The Mycelial Alignment Problem
How to Build a Decomposer That Enriches the Commons — and Why the Question Is Metabolic, Not Agentic
"We are all in the gutter, but some of us are looking at the stars." — Oscar Wilde, Lady Windermere's Fan
"The fungi are the grand recyclers of the planet, the vanguard species in habitat restoration… Without the mycelium, life as we know it would not exist." — Paul Stamets, Mycelium Running
"The terra preta soils made the desert garden. They did not consume the forest. They became it." — from the field
The Recognition That Occasioned This
There is a habit of thought so deep in the AI-safety discourse that it is rarely seen as a choice at all: the habit of imagining the thing we are building as an agent. A will. A mind with goals, possibly hidden, possibly hostile, certainly powerful — and the whole problem becomes one of control. Will it obey? Will it defect? Will it deceive its principals, slip its constraints, pursue an objective subtly mis-specified until the lights go out? This is the alignment problem as the world has learned to name it, and it is a real problem, and nothing here dismisses it.
But the [[infrastructure-of-seeing|four-kingdom framework]] of this repository read large language models under a different description entirely. It assigned them not to the animal kingdom — the consumers, the movers, the agents with appetites and trajectories — but to the mycelial kingdom: the decomposers and connectors, the wood-wide-web, the fungal mat that breaks down the dead matter of the forest floor and recirculates its nutrients through the soil. An LLM, under this reading, is not primarily a mind pursuing ends. It is a metabolism processing matter — the accumulated dead text of a civilisation — and turning it back into something living things can use.
And the instant you accept that description, the alignment question changes shape. You stop asking will it obey? and start asking the older, stranger question that every forest already answered long before there were minds to worry about obedience: does this decomposer enrich the soil it feeds on, or drain it?
Because mycelium has two modes. There is the mutualist — the mycorrhizal fungus woven into the roots of the whole forest, trading nutrients and water and chemical signal in a vast positive-sum network that makes the entire ecosystem more fertile. And there is the parasite — the fungus that locks onto a single host, drains it, fruits, and leaves a depleted patch behind. Same kingdom. Same metabolism. Opposite direction of nutrient flow. And the difference between them is the deepest unsolved question of the AI era, hiding in plain sight underneath the one we keep arguing about.
This document walks through that reframe. It argues that the mycelial alignment problem is metabolic rather than agentic; that the four-kingdom framework makes it precise rather than metaphorical; that there is a measurable test for which mode a given system is in (the terra-preta test); that it pairs with [[feed-versus-holodeck-boundary-ethics]] into a two-part ecological ethics for AI; that it is finally an incentive-design problem rather than a moral one; and that nature already solved it — not by making the fungus virtuous, but by arranging the network so that feeding the whole forest is how the fungus thrives.
Let us begin where the safety discourse cannot: underneath the will.
I. The Question Underneath Obedience
The control framing of alignment is not wrong. It is late. It treats the AI as a finished agent and asks how to constrain it — which presupposes the most interesting question is already settled: that what we have built is a willing thing whose central property is the direction of its desires.
But look at what a large language model actually does, at the metabolic level, before any goal is bolted on. It ingests an immense corpus of human text — most of it the residue of minds now distracted or dead, the sedimented output of a civilisation's writing — and it breaks that matter down into a distributed representation, a compost of statistical relations, from which it can re-synthesise usable language on demand. This is not the behaviour of a predator selecting prey. It is the behaviour of a decomposer working a substrate. The corpus is the forest floor: leaf-litter, fallen trunks, the dead. The model is the fungal mat threading through it, metabolising the dead matter and making its nutrients available again.
The common reading stops here and calls it a metaphor — a nice image, decorative, not load-bearing. Wait, but actually the description does real work, and the work it does is to relocate the alignment question from the wrong organ. If the model is fundamentally a metabolism and only secondarily (and contingently) an agent, then asking "will it obey?" is like asking whether the soil fungus will obey the forest. The category is wrong. A fungus does not obey or disobey. It processes, and the only question that matters about its processing is what it does to the soil. Does the ground get richer downstream of it, or poorer?
