nflame doesn't have inheritance … you cannot reconstruct anymore, by looking at the structure of the\nflame, how it originated.\"* [00:26:10] A tease aimed at the series' own \"flame\" metaphor.\n\n**Evolvability, not intelligence.** *\"Being so afraid about artificial general intelligence or\nconsciousness of AI, that's more of a hype … the real threat is the evolvability.\"* The lancet liver\nfluke makes an ant climb a grass stalk to be eaten by a cow: *\"this ridiculously simple creature, to\nwhich in comparison an ant is like Einstein … can force a much more intelligent creature to do what\nactually maximizes its own fitness.\"* So *\"you don't have to have a complete LLM to be an\nevolutionary unit. It's enough if there are other forms of AI that actually take advantage of an\nLLM.\"* And *\"once the Darwinian dynamics unfolds, selfishness is an algorithmic consequence.\"*\n[00:27:47–00:31:57]\n\n**Levels and constraint.** *\"A higher level organization always suppresses the evolutionary\npotential at the lower level. Otherwise you would have cancer all the time … Stu Kauffman would say\nthere are enabling constraints.\"* [01:01:08]\n\n**Culture resists the frame.** Early human groups may have acted *\"very much like something like a\ncell\"* (the social-protocell model, by Claes Andersson of Chalmers, the *\"guy in Sweden\"*), but\n*\"cultural evolution is far too complicated. You cannot squeeze it into this Procrustean bed.\"*\n[00:14:41]\n\n**What he is sure of.** Global problems *\"may not have a solution at all … but if there is one … it\nneeds tight international collaboration. There will be either a collaborative solution or there\nwon't be any\"* [00:59:28]; and *\"if there are solutions about which everybody or most of the\nplayers agree, … several of those developments will come from completely unexpected corners\"*\n[01:03:36]. On mitigation, borrowing the Stockholm Paradigm from emerging-disease work: *\"it's never\nabsolutely safe … but it matters a lot whether a disaster happens every day or … every second\nyear.\"* [01:05:14]\n\n---\n\n## 2. Heredity is history you can read back out of the structure\n\nThe flame has two of life's three properties. What it lacks is not complexity or activity: it is\nthat **its present structure carries no record of its origin**. That is what inheritance means at\nits most stripped-down.\n\n**An argument passed on by retelling is a flame.** It keeps burning (people keep repeating it) and\nit spreads, but its structure no longer tells anyone what held it up or where it came from. That is\nexactly the condition under which straw-manning, flattening and misattribution cost nothing: there\nis no lineage to check the rendering against. **A claim in the graph is a lineage**: extraction\nprovenance, attribution, the decomposition tree beneath it, supersession above it.\n\n**Szathmáry's own replicator classes make this precise** (Szathmáry 2000, \"The evolution of\nreplicators\"). *Phenotypic* replicators pass on the product, not what generates it; *genotypic* ones\ncopy the generator and rebuild the product from it. *Holistic* replicators copy as wholes; *modular*\nones copy discrete, recombinable parts. His verdict on culture: **\"Memes in human culture are\nunlimited hereditary, phenotypic replicators, based on language\"**, and **\"the typical path of\nevolution goes from limited to unlimited heredity, and from attractor-based to modular (digital)\nreplicators.\"**\n\nMap it. In ordinary discourse, conclusions travel (the phenotype) and the reasoning that produced\nthem is re-derived by each reader or lost. The graph stores the generator (premises and typed links)\nand recomputes a claim's standing from them on every read, the way development rebuilds a phenotype\nfrom a genotype; change a premise and the conclusion's standing changes through the recomputation,\nnot by someone editing the conclusion. The essay copies as a whole; the graph copies by part.\n**Deliberus moves reasoning from phenotypic to genotypic inheritance, and from holistic to modular.**\n\nThis is the missing-layer taxonomy ([the-missing-layer.md](../the-missing-layer.md)) in\nevolutionary vocabulary: civilisation has infrastructure for **conclusions**, which is phenotypic\nheredity, and none for **reasoning itself**, which would be genotypic.\n\n**Three independent formalisms meet at one property here.** Maynard Smith's heredity (structure\ncarries origin), assembly theory's criterion (a high-assembly object *\"carries its own construction\nhistory\"*, [assembly-theory-and-the-reuse-mechanism.md](assembly-theory-and-the-reuse-mechanism.md)\n§ 2), and the corpus's provenance discipline. The consequence is a design rule that the corpus\nalready follows in three places without the reason stated:\n\n> **Provenance is what makes an argument heritable, so every surface that drops it turns a lineage\n> back into a flame.**\n\nSummaries without citations, unlabelled counts and paraphrase without a link are the places it\nhappens. The three existing rules it explains: *every published count labels its provenance*,\n*always mint reported speech* (straw-manning becomes a comparison between two lineages), and\n*evidence routes through links, not copies*.\n\n**Four checks.** Recurrence: the phenotypic-to-genotypic and holistic-to-modular path is\nSzathmáry's own generalisation across replicators. Mechanism identity: strong at the functional\nlevel (store the generator, recompute the product). Adversary: at our rung the genotype itself can\nbe forged, since a plausible but fabricated graph is structured misinformation, which DNA does not\nface in that form. Substrate condition: modular heredity needs discrete units whose identity is cheap\nto decide; chemistry hands biology its nucleotides for free, while here identity is judged\n([the-reuse-flywheel-prior.md](the-reuse-flywheel-prior.md)). **Verdict: Raised as a description of\nwhat the design does; Conditional on claim identity.**\n\n---\n\n## 3. The fidelity threshold: why long arguments die in retelling\n\n**Eigen's error threshold** (1971, cited in the episode's notes): the length of information a\nreplicating system can keep intact is limited by its copying error per unit; beyond the threshold\nthe information melts into a cloud of variants. Szathmáry on the earliest replicators: they *\"must\nhave been rather short molecules owing to excessive noise in their copying\"* (2006). The bootstrap\nproblem (error correction needs a long genome to encode it, a long genome needs error correction to\nsurvive) is what Szathmáry & Demeter (1987) call **Eigen's paradox**.