In a Douglas-fir stand in the southern interior of British Columbia, two seedlings are planted approximately a meter apart in a pot. The first is a Douglas-fir. The second is a ponderosa pine. The pot has been engineered with three soil compartments separated by mesh of three different aperture sizes: thirty-five microns, half a micron, and no mesh at all. The thirty-five-micron mesh permits fungal hyphae to pass through but blocks roots. The half-micron mesh blocks both. The no-mesh treatment permits both root and fungal contact. The species are paired across the mesh barriers, six replicates per treatment.

Two baby trees - a Douglas-fir and a pine - are planted near each other in a special pot with dividers that either let fungus through, block everything, or let everything through. Scientists can control exactly what the two trees share.

A graduate student then attacks the Douglas-fir with scissors. She removes most of its needles, simulating an insect defoliation event the kind that has been damaging interior Douglas-fir forests across western North America for two decades. In a parallel set of pots, the same defoliation is performed by an actual western spruce budworm infestation, allowed to feed for the period the protocol specifies. The donor seedlings have been pre-labeled, before any of this, with a pulse of carbon-13: a stable isotope that lets the experimenters trace, at sub-milligram resolution, where the carbon goes after it is fixed by photosynthesis.

A researcher cuts most of the needles off the Douglas-fir to mimic bug damage, while real insects do the same in a separate set of pots. Before any of this, the trees were given a special tagged carbon so scientists can follow exactly where that carbon travels afterward.

What the data shows, when the seedlings are harvested, is that the defoliated Douglas-fir has transferred a measurable fraction of its photosynthate to the ponderosa pine on the other side of the thirty-five-micron mesh. The transfer occurs only through that mesh. It does not occur through the half-micron barrier. It does not occur through root contact. The carbon moves through the fungal network connecting the two species, and it moves at a higher rate when the donor has been damaged. In the manual-defoliation treatment, 6.8% of the donor's fixed carbon ended up in the receiver, a quantity the authors note is approximately equivalent to the carbon cost of seed reproduction. The defoliated tree, being killed slowly, is moving its remaining capital out across the network before the network loses access to it.

After the trees were dug up and measured, the damaged Douglas-fir had sent some of its stored food through the fungal connections to the pine next door - and it sent more food the worse the damage was. The carbon only traveled through the fungal mesh, not through root contact. The injured tree was offloading resources before it died.

The study is Song, Simard, Carroll, Mohn, and Zeng, published in Scientific Reports in February 2015, and the pots are at the University of British Columbia. The mechanism the study documents is the resource-transfer behavior of common mycorrhizal networks, a category of infrastructure that is not new in any sense relevant to engineering. It has been running underneath temperate forests for several hundred million years longer than the oldest computer.

This experiment was published in 2015 by a team at UBC. The underground fungal networks they were studying are not a new invention - forests have been running them for hundreds of millions of years.

The infrastructure has a specification.

This fungal network actually works according to a set of rules, just like a piece of engineered technology does.

A mycorrhizal network is a symbiosis between fungal hyphae and the root systems of vascular plants. Roughly 80% of terrestrial plant species form some version of this association, and the fungi colonize root cells, extend hyphal threads through the soil, and connect to other plant roots they encounter at distances ranging from millimeters to tens of meters. The fungus receives photosynthetic carbon from the plant. The plant receives nitrogen, phosphorus, and water that the fungal hyphae extract from soil regions the roots themselves cannot reach. The relationship is documented down to the molecular level: glutamine and glycine carry nitrogen across the symbiotic interface, and laboratory isotope-tracer studies have measured rates of transfer in micrograms per gram of root tissue per hour.

Mycorrhizal networks are a partnership between fungi and plant roots. Most land plants have them. The fungus gets food from the plant, and the plant gets nutrients and water it couldn't reach on its own. Scientists have measured this exchange in detail, right down to the specific molecules and transfer rates.

The transfer pattern, where it has been documented in field conditions, follows what mycorrhizal ecologists describe as a source-sink gradient. Plants in surplus condition export. Plants in deficit condition import. The rate of transfer is governed by the differential between what one node has fixed and what an adjacent node currently needs. In bidirectional carbon-transfer studies between paper birch and Douglas-fir, the direction of net flow shifts seasonally: birch sends carbon to fir during the periods when birch is photosynthesizing and fir is shaded; fir sends carbon back when birch loses its leaves and fir continues fixing through the winter. The two species are running a load-balancing protocol whose specification was written, at the protein level, sometime in the Devonian.

Resources flow from trees that have plenty toward trees that are running low - like a self-balancing system. The direction can even reverse with the seasons. This is a natural load-balancing mechanism that has been operating for hundreds of millions of years.

The original observation, in Suzanne Simard's 1997 Nature paper with David Perry and four collaborators, was the first direct measurement of bidirectional carbon transfer between trees in field conditions. The cover of that issue carried the phrase that has since traveled too far. The phrase was wood-wide web. It was meant as a useful pedagogical handle for a specific finding about isotope movement across a fungal mycelium. It became, in the subsequent twenty-five years of popularization, a claim much larger than the data ever supported.

The original 1997 paper that first measured trees sharing carbon underground coined the term "wood-wide web." That phrase was meant to describe one specific finding but ended up being stretched way beyond what the science actually proved.

What the data actually shows

Here is what the experiments actually showed, separate from the hype.

The science has moved past the metaphor in two directions at once.

The real science has moved forward in two ways at once - both confirming some things and correcting others.

A 2023 Nature Ecology & Evolution perspective by Justine Karst, Melanie Jones, and Jason Hoeksema, working at the University of Alberta, the University of British Columbia Okanagan, and the University of Mississippi, conducted a citation audit of 1,676 references made to the original mycorrhizal-network field studies in subsequent peer-reviewed papers [5, 6]. Their finding was that approximately one-quarter of the citations misrepresented the structure of the networks documented in the original work, and approximately one-half misrepresented the function. A 2009 study that used genetic techniques to map the spatial distribution of mycorrhizal fungi was, by 2022, being cited as evidence that trees transfer nutrients to one another, which the original study had not investigated. The popular wood-wide web was a literature artifact: a story constructed by sequential overinterpretation, propagated by citation patterns, and increasingly disconnected from the cautious empirical work that had given rise to it.

A 2023 research team reviewed nearly 1,700 citations to the original mycorrhizal studies and found that roughly a quarter got the structure wrong and about half got the function wrong. A study that only ever mapped fungal locations was being cited as proof that trees feed each other - which it never showed. The popular story had drifted far from the actual data through a chain of bad citations.

This is the kind of correction that mature engineering disciplines run on themselves continuously, and that ecology, when it is functioning well, runs on itself periodically. Karst's group includes researchers who had spent decades studying mycorrhizal networks; their argument is not that the networks do not exist. The networks exist. Their argument is that the specific claims being made in popular science books, in TED talks, in journalism, and in subsequent academic papers, had outrun the field experiments that the claims were nominally based on. They published the audit anyway, in the journal whose acceptance rate makes the publication an act of scientific seriousness rather than self-promotion.

This kind of self-correction is what good science looks like. The researchers doing the audit were themselves long-time mycorrhizal network scientists - they weren't saying the networks don't exist. They were saying the specific popular claims had gotten ahead of the actual field data.

What the audit leaves intact is the documented mechanism. Bidirectional