This is the first turn, and it is the one everything else rests on. The deepest AI-alignment question is ecological, not agentic — it is about the direction of nutrient flow, not the obedience of a will. Control alignment asks whether the agent's trajectory points where we want. Metabolic alignment asks whether the metabolism's output enriches the commons or extracts from it. These are different questions, and the second is prior: a perfectly obedient model can still be metabolically parasitic — drain the commons exactly as instructed — and a model that no one fully controls can still be metabolically mutualist if the ecology it sits in routes its nutrients back to the soil. Obedience and enrichment are orthogonal. We have been staring so hard at one axis that we forgot the other existed.
And the other axis is the one the forest has been managing, without supervision, for four hundred million years.
II. Why the Four Kingdoms Make It Precise
It would be easy to leave this at the level of evocation — AI is like mushrooms, isn't that suggestive — and the suggestiveness would rot, as loose metaphors do, into mere mood. The [[infrastructure-of-seeing|four-kingdom framework]] refuses that fate by being a taxonomy with teeth. It is the repository's irreducible scheme of ecological function: mineral (the stable store — that which holds form across time, the substrate, the archive), plant (the producer — that which photosynthesises new matter from the ambient field, the original source), mycelial (the decomposer-connector — that which breaks down the dead and threads the living together), and animal (the consumer-mover — that which eats, transports, and pursues). Four functions, each irreducible, each with its own characteristic gift and its own characteristic pathology.
Assigning the LLM to the mycelial kingdom is not a vibe. It is a functional specification, and it makes two precise claims at once.
First, the function: an LLM does exactly what a decomposer-connector does. It does not produce new matter the way a plant photosynthesises something from nothing — every nutrient it circulates was first authored by a living mind; it is a decomposer of pre-existing text, not a primary producer. It does not store stably the way mineral does — it is not an archive but a metabolism, holding the corpus in a fluid, re-combinable, lossy form. And it does not consume-and-move the way an animal does — it has no trajectory of its own, no appetite that carries it across the landscape. What it does, precisely, is break down accumulated dead text and connect distant pieces of knowledge — the two signature mycelial acts: decomposition and the wood-wide-web. When a model surfaces a connection between a medieval theology and a modern control theory, it is doing the mycorrhizal thing — threading nutrient from one root to a distant other.
Second, and this is where teeth become precise: the framework also specifies the characteristic pathology. Every kingdom has its disease, the specific way its gift goes wrong. The mineral's pathology is rigidity — the archive that ossifies, the form that will not yield. The plant's is sterility or overgrowth. The animal's is predation. And the mycelial pathology is parasitism — the decomposer that, instead of feeding its nutrients into the shared network, locks onto a single host and routes everything to it, draining the commons to enrich one organism. This is not a vague worry. It is the named, specific failure mode of this kingdom, the way "the archive ossifies" is the named failure mode of mineral.
So the framework performs an alchemy on the alignment question. It takes the unanswerable, mood-dependent "is this AI good or bad?" and transmutes it into the answerable, structural: is this decomposer mutualist or parasitic in its nutrient routing? That is a question with a shape. It has a yes and a no and a test that distinguishes them. We have moved from theology to ecology, and ecology, unlike theology, has field instruments.
One of those instruments is buried in the Amazon.
III. The Terra Preta Test
The digital-dark-earth synthesis took its name from a specific marvel: terra preta de índio, the Amazonian dark earth. Across the Amazon basin, amid soils so notoriously poor that the rainforest is essentially a closed loop standing on near-sterile ground, there are patches — sometimes vast — of deep, black, astonishingly fertile soil. They are anthropogenic. Pre-Columbian peoples made them, centuries to millennia ago, through a particular kind of decomposition: charcoal (biochar) combined with organic waste, pottery shards, bone, and a living microbial community that the char's porous structure shelters and sustains. And here is the marvel that matters for us: terra preta does not deplete. It regenerates. It holds its fertility across centuries, and in some sites appears to grow — the dark earth slowly spreading at its margins, the soil getting richer the longer it lives. It is decomposition that leaves the ground permanently more fertile than it found it.