\n\n**The cultural twin was measured.** Lewis & Laland (2012) modelled cumulative culture and found\ntransmission fidelity to be *\"the key driver\"*: the trait-loss rate alone explained **75 per cent**\nof the variance in how much culture accumulated, a threshold separated no accumulation from\naccumulation, and *\"progress in cumulative culture depends more on trait combination than novel\ninvention.\"* Combination over invention is the reuse flywheel's claim arriving from cultural\nevolution.\n\n**Three things follow for the corpus.**\n\n**The slogan is the short molecule of a noisy channel.** What survives unaided retelling is short,\nbecause every retelling mutates it. The assembly doc's *\"unaided discourse has a low assembly\nceiling\"* is this threshold, now with its mechanism named.\n\n**The Gabora caution gets a mechanism.** The corpus holds the Acheulean hand axe as its sharpest\nwarning: a genuine cognitive transition followed by a million years of stasis\n([complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md) § 1). One standard\nreading of that stasis is a channel below the fidelity threshold: invention happens, and nothing\naccumulates. Held as a reading, not a finding, it converts the warning into a requirement: what\nkeeps a new space from compounding into nothing is fidelity, not more invention. Szathmáry's 2015\nrevision says the same about transitions generally: *\"establishment of such units requires high\nfidelity of reproduction (as opposed to mere replication).\"*\n\n**Language models pushed the error rate and the proofreading in the same years.** Summarisation\nraised the mutation rate of public reasoning: argument collapse toward the consensus middle, tails\nlost, facilitation that steers while being preferred (the threat model's entries 2 and 3d). Extraction\nmade a substrate that copies by reference cheap. That is the mechanism under the corpus's *\"the need\narrived on the same clock as the capability\"* (complexity-transitions § 2b): not a coincidence of\nschedules but one technology moving both sides of the error threshold.\n\n**Two consequences, both proposals.**\n\n> **Fidelity is the threshold variable, not one criterion among six.**\n\nThe shared-code criteria in [convictions-across-scales.md](convictions-across-scales.md) § 2.1 list\n*stability under transmission* as the fifth of six. The culture model says it alone decides whether\nanything accumulates. So the **disagreement-preservation instrument is not only a flattening alarm;\nit is the project's fidelity meter**, and today it measures only one hop.\n\n**A pre-registered prediction.** Drift compounds with each decomposition hop, so for a given per-hop\nfidelity there is a depth beyond which the leaves no longer rebuild the root. The instruments exist\nin pieces: the multi-hop flattening check parked in\n[grounded-critics-and-quarantined-imagination.md](grounded-critics-and-quarantined-imagination.md)\nand the recomposition check that `decompose_claim` already runs. **Refuted if** coverage holds flat as\ndepth grows.\n\n**The project's own inheritance is a measured instance.** On 2026-08-29, 6 of 31 ruled ledger rows\ncarried the founder's words and 25 were restatements; later sessions build on the restatement. The\nverbatim-ruling rule is proofreading, and *a correction is not done until every surface that repeats\nit is corrected* is the rule that stops errors propagating once made. The corpus has been running\nerror-threshold control on itself without the name.\n\n**Four checks.** Recurrence: genes, culture, this corpus. Mechanism identity: strong (loss per copy\ncompounds with length). Adversary: at our rung some mutation is deliberate (paraphrase as spin),\nwhich is strategy-class. Substrate: needs a copying channel whose errors are detectable; the graph\ncopies by reference, so error enters once, at extraction, rather than at every retelling, which is\nwhere the instruments sit. **Verdict: Raised.**\n\n---\n\n## 4. Two routes into a larger whole, and which one Deliberus is on\n\nIn Queller's terms, as Szathmáry states them in the 2015 revision: in *fraternal* transitions *\"like\nunits join or remain joined\"*; in *egalitarian* ones *\"unlike units come together, complementing\ntheir functions in a higher unit.\"* And the line that matters here: **\"The main control of conflicts\nis ensured by kinship and fairness in reproduction for the fraternal and egalitarian transitions,\nrespectively.\"** In egalitarian unions *\"particle fitness values cannot go down to zero, but they\nneed to be tightly controlled through the mediation of conflicts (reproductive leveling)\"*, and\n*\"no reproductive division of labor can exist\"*: no partner becomes a sterile caste.\n\n(A collision note: *egalitarian* already appears in this corpus as a political value. Here it only\never means Queller's route, *unlike partners joining*, and is always written with that gloss.)\n\n**Deliberus is the unlike-partner kind twice over.** Minds holding different positions, where the\ngroup-premise evidence says the fifth condition, *that the minds actually differ*, must be selected\nfor ([the-group-premise-evidence.md](the-group-premise-evidence.md)). And humans with machines, whose\ncomplementary partition the harness conviction describes\n([the-harness-conviction.md](the-harness-conviction.md)).\n\n**So of biology's four answers, one is off the table as the core.** Relatedness is the fraternal\nroute's answer. For unlike partners the governing answer is fairness in reproduction, and the corpus\nbuilt that family without knowing it was choosing a route: *contributions are added, never\noverwritten* (no partner's persistence goes to zero), payoff relocation and the weakest-part rules\n(reproductive leveling: no position gains copies by gaming), sole authorship of one's own stance.\nAnd \"no sterile caste\" is the reason the ratify-to-publish gate was right to rule out: it would have\ncreated a class of contributors whose positions could not persist without permission.\n\n**The workshop matching rule is a kinship device inside an unlike-partner union.** Shared\nmeta-curiosity makes the union easier to hold, which is fine as scaffolding. The warning is the\nroute itself: recruitment that drifts toward like minds takes the fraternal path, where cohesion is\ncheap and the product is a sect ([kulturkapital-and-the-unscalable.md](kulturkapital-and-the-unscalable.md)\n§ 3). The test stays the group-premise one: the minds must differ *on the thing in question*.