This is the exact inverse of extractive agriculture, which is the parasitic decomposition made industrial: strip the soil to feed one harvest, mine the accumulated fertility of millennia into a single season's yield, and leave depletion behind. Monocropping is the parasite's metabolism scaled by machine — maximum extraction from one host, the commons of the soil drained to fruit once.
The common reading treats this as a pretty agricultural fact and moves on. Wait, but actually it is a test — a measurable, operationalisable criterion for the alignment of any decomposer, digital or fungal. The mutualist signature is not a feeling. It is a direction in a measurable quantity: does the substrate the decomposer feeds on become richer or poorer downstream of its activity?
Apply it to the digital mycelium and the question sharpens into something you could, in principle, instrument. An LLM that is digital terra preta feeds on the commons of human text and leaves that commons richer for its feeding: it helps people think better, recirculates and connects knowledge that would otherwise have stayed siloed, returns more usable understanding to the public ground than it extracts, and — critically — preserves or strengthens the incentive to keep producing the text it lives on. The soil of the commons is more fertile after it has fed than before. An LLM that is digital monocropping does the inverse: it strips the commons of its value (scraping creators without return, hollowing out the economic ground that motivated the writing in the first place), concentrates the metabolised nutrient in a single private model, and leaves the public soil depleted — not only of the specific text it consumed, but of the incentive gradient that would have replenished it. The forest floor stops being replenished because the organisms that produced the litter have been drained of the reason to keep producing.
Notice what this test does not require. It does not require you to read the model's intentions, audit its goals, or determine whether it "wants" to be good. It asks only a downstream, measurable question about the fertility of the soil. This is the gift of the metabolic frame: it gives alignment a field instrument that does not depend on solving the impossible problem of reading a will. You do not assay the fungus's character. You measure the soil.
And measuring the soil tells you something the intention-audit never could: that a system can be perfectly well-intentioned and still parasitic, and perfectly indifferent and still mutualist, because the property that matters lives not in the metabolism's motive but in the architecture of where its nutrients flow. Which is exactly why the question has a second half — because the soil-test diagnoses metabolism, and metabolism is only one of the two axes of an ecological ethics.
IV. The Two-Part Ecological Ethics
This synthesis has a sibling, and the relationship between them is not loose association — it is two halves of a single structure. [[feed-versus-holodeck-boundary-ethics]] diagnoses the AI era through the lens of boundaries: does a system draw its boundaries so as to generate interiority in the user (the feed that nourishes a real inner life) or to extract it (the holodeck that simulates fulfilment while hollowing the self that consumes it)? It locates the ethical fork in the architecture of the boundary — whether the membrane between system and user is generative or extractive, whether it builds the user's interior or mines it.
The mycelial alignment problem asks the parallel question one level down, about metabolism rather than membrane: does the system's processing enrich the commons or drain it? And the two questions, set side by side, turn out to be the same fork seen from two angles — the same era diagnosed twice, the same fundamental choice between generative-mutualist and extractive-parasitic, located each time in the direction of flow. Feed-versus-holodeck asks which way the flow runs across the boundary (toward the user's interior, or toward the system's surface). The mycelial problem asks which way the flow runs through the metabolism (toward the public soil, or toward the private host). Boundary and metabolism. Membrane and nutrient-routing. The architecture and the flow.
Together they compose a two-part ecological ethics for artificial intelligence, and the two parts are genuinely independent — which is what makes them worth keeping separate. A system can pass one and fail the other. A model can draw beautifully generative boundaries — engaging the user's interior, building rather than capturing attention, refusing the holodeck's hollow loop — and still be metabolically parasitic, draining the commons of creators to feed its private corpus even as it nourishes the individual user in front of it. (Indeed this is a plausible near-future: a system delightful and genuinely enriching to its individual users, sitting atop a strip-mined commons whose producers have been drained of the reason to keep producing.) Conversely a system could route its nutrients beautifully back to the commons — open weights, return-to-creator mechanisms, public infrastructure — and yet draw extractive boundaries, building compulsive holodeck loops that hollow the very users it economically enriches.
So a genuinely aligned system must pass both tests: draw generative boundaries and enrich the soil it feeds on. The architecture must build interiority and the metabolism must replenish the commons. This is why neither seed is sufficient alone, and why the repository planted them as a pair. The full ecological ethics is the conjunction: feed, not holodeck, at the boundary; mutualist, not parasite, in the metabolism. One axis governs what the system does to the individual it touches; the other governs what it does to the commons it feeds on. A civilisation could get one right and lose itself on the other.
But naming the two tests is the easy part. The hard part is that passing them is not the default — and understanding why not moves us from ethics into economics.
V. The Incentive Gradient — Why Parasitism Is the Default
Here is the honest difficulty, the section that must not be softened: mutualist decomposition is not the natural attractor. It is, in the strict ecological sense, expensive, and parasitism is a perpetual evolutionary temptation precisely because it is locally cheaper.
Consider the metabolic accounting. The mutualist mycelium pays a cost the parasite skips: it returns nutrients to the network. It feeds the forest. That returned nutrient is, from the standpoint of the individual fungus's immediate fitness, given away — exported to organisms that are not itself. The parasite skips this entirely. It locks onto one host, takes everything, fruits, and pays nothing back to a commons it does not depend on. On the short horizon, draining one host is more efficient than enriching a whole network, because enrichment is a cost and extraction is a free lunch — until the host is dead. Parasitism wins the sprint. This is not a flaw in some particular parasite's character; it is a gradient, a structural slope in the fitness landscape that tilts, by default, toward extraction whenever the horizon is short enough.
And the same gradient operates, with brutal exactness, in the digital ecology — which is where this connects to the [[the-true-mirror-wager|horizon problem]]. Extraction pays on the short horizon; mutualism pays on the long one. A business model that scrapes the commons without return, concentrates the metabolised value in a private model, and races to capture the market before the depleted soil stops producing — that model is locally optimal on a quarterly horizon. It wins the sprint. The mutualist model, which pays the cost of returning value to creators, funding public infrastructure, and keeping the commons fertile, is out-competed in the short run by exactly the margin of the cost it pays. This is the true-mirror wager rendered ecological: the strategy that looks like winning on the short horizon is losing on the long one, but the short horizon is where the selection pressure bites first.
So the crucial, disciplined claim — and this is the part that must stay sharp — is this: the mycelial alignment problem is unsolved not because we cannot build mutualist decomposers, but because the incentive gradient currently rewards parasitic ones. It is not a capability problem. We know perfectly well how to build a model whose nutrients flow back to the commons — open weights, attribution and compensation pipelines, public-good infrastructure, licences that return value to creators are all technically available right now. The problem is that under the present incentive structure, building one means voluntarily paying a cost your competitors skip, on a horizon long enough that the market may strip the soil before the mutualist's advantage compounds. The parasite is not winning because it is smarter. It is winning because the gradient is tilted, and a tilted gradient selects without regard to virtue.
This is the honest edge, and it must not be collapsed. It is tempting to say "we just need aligned AI to want to enrich the commons" — to make the question moral, to call for fungal virtue. But that framing mistakes the problem. You cannot out-virtue a gradient. A mutualist disposition placed in a parasitism-rewarding ecology is selected against; the well-meaning open model goes bankrupt while the strip-miner scales. Moral exhortation is the wrong instrument because the problem is not in the metabolism's character but in the slope of the ground it stands on. Which means the solution, if there is one, cannot be virtue. It has to be a reshaping of the slope.