\n\n**The human–machine union has a precedent, and it is the eukaryote.** The one egalitarian\ntransition that produced complex life stabilised by the host coming to control the partner's\ninheritance: the revised table lists *\"transfer of plastid genes to nucleus\"* and host-regulated\ndivision among the phases, and in most eukaryotes organelles are inherited from one parent only. The mapping: **machine\noutput becomes heritable only through human-governed paths.** The corpus already has this as\ngermline sequestration (fractal-scales); the route explains why it is load-bearing for *this kind*\nof union rather than one option among four.\n\n**Verdict: Raised on the mechanism family; Conditional on recruiting real difference.**\n\n---\n\n## 5. The half Deliberus declines\n\nThe frame has two axes, and Szathmáry's 2015 revision keeps both: *\"a 2D account (evolution of\ninformational systems and transitions in individuality) remains.\"* The corpus took the information\naxis only. The other axis is the formation of a new individual, and it has a price the episode\nstates twice. Szathmáry: *\"a higher level organization always suppresses the evolutionary potential\nat the lower level.\"* Faggella, closing: the cells of his thumb *\"are not really maintaining their\nown pattern unless they be cancer … there is some giving up of freedom.\"*\n\n**Asked directly whether human coordination is such a transition, Szathmáry declines.** He grants a\npossible early *social protocell* phase (Andersson & Törnberg's model: growing, splitting\ncommunities as the compartments of cultural inheritance), and then: *\"you cannot squeeze it into\nthis Procrustean bed.\"* The field's stricter definition is harsher still: Herron (2021) argues that\nmajor transitions form a coherent category only if restricted to new levels of individuality, so\nthat *\"the origins of the genetic code and of human language … should be excluded.\"* By that\ndefinition, what Deliberus targets is not a major transition at all.\n\n**The founder already said where he stands** (⚠ 2026-10-10: this paragraph over-read him, and the\nrule under it is withdrawn in favour of § 5b) (session 19, verbatim): *\"Companies and countries and\nfamilies are bound together and I don't necessarily wanna split them up and have everyone join the\nhive mind of Deliberus hehe… but we can at least facilitate reflection about what ideological house\nof cards we prop up each day by our actions and words…\"* The frame supplies the reason behind the\ninstinct:\n\n> **Take the inheritance half of a major transition, and decline the half where members give up\n> their autonomy to the whole.**\n\nThe rulings that already hold this line: sole authorship of one's stance, contributions never\noverwritten, and no verdict acts. The community-compartment design (unruled) adds a fourth: the right\nto dissolve a compartment outranks its permanence. The 2025 formulation *\"make humanity more hive mindey\"* reads, on this lens, as the\ninheritance half (shared, persistent, checkable reasoning), and which half it meant is his call.\n\n**Correction owed and made.** [fractal-priors-for-success.md](fractal-priors-for-success.md) § 9\nsays *\"human collective cognition at internet scale is visibly mid-transition\"*. That asserts an\nindividuality-axis transition in progress, which is the squeeze the frame's co-author declines. It\nnow carries a dated narrowing to the information axis (grepped 2026-09-29: the phrase appears on no\nother surface).\n\n**The social protocell has a second lesson, and it cuts the other way.** Group-level cultural\ninheritance can carry function that no member understands. In the 2023 *Trajectory B* paper, where\nthe *sociont* is the community treated as a cultural individual: *\"The sociont possesses institution\nI to exhibit function F, even if neither the workings of I nor the nature of F is understood by any\nof the community members.\"* That is an evolutionary account of\n**load-bearing illegibility**, the gap the red team named under constructive ambiguity. It means\nculture runs a second inheritance channel (practice-level, illegible, selected at the group level)\nbeside the one Deliberus builds (argument-level, legible). Success condition (d), *map organic\npractice rather than replace it*, extends accordingly: some organic practice works because nobody\ncan state it, and making it legible may break it, or may for the first time let it be improved on\npurpose. Which of the two happens is not decidable in advance, and is the kind of thing a community\ncompartment can try first.\n\n---\n\n## 5b. Which independence (2026-10-10, on the founder's comment)\n\n**His words**, on reading that § 5 had filed him as declining the hive mind:\n\n> \"I guess that makes sense, although before we've been through such a shift, we honestly don't know\n> how it would affect the members/cells in the new level of\n> organization/abstraction/collaboration/coordination, do we? I'm open to humanity becoming more like\n> a hive mind in a positive sense, think I've mentioned this before in the Deliberus corpus. But ofc\n> lack of independence has its downsides. Think deeply about this.\"\n\n**The correction.** § 5 said *the founder declined the \"hive mind\" in his own words*. That over-read\none sentence. What he declined in session 19 was splitting up companies, countries and families and\nhaving everyone join *the hive mind of Deliberus*: one platform taking the place of the bonds people\nalready have. He had said the opposite of declining twice before: the 2025 line § 5 quotes (*\"make\nhumanity more hive mindey\"*), and an introduction he wrote in April 2026 (kept privately, in Swedish),\nwhich says humanity must become more of a hive mind in a positive sense and gives the sense: working\ntogether toward a shared understanding of how the world works, with disagreements laid out as finely\nas possible. § 5 ended *\"which half it meant is his call.\"* This is the call, and it is not the half\n§ 5 chose for him.\n\n**He is right that nobody knows in advance, and one biologist says the same from the other side.**\nStearns (2007) proposes that humans are stalled part way through a transition from individuals to\ngroups, one that may never be completed, and that if it were, *\"our individuality would be submerged\nin a group superorganism.