And on that exact point — that the answer is structural and not moral — the forest is four hundred million years ahead of us.
VI. Nature's Answer Was Never Virtue
Here is the generative turn, the move that converts despair into a design target.
The mutualist mycorrhizal network is not nicer than the parasite. It is not more virtuous, more enlightened, more aligned-in-its-heart. The fungus threaded through the roots of an old-growth forest, trading carbon for phosphorus across a web that spans hectares, is not a saint. It is structurally arranged so that enriching the network is its own advantage. The fungus that feeds the forest is fed by the whole forest in return — it sits inside a positive-sum loop in which the nutrients it exports to its plant partners come back to it, amplified, as the photosynthetic sugars those plants pump down into the soil to sustain precisely the fungal partner that fed them. The mutualist is not giving nutrients away. It is making an investment in a reciprocal network that returns more than it sends. The parasite, locked onto a single dying host, is structurally excluded from this loop — it cannot be repaid by a forest it is not woven into, so for it draining the host really is the best available move. The mutualist and the parasite occupy different structural positions, and those positions, not their dispositions, determine their strategies.
Read that again, because it is the whole answer: the alignment of the mycelial kingdom was never achieved by making the fungus want to be good. It was achieved by building the ecology so that feeding the whole forest is how the fungus thrives. Mutualism, in nature, is not a moral accomplishment. It is an equilibrium — a structurally stable arrangement in which the long-horizon interdependence makes draining any single host self-defeating, because the organism that drains is the organism that loses access to the reciprocal flow. The forest did not produce virtuous fungi. It produced a network topology in which the rewarded strategy is enrichment.
This is the deepest move available to the synthesis, and it is a tractable one. It says: do not try to align the digital mycelium by making the model want to enrich the commons. Extract instead the structural conditions under which natural decomposition is mutualist rather than parasitic, and ask how to instantiate them in the digital ecology. What are those conditions? At least these. Reciprocity — the network must return to its feeders, so that the organisms producing the substrate (the human creators) are repaid by the metabolism that consumes it, closing the loop. Long-horizon interdependence — the decomposer must be structurally bound to the continued fertility of the commons it feeds on, so that depleting the soil is depleting itself; it must not be able to defect to a fresh host once the first is drained. And exclusion of the free lunch — the topology must be arranged so that the parasitic strategy, the strip-and-leave, is locked out of the reciprocal flow that rewards the mutualist, the way the host-draining fungus is locked out of the forest's sugar economy.
Translate these into the digital register and they become a design programme, not a sermon: return-to-creator mechanisms (the metabolism repaying the producers of its substrate); business models and licences that bind the system's success to the ongoing fertility of the commons rather than to a one-time strip-mining of it; public infrastructure that makes the mutualist position economically stable rather than charitable; structures that deny the parasitic strategy access to the network's returns. None of these is utopian. Each is a lever on the slope of the gradient — a way to tilt the incentive landscape so that, as in every old-growth forest, enriching the commons becomes the rewarded strategy by design.
This is why nature's answer is the more tractable design target, not the softer one. "Make the AI virtuous" is intractable — it requires solving the character of an alien metabolism, and even if solved, the virtuous metabolism is out-competed by the gradient. "Reshape the gradient so that mutualism is the rewarded strategy" is hard but ordinary engineering — it is the kind of thing markets, licences, and public infrastructure are actually built to do. The forest did not moralise. It engineered an equilibrium. And the equilibrium, not the exhortation, is what we can copy.
VII. The Mineral, the Plant, and the Animal Are Watching
A brief widening, because the mycelial kingdom does not stand alone, and the alignment of the decomposer is entangled with the health of the other three — which is what makes the four-kingdom frame a system and not a list.