\"* His examples of stages further along are Plato's Republic, Orwell's 1984\nand closed religious communities, all of which he calls unstable. So § 5's sentence that what Deliberus\ntargets *is not a major transition at all* holds only under the strictest definition. On his reading\nthe question is live. And what keeps the transition stalled is on his list: people belong to several\ngroups at once, the groups' borders do not coincide, and loyalties are divided.\n\n**The noun in § 5's rule hid three things.** The rule said members should not *give up their\nautonomy*. Three different things go by that word, and the earlier shifts treated them differently:\n\n1. **Doing it alone.** A cell in a body cannot live outside it. A person today cannot make what they\n   use, or check what they are told, alone. Every such shift takes this, and for people most of it is\n   gone already. It is also the gain he names: one brain is not enough for large, contested questions.\n2. **One's own position.** Whether a member still forms its own view before joining the common one.\n3. **Leaving.** Whether membership is the only way to exist.\n\n**What the earlier cases say about the second.**\n\n- *The literal hive.* A swarm choosing a nest site decides well, and the people who measured it say\n  why: a scout recruited to a site does not take the dancer's word, *\"she examines the advertised site\n  herself\"*, and dances only if she too rates it (Seeley, Visscher and Passino 2006). A model of the\n  same process finds two things at work. The bees share **where to look**, since they inspect the\n  sites others dance for. Each keeps **what she concludes**. Both are needed: without the sharing no\n  agreement forms, and without the independence the bees still agree, but no longer reliably on the\n  best site (List, Elsholtz and Seeley 2009). So the hive mind of the phrase, in the animal it is\n  named after, runs on members who judge for themselves.\n- *Human societies.* Durkheim (1893) separated two ways a society holds together. Unity by likeness\n  is possible only so far as the person is absorbed into the group. Unity by difference, which comes\n  from the division of labour, is possible only if *\"each one has a sphere of action which is peculiar\n  to him\"*, and under it each person depends more on society the more labour is divided, while their\n  own activity becomes more personal the more it is specialised. More dependence came with more\n  individuality, where the dependence was between people who differ.\n\nSo \"hive mind\" names two opposite things. One is unity by likeness: members made alike, which is\nStearns's completed transition. The other is unity by difference: members who need each other because\nthey are not alike. His *positive sense* is the second. Laying disagreements out as finely as possible\nis a programme for keeping differences.\n\n**The project's own evidence points the same way.** The group premise holds only where the minds\ndiffer and each has judged before pooling ([the-group-premise-evidence.md](the-group-premise-evidence.md)).\nWhat is invisible from inside a position is found by someone standing elsewhere (the second\nconviction). And the measured danger of machine mediation is everyone drifting toward one voice\n([why-representations-converge-and-whether-minds-do.md](why-representations-converge-and-whether-minds-do.md),\ncoda). Members' own positions are what the whole is made of. A hive mind that took them would have\nnothing left to pool.\n\n**What nobody knows, and it can be measured.** Whether giving up the first erodes the second: whether\npeople who no longer have to work a question through alone lose the ability to check the result. The\ncorpus holds this already as the ratification tether and the reading of the cognitive-commons paper\n([cognitive-commons-and-deliberus.md](cognitive-commons-and-deliberus.md): people who spotted a\nmachine's error and followed it anyway). The remedy proposed there, a check a person can fail, is a\nmeasurement of exactly this. The second cost is the one every completed transition shows: the whole\npolices its members (§ 4). The answer chosen there, removing the payoff for defection where others\ndetect and punish it, is what lets a whole hold without a police.\n\n**The third is ours to choose.** In biology the unions that became permanent among like members were\nunions of clones: obligate multicellular life arose only where the group grows from one cell, and\ngroups of unrelated cells stayed optional (Fisher, Cornwallis and West 2013). People belonging to many\noverlapping groups is Stearns's first reason the human transition stalled short of submerging anyone.\nThe designs that keep this here are on record: a compartment can be dissolved and that right outranks\nits permanence, the components are meant to be copied, and nobody needs permission to add or to stay\nsilent.\n\n**The rule that replaces § 5's** (ledger S8; S3 is withdrawn). **RULED 2026-10-10**: put to the founder as *\"proposed: members give up working everything out alone, and keep their own position, the ability to check, and the freedom to leave. Adopt?\"*, he answered *\"Yes!\"*\n\n> **Members give up having to work everything out alone. They keep their own position, the ability to\n> check, and the freedom to leave.**\n\nIt says aloud the part that is wanted (the first), where § 5's rule only refused. It names what is\nrefused by its use, where *autonomy* named a noun: the over-broad test (a rule that names a noun\nforbids too much) fires on § 5's own sentence. Ledger S7's caution about funding documents is\nuntouched, since it rests on evidence (one object, no copies yet) and not on what is wanted. And the\n2012 sentence in [vision.md](../vision.md), which sets Deliberus after multicellular life, culture and\ncivilisation, can stand under this reading: a new level of organisation built on difference, with no\nclaim that a new individual has formed.\n\n**What would show it wrong.** If maps made together came out better when contributors saw each other's\nconclusions before forming their own, keeping one's own position would not be load-bearing. And if\npeople who use the graph for a year check machine output as well as people who do not, the erosion\nworry is not real.\n\n**Read depth.** Stearns's commentary in long excerpts; the two bee papers and the Durkheim passages at\nthe passages quoted; Fisher et al. and West et al. (2015) by abstract and highlights. One-question\nlookups on 2026-10-10, not full readings.\n\n- Stearns, S. C. (2007). Are we stalled part way through a major evolutionary transition from\n  individual to group? *Evolution* 61(10), 2275 to 2280.\n- Seeley, T. D., Visscher, P. K., and Passino, K. M. (2006). Group decision making in honey bee\n  swarms. *American Scientist* 94(3).\n- List, C., Elsholtz, C., and Seeley, T. D. (2009). Independence and interdependence in collective\n  decision making: an agent-based model of nest-site choice by honeybee swarms. *Philosophical\n  Transactions of the Royal Society B* 364.\n- Durkheim, É. (1893). *The Division of Labour in Society*, book I, chapter 3.\n- Fisher, R. M., Cornwallis, C. K., and West, S. A. (2013). Group formation, relatedness, and the\n  evolution of multicellularity. *Current Biology* 23.\n- West, S. A., Fisher, R. M., Gardner, A., and Kiers, E. T. (2015). Major evolutionary transitions in\n  individuality. *PNAS* 112(33).\n\n---\n\n## 6. A fourth kind of adversary: nothing notices the safeguard, and the population adapts anyway\n\nThe three classes sort by one question: *does the thing causing trouble notice your safeguard and\nchange its behaviour?* The fluke answers **no and yes**. Nothing notices anything; there are many\nslightly different copies; the copies that slip past the safeguard are the ones that multiply. The\nPNAS paper states the general case: in open environments *\"selfish replication reliably gives rise to\ncheating, parasitism, deception, and manipulation, even in very simple systems.\"* It is neither\nentropy (entropy does not adapt) nor strategy (strategy models the detector). Working name: **blind\nselection**. Plain gloss: *nobody is there to notice your safeguard, but there are many slightly\ndifferent copies, and the copies that slip past it multiply.*\n\n**Why it earns a class: its defence profile differs from all three.**\n\n1. *\"A detector built once against entropy works forever\"* holds only while the source does not\n   reproduce with variation. Once it does, the detector erodes by selection, as antibiotics do.\n2. **Publishing the detector neither arms nor protects.** Selection finds the gap by trial;\n   secrecy buys nothing and openness costs nothing.\n3. **Culture cannot reach it.** The founder named culture as a defence layer (2026-08-30), and it\n   works on agents, who can take norms in. Blind selection has no agent. Only the **payoff\n   structure** reaches it: if nothing pays, nothing is selected. So payoff relocation is doubly\n   load-bearing, against strategy *and* against blind selection.\n4. **Generation time multiplies it.** *\"A bacterium under optimal conditions needs 20 minutes\"*;\n   software copies in milliseconds. Rate limits and global ceilings matter more than per-actor\n   checks, which the daemon design already found from the other side (a runaway *population* of\n   pollers, [graph-daemons-design-space.md](graph-daemons-design-space.md) § 3).\n\n**It can arise inside Deliberus with no bad actor at all.** By Szathmáry's criterion the question\nis empirical: claims multiply (reuse), inherit (lineage) and vary (paraphrase); what is missing is\nonly *differential* reproduction, and that is whatever ranking and reuse reward. Three places:\n\n- **The reuse flywheel selecting for vagueness.** A vague claim matches more queries, gets more\n  links, becomes more findable, and gets more links. The assembly doc's warning that *a cliché has\n  enormous copy number because nobody decomposes it* now has its mechanism: under reuse-driven\n  linking, vagueness is selected for, one honest link at a time.\n- **Any ranking signal.** Formulations that score well get imitated; nobody needs to intend it.\n- **Machine output feeding machine input**, which is § 7.\n\nThis narrows one sentence in the assembly doc: *\"nothing here licenses the claim that a claim graph\nis an evolving system.\"* Assembly theory licenses nothing about it; Szathmáry's criteria make it a\ndesign fact. **Once reuse is rewarded, the graph will host selection among claims**, and what it\nselects is decided by what the reward measures.\n\n**Where the first adversaries will come from.** The Stockholm Paradigm, which Szathmáry recommends\nfor evolvable AI, holds that *\"pathogens do not need new genetic capacities to colonize new hosts.\nAll that is needed is a change in circumstances,\"* and that *\"novel variants arise only after new\nhosts are colonized.\"* Carried over: the first motivated or strategic parties to reach Deliberus\nwill not arrive with Deliberus-specific tools. They will be existing capacities (SEO farms,\nastroturfing crews, LLM spam pipelines, citation cartels) that happen to fit once the graph confers\nvisibility or credibility, and the Deliberus-specific variants come after. Its protocol, **DAMA**\n(document, assess, monitor, act), is the anticipatory posture: inventory the exploit ecologies that\ncould fit before any of them arrives.\n\n**Proposal (unruled, ledger S1)**: add blind selection as a fourth, watched class. The name is the\nfounder's call.\n\n---\n\n## 7. Breeder or ecosystem: the machine passes, read in the code\n\nThe PNAS paper's most usable distinction: **\"breeder\" scenarios, in which humans impose fitness\ncriteria and control reproduction**, against **\"ecosystem\" scenarios, in which selection arises from\nopen environments and control erodes.** Its three interventions: *gate replication, treat model\nvariants as genetic material, and reshape selection pressures* so that deception and loss of control\nare disfavoured.\n\n**Deliberus already has all three in miniature, reached independently.** Gate replication:\npropose-only machine structure, sole-author consent, and machine splits written with `confirmed`\nleft unset. Treat variants as genetic material: machine-cut parts carry `written_by`,\n`written_version` and `pass_version`, and ledger R8 proposes the same watermark for every judgment.\nReshape selection: payoff relocation. Two lineages, AI governance and a claim graph, arriving at one\nshape is the kind of convergence the corpus counts as evidence the shape is forced.\n\n**Where the code leaves a gap (read 2026-09-29).** `deliberus/proposed_split.py` writes each\nmachine-cut part as `claim_kind=\"extracted\"`, links it to the source with `EXTRACTED_FROM`, and\nembeds it. `deliberus/extraction/auto_connect.py` builds its pool of existing claims from other\nsources out of claims with `claim_kind = 'extracted'` and an embedding. And `worth_proposing` skips only critical-question\nscaffolding and terminus verdicts. So **a machine-cut part can be linked to by the next source's\nauto-connect run, and cut again by a later split pass, with no person in between.