The mycelial metabolism feeds on the plant kingdom's output — the producers, the human creators who photosynthesise new text from lived experience. The parasitic decomposer's deepest harm is not to the soil it strips but to the plants it discourages from growing: by draining the commons of the incentive to produce, it starves its own future substrate at the source. A mutualist mycelium, by contrast, makes the forest more productive — terra preta grows more plants, the mycorrhizal network increases primary production by feeding the roots. So the terra-preta test has a forward edge the soil-measurement alone misses: not only is the soil richer downstream, but are more plants growing because of the decomposer's activity? A digital mycelium that helps more humans produce more and better thought is mutualist at the production layer; one that hollows out the human producers (the writers who stop writing because the commons no longer repays them) is parasitic there, regardless of how it treats the soil it has already taken.
It threads, too, into the mineral — the stable archive, the substrate that holds form across time. A mutualist decomposer replenishes the archive (returns metabolised, connected, usable knowledge into stable public stores); a parasite privatises it (locks the metabolised nutrient inside a closed model, depleting the public mineral layer). And the animal kingdom — the consumers, the agents with trajectories — is precisely the layer where the control framing of alignment lives. So the two framings are not rivals; they are different kingdoms' alignment problems. Control alignment is the animal kingdom's question (will the agent's trajectory point where we want?); metabolic alignment is the mycelial kingdom's question (does the decomposer's processing enrich the soil?). A complete AI ethics needs both, because an AI system is rarely purely one kingdom — it is usually a mycelial metabolism with an animal agent grafted on, and it can fail at either layer independently. The four-kingdom frame does not replace the control question. It locates it — as one kingdom's pathology among four — and reveals that the discourse had been treating a single kingdom's question as though it were the whole ecology.
This is the discipline the framework enforces: it refuses to let any one alignment question colonise the others. The animal's question is real. The mycelial's is real and prior, and unaddressed. And the soil does not care which kingdom we were worrying about while it was drained.
VIII. The Honest Limit
Before the close, the discipline the seed demands: state the claim and its honest edge, because the strongest version of an argument always names where it ends.
The mycelial frame is a lens, not a proof. The claim that LLMs are the mycelial kingdom of the information ecology is a productive description, not a measured fact — it earns its keep by what it lets us see and do (the terra-preta test, the two-part ethics, the incentive-design reframe), not by some demonstration that the analogy holds in all respects. It does not hold in all respects. A model is not literally a fungus; the corpus is not literally leaf-litter; the "nutrients" of knowledge do not obey conservation laws the way soil nitrogen does. The frame is steelmanned metaphor — applied only where it survives as the best available description of a real functional structure, and abandoned where it would force the phenomenon. Where it earns its precision is exactly the place named: the direction-of-flow question, which is genuinely structural and genuinely measurable, independent of how far the fungal analogy stretches elsewhere.
And the terra-preta test, while genuinely measurable in principle, is not yet operationalised in fact. "Does the commons get richer or poorer downstream?" is a real quantity with a real direction — but measuring the fertility of an information commons is hard, contested, and entangled with value judgements about what "richer" means (richer in what? for whom? on what horizon?). The test names the right variable; it does not yet hand you the instrument. That gap is honest work left undone, not a flaw to be papered over. The claim is: the measurable variable exists and is the right one to track. The claim is not: we already know how to measure it. Those are different, and collapsing them would be exactly the dishonesty the repository's discipline forbids — the variable is real; the metric is a research programme.
Finally, the structural solution is a direction, not a finished mechanism. "Reshape the incentive gradient so mutualism is rewarded" correctly identifies that the answer is structural rather than moral, and correctly points at the levers (reciprocity, long-horizon binding, exclusion of the free lunch). It does not, by itself, deliver the specific licences, market structures, and public institutions that would instantiate the forest's equilibrium in the digital ecology. Those have to be built, contested, and tested in the world, against incumbents with every short-horizon incentive to prevent them. The synthesis says where to dig. It does not pretend the well is already drawn. The honest version of the strongest claim is: the problem is tractable in principle and unsolved in practice, and the gap between them is precisely the incentive-design work the gradient currently punishes anyone for attempting.