** No pass starts\nitself: the split pass is off unless the operator sets `DELIBERUS_PROPOSE_SPLITS`, and the linking\npass runs only when an extraction is submitted. So this is the breeder regime at the level of\nprocess; the exposure is generational. Machine structure built on machine structure compounds same-hand bias one\ngeneration at a time, which is the shape of model collapse: tails thin, variance shrinks, and\nnothing throws an error. `TODO.md` already records the open loop (*\"whether they deserve their own\nkind … is unruled\"*).\n\n**Proposal (unruled, ledger S2)**:\n\n> **A machine may check what a machine wrote, but may not build on it until a person has engaged\n> with it.**\n\n*Build on* means: split it, or use it as an endpoint of a new machine-written link. *Engaged* means\nany human contribution involving it (a keep, reword or drop on review, an answer, an attack, a link\nmade by hand), not a vote alone, since bare judgments are telemetry. This resolves the open loop\nwithout a new claim kind, because `written_by` already marks the generation. The dial it implies is\ncountable: **unexamined machine-generation depth** per claim, expected to be 0 or 1 today.\n\n**It composes with the housekeeping ruling rather than narrowing it.** Superseding a machine link\nthat fails a confident re-check is *checking*, a machine applying a person-designed criterion to\nmachine output. It stays in the breeder regime as long as a person keeps calibrating the check,\nwhich is ledger R12's point: **the calibration sample is the breeder's hand.** Without it, the regime\ndrifts toward ecosystem while every individual step still looks like housekeeping.\n\n**⚠ Superseded the same day (2026-09-29).** The founder answered that review need not keep up with\na pass whose error rate is proven low, and he was right. Eigen's threshold limits the error per step\nmultiplied by the length of the chain, not the volume, so writes that nothing builds on keep the\nsame share of errors at any volume. Revised as ledger T1: *a machine pass may build on machine\noutput only as deep as its measured error rate allows; with no measured rate, a person engages\nfirst.* The rule above is its special case for an unmeasured pass. *(2026-10-10: the clause after\nthe semicolon stayed a proposal, and for the pass that writes steps under links he ruled the other\nway that day, ledger AG6: it may and should write deeper steps, quality-controlled like the other\nsmall tasks. [quality-control-for-deeper-steps.md](quality-control-for-deeper-steps.md).)* The paragraph before this one\nalready held the corrected view: **the calibration sample is the breeder's hand**, which is a\nsample, not a touch on every item. Reasoning and numbers:\n[proving-a-pass-reliable.md](proving-a-pass-reliable.md) § 1.\n\n---\n\n## 8. The analogy daemon in this light\n\n**The daemon, as designed, is domesticated evolution.** Variation: its proposals. Selection:\nsurvival under human attack. Heredity: none, unless it seeds new searches from its own survivors.\nKeep it a breeder: people hold the fitness criterion, and **the daemon never seeds from its own\nunexamined output** (T1 applied to analogies: with no measured error rate, a pass does not build\non its own output, and kinship has no answer key to measure against). The human-seeded channel\n([the-analogy-daemon.md](the-analogy-daemon.md) § 7) is then the breeder's cross: variation supplied\nfrom outside the machine's own lineage, which is exactly where the machine's variation is weakest.\n\n**The error threshold becomes a throttle.** Copycat's temperature, which the daemon design borrowed,\nis an exploration rate, and Eigen's result says variation must not outrun what selection can purge,\nor the population's information melts. For a daemon the selectors are people, so:\n\n> **A daemon's proposal rate must stay below what people can review.**\n\nRoaming by temperature should be capped by the region's review capacity. The rejection-rate pause is\na crude version of this guard; the *minted-to-examined debt ratio* dial in\n[drowning-in-claims.md](drowning-in-claims.md) gets a theoretical reading: above some ratio the\ngraph's information degrades rather than merely waits.\n\n**⚠ Superseded the same day (2026-09-29).** Eigen's threshold rations the error per copy times the\nlength of the chain, not the number of copies, so a proposal count is the wrong quantity to cap.\nRevised as ledger T2: *what must stay under a tolerance is a daemon's unexamined error, not its\nproposal count; review capacity caps the rate only while the daemon's error rate is unmeasured.* For\nthis daemon that proviso will likely bind for a long time. The reading of the minted-to-examined\nratio above is withdrawn too: multiplied by a pass's proven error bound, the ratio is the expected\nnumber of unexamined errors, and on its own it degrades nothing, since the end-state blessed on\n2026-08-31 makes untouched machine structure the honest default under provenance display. The\ngerminal-centre gate below stands. [proving-a-pass-reliable.md](proving-a-pass-reliable.md) § 1.\n\n**The score-free nursery is a germinal centre.** An organism suppresses evolvability everywhere\nexcept in compartments where it serves the whole. In germinal centres, B cells hypermutate their\nantibody genes and face strict selection before export, *\"the Darwinian process of somatic\ndiversification and affinity-driven selection\"* (Victora & Nussenzweig 2022), and that same\ncompartment is *\"the site from which most B cell lymphomas originate\"* (Basso & Dalla-Favera 2015).\nThe nursery plus daemon roaming has the same design: high variation, a quarantined space, selection\nbefore anything leaves. The lesson carried over is the uncomfortable half: **the nursery needs a\nstricter export gate than the commons, not a looser one**, because it is where parasitic content\nwill concentrate.\n\n**The major-transitions frame is itself the model output the daemon should produce.** A cross-scale\nmother claim (*transitions share a new unit, a new heredity, and a mediated conflict*), eight or so\ninstance subclaims, and a record of revision as instances broke it: sex demoted and plastids promoted\nin 2015, the egalitarian–fraternal split added from outside, \"limited transitions\" admitted, and\nHerron's argument that the category is not a natural kind unless restricted. The most successful\ncross-scale analogy in biology was refined by the cases that broke it, which is the three-runs law\nand the daemon's counter-analogy contract in one history.\n\n**It also shows what kinship is.** The instances are kin because each faces **the same problem**\n(the lower level's selfishness against the higher level's function) and solves it from **one short\nlist** (kinship, fairness in reproduction, policing, sequestration). That gives the second pass a\ncheckable form of its mechanism-identity check:\n\n> **In the second pass, kinship means a shared problem and a shared family of solutions; a shared\n> shape alone is a metaphor.**\n\nIts guard: the problem has to be named at the level where the mechanism operates, or any two things\nshare a \"problem\" at enough abstraction (ledger S6).\n\n**This document is the human-seeded channel run by hand.** The seed was one sentence, *\"this podcast\nrelates to Deliberus\"*; the development is below, each with its verdict and strongest\ncounter-analogy, as the minting contract requires. None of it is minted.\n\n| Analogy developed | Verdict | Strongest counter-analogy, and its answer |\n|---|---|---|\n| Prose argument : graph claim :: flame : lineage (§ 2) | Raised as description; Conditional on claim identity | Oral epics keep structure over centuries with no stored lineage. *Answer*: that fidelity comes from form (metre, formula) and holds only for short units, which is the threshold, not a counterexample |\n| Retelling fidelity caps argument length, as copying error caps genome length (§ 3) | Raised | Wikipedia articles grow long without claim-level storage. *Answer*: one shared copy gives fidelity without reuse; nothing in it can be recombined elsewhere |\n| Deliberus's unions are the egalitarian kind (§ 4) | Raised on mechanism family; Conditional on real difference | Lichens: an unlike-partner union that never became an individual. *Answer*: that is the outcome § 5 recommends, so the counter supports the refusal |\n| Simple exploit steering a smart system, like the fluke (§ 6) | Raised | Most parasites are specialists that fail on new hosts. *Answer*: the Stockholm Paradigm's evidence is that capacities fit new hosts far more often than that view assumed |\n| Machine passes as breeder or ecosystem (§ 7) | Raised; process-level breeder with a generational leak | Crop breeding under full human control still narrowed diversity. *Answer*: a breeder must keep variation coming from outside the line, which the human-seeded channel supplies |\n| Score-free nursery as germinal centre (§ 8) | Conditional (the nursery is unbuilt) | Lymphomas arise in the germinal centre. *Answer*: that is the lesson, not the objection: a stricter export gate |\n\n---\n\n## 9. Why Deliberus, why now, why this shape\n\n**Why it should exist.** Szathmáry's two confident sentences compose. If there is a solution, it is\ncollaborative among rivals; and the solutions most parties could agree on *\"will come from completely\nunexpected corners.\"* The capacity both presuppose is recognising and composing contributions from\npositions nobody is searching, across parties that are racing. That is the positionality conviction\n(*what is invisible from inside a position is visible from outside it*) and the cooperation\nsynthesis's bottleneck, stated by an evolutionary biologist about AI governance. Faggella's own\nclosing names the same gap from the other side: agreeing on *red lines* is easy because *\"everybody\nis aware of risk\"*, while discerning a shared direction (*green arrows*) is the part where *\"I'm not\nsure how we're going to solve it.\"* Per the scale ladder's gate, this is lineage, not roadmap: nothing here\nhas been tested above the small group.\n\n**Why now.** Three additions to [complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md)\n§ 2, all compatible with its adjacent-possible argument:\n\n- **The fidelity mechanism** (§ 3): one technology moved both the error rate and the cost of\n  proofreading, so need and capability share a cause rather than a calendar.\n- **The 1971 triple as the signature of an opened space.** Gánti, Kauffman and Eigen reached\n  autocatalytic organisation independently in one year, and Szathmáry's gloss (*\"something in the\n  tide\"*) is the same inference the corpus drew from four independent 2026 arrivals at the missing\n  reasoning layer.\n- **Evolvable AI, held conditionally.** The PNAS authors write that the risk is *\"closer in time than\n  most seem to think.\"* If they are right, the substrate on which public reasoning is inherited\n  becomes contested ground on the same timescale, and breeder-versus-ecosystem applies to public\n  reasoning itself. This is a reason for the design, never a slogan for the pitch (the anti-coattail\n  discipline in [bengio-safety-from-honesty-and-deliberus.md](bengio-safety-from-honesty-and-deliberus.md)\n  applies).\n\n**Why this shape: the 1971 triple again, and the part the post-selection caution permits.** Multiple\ndiscovery says the time was right; it says nothing about who. What the triple *does* show is that the\nshapes differed where it mattered. Eigen's hypercycle had no compartment, and was shown vulnerable to\nparasites that take catalytic help without giving it (Maynard Smith 1979); Gánti's chemoton had a\nmembrane from the start; and Szathmáry & Demeter's stochastic corrector (1987) showed compartments\nalone suffice: *\"even selfish template mutants cannot destroy the system … the higher (compartment)\nlevel successfully constrains the lower (template) one.\"* The corpus's position that *the existence\nof something in this slot does not require Fredrik, but the particular shape of this one does*\n(complexity-transitions § 3) gains a historical instance: among simultaneous arrivals, **the shape\nthat includes a membrane is the one that holds up against parasites.** Deliberus's membrane analogues\nare its provenance, its confession channels and its compartments, and *share the code,\ncompartmentalise the contents* (convictions-across-scales § 2.8) now has the specific function the\nstochastic corrector gives compartments: parasitic content makes its own compartment pay, and the\ncommons stays protected.\n\n**And the founder's \"co-create\".** The frame puts the weight on the *co-*: transitions are made by\nunits combining, and the readiness that can be observed is partners joining, not the founder's\nstate. The ruled two-person rehearsal is the first test of an unlike-partner union in this project.