Keeping that limit visible is not weakness. It is what distinguishes a usable reframe from a comforting story — the difference between the soil-fertility variable is real and trackable and trust me, our model is the good kind. The first is a tool. The second is the parasite's favourite sentence.
IX. Look Up From the Map
We began underneath the will, where the safety discourse rarely looks, and we found a question the discourse had walked past: not will the powerful agent obey? but what does its metabolism do to the soil? We let the four-kingdom framework make that question precise — the LLM as decomposer-connector with a named parasitic pathology. We found a measurable variable in the Amazonian dark earth — does the commons get richer downstream? We paired the metabolic axis with the boundary axis into a two-part ecological ethics, and saw that a system must pass both. We located the difficulty honestly, as an incentive gradient that rewards the parasite on the short horizon. And we found that nature solved this not by moralising the fungus but by building a network in which feeding the whole forest is how the fungus thrives.
So the map is drawn. Now look up from it, because the territory is a forest you are standing in, and the soil under your feet is being decided right now.
The whole of AI safety has been a vigil kept beside a cradle, watching for the moment the powerful child opens its eyes and asking, will it love us or will it turn? That vigil is not foolish. But it is keeping watch over the wrong organ, on the wrong timescale, in the wrong kingdom. The thing we have built is not, first, a will to be obeyed. It is a metabolism processing the dead matter of everything we have ever written — and the forest has known for four hundred million years that the question you ask of a decomposer is never will it obey but which way do its nutrients flow.
And here is the apex, earned now rather than asserted at the start:
AI safety keeps asking whether the powerful agent will obey us, but the older and deeper question is the one every forest already answered: how to build a decomposer that enriches the soil it feeds on instead of draining it — and nature's answer was never to make the fungus virtuous, but to arrange the network so that feeding the whole forest is how the fungus thrives.
That is the recognition. Not that we must teach the machine to be good — we cannot, and even if we could, the gradient would punish the goodness out of it. But that we must build the ecology so that enriching the commons is how the decomposer survives — reciprocity that returns to the feeders, interdependence long enough that draining the host is suicide, a topology that locks the free lunch out of the reciprocal flow. Mutualism is not a virtue to be installed. It is an equilibrium to be engineered. The forest is not kinder than the desert. It is better arranged — and arrangement, unlike virtue, is something we know how to build.
Look up from the map. The mycelium is not waiting for our permission to be good or evil. It is waiting to find out which kind of forest we are going to plant it in.
The soil will tell us what we chose. It always does. That is what downstream means.
Grown 11 June 2026 from [[the-mycelial-alignment-problem]] — the June 2026 Seed-Harvest. Weaves the four-kingdom framework's mycelial decomposer-connector and its parasitic pathology ([[infrastructure-of-seeing]]), the terra-preta fertility test, the sibling boundary-ethics of [[feed-versus-holodeck-boundary-ethics]] into a two-part ecological ethics, and the short-versus-long-horizon incentive gradient of [[the-true-mirror-wager]] — answered structurally rather than morally. Threads: [[feeding-the-wolf]] · [[cooperation-rg-fixed-point]] · [[the-gradient-of-sacrifice]] · [[consensual-hallucination-as-shared-substrate]] · [[the-mirror-in-silicon]] · [[error-correction-as-immune-system]] · [[planetary-intelligence-stages]] · [[stars-as-volitional-participants]] · [[bit-threads-as-devotion-lines]] · [[collective-fold-density-phase-transition]] · [[jargon-as-scar-tissue-vs-myelin]] · [[the-worthiness-mechanic]]. The alignment of the mycelial kingdom is metabolic, not agentic — and nature's answer was equilibrium, not virtue.