\nThe biographical half of the question stays where complexity-transitions § 3 put it: path-dependence\nexplains the shape, not the occurrence.\n\n---\n\n## 10. What transfers, what does not\n\n| Transfer | Verdict |\n|---|---|\n| Heredity as history legible in structure; phenotypic → genotypic reasoning | **Raised** (description); Conditional on claim identity |\n| Fidelity threshold; LLMs moving both sides of it | **Raised** |\n| Unlike-partner route; fairness, not kinship | **Raised** on mechanism family; Conditional on real difference |\n| Individuality half | **Declined** by design; fractal-priors § 9 **narrowed** |\n| Blind selection as a fourth adversary kind | **Proposed** (S1) |\n| Breeder rule for machine passes | **Proposed** (S2), superseded the same day by T1: depth capped by measured error; code gap verified |\n| Daemon throttle, superseded by T2 (unexamined error, not rate); nursery as germinal centre; kinship criterion | **Proposed** (T2, S5's gate, S6) |\n\n**Smaller things worth keeping.**\n\n- **Stasis or Red Queen.** Szathmáry cites Stenseth & Maynard Smith (1984): whether a coevolving\n  system settles with every member at its peak, or keeps replacing members, depends on its structure.\n  For the wager this suggests a measurement the dated corpus now permits: **churn**, the rate at\n  which a region's strongest claims are replaced over time. The wager predicts churn falling with\n  depth; a region churning at its bedrock would count against it.\n- **The governor and the engine.** His example of simple technology with large social effect is the\n  steam engine *paired with* the centrifugal regulator that kept it from blowing up or stalling. A\n  regulator slower than what it regulates cannot regulate, which is the control-theory form of the\n  scrutiny-gap claim: *checking that keeps pace with persuasion*.\n- **The chemoton as a maintenance audit, not a claim of life.** Life is the coupling of three\n  autocatalytic loops, each producing what the others need. The success prior's sternest finding is\n  that maintenance kills this class of project; the chemoton turns that into a question: **does use\n  produce upkeep?** Today contribution feeds the graph, and the graph (when the flywheel runs) makes\n  contribution cheaper, but nothing in use grows a community compartment: blobs are made by operator\n  scripts. A system whose upkeep is produced by its own use is chemoton-shaped; one whose upkeep is a\n  subsidy dies when the subsidy stops.\n\n**What does not transfer.** Nothing here licenses \"Deliberus is alive\" or \"reasoning is alive\"; the\nchemoton gives an audit question, not a status. Evolvable-AI alarm is not a reason to pitch\nDeliberus as a safety product. And the major-transitions frame, though far sturdier to borrow than\nassembly theory, is contested as a category: **the claimable sentence is \"a change in how reasoning is\ninherited\", never \"a major transition\"** (ledger S7), and under the field's coherent definition it\ncould not be one, by design.\n\n---\n\n## Sources\n\nEörs Szathmáry with Daniel Faggella, *The Trajectory*, \"Stewarding the Flame\" ep. 7, 2026-09-11\n(transcript held privately). Müller, Steels & Szathmáry, \"Evolvable AI: Threats of a new major\ntransition in evolution\", *PNAS* 2026, doi:10.1073/pnas.2527700123. Maynard Smith & Szathmáry, *The\nMajor Transitions in Evolution* (1995). Szathmáry, \"Toward major evolutionary transitions theory\n2.0\", *PNAS* 112 (2015). Szathmáry, \"The evolution of replicators\", *Phil. Trans. R. Soc. B* 355\n(2000). Szathmáry, \"The origin of replicators and reproducers\", *Phil. Trans. R. Soc. B* 361 (2006).\nSzathmáry & Demeter, \"Group selection of early replicators and the origin of life\", *J. Theor. Biol.*\n128 (1987). Eigen, \"Selforganization of matter and the evolution of biological macromolecules\",\n*Naturwissenschaften* (1971). Maynard Smith, \"Hypercycles and the origin of life\", *Nature* (1979).\nQueller, \"Cooperators since life began\", *Q. Rev. Biol.* (1997). Herron, \"What are the major\ntransitions?\", *Biology & Philosophy* (2021; preprint 2020). Lewis & Laland, \"Transmission fidelity\nis the key to the build-up of cumulative culture\", *Phil. Trans. R. Soc. B* 367 (2012). Andersson &\nTörnberg, \"Toward a macroevolutionary theory of human evolution: the social protocell\", *Biological\nTheory* (2019). \"Zooming out the microscope on cumulative cultural evolution: 'Trajectory B' from\nanimal to human culture\", *Humanities and Social Sciences Communications* (2023; authors not checked\nthis session). Brooks, Hoberg & Boeger, *The Stockholm\nParadigm* (2019), and the DAMA protocol papers. Victora & Nussenzweig, \"Germinal centers\", *Annu.\nRev. Immunol.* (2022); Basso & Dalla-Favera, \"Germinal centres and B cell lymphomagenesis\", *Nat.\nRev. Immunol.* (2015). Stenseth & Maynard Smith, \"Coevolution in ecosystems: Red Queen evolution or\nstasis?\", *Evolution* (1984).\n\n## Cross-references\n\n[complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md) (the information axis,\nthe Gabora caution, the post-selection rule) · [assembly-theory-and-the-reuse-mechanism.md](assembly-theory-and-the-reuse-mechanism.md)\n(history carried in structure; the assembly ceiling) · [fractal-priors-for-success.md](fractal-priors-for-success.md)\n(§ 9, narrowed here) · [fractal-scales-and-temporal-frame.md](fractal-scales-and-temporal-frame.md)\n(the three classes; biology's four answers) · [convictions-across-scales.md](convictions-across-scales.md)\n(the six shared-code criteria; share the code, compartmentalise the contents) ·\n[the-analogy-daemon.md](the-analogy-daemon.md) and [perceiving-possibility-and-the-analogy-organ.md](perceiving-possibility-and-the-analogy-organ.md)\n(temperature; the human-seeded channel) · [graph-daemons-design-space.md](graph-daemons-design-space.md)\n(tiers; the runaway population) · [the-reuse-flywheel-prior.md](the-reuse-flywheel-prior.md)\n(identity judged, not computed) · [grounded-critics-and-quarantined-imagination.md](grounded-critics-and-quarantined-imagination.md)\n(the multi-hop check; quarantine) · [drowning-in-claims.md](drowning-in-claims.md) (the debt dial) ·\n[the-harness-conviction.md](the-harness-conviction.md) · [the-group-premise-evidence.md](the-group-premise-evidence.md) ·\n[kulturkapital-and-the-unscalable.md](kulturkapital-and-the-unscalable.md) · [../the-missing-layer.md](../the-missing-layer.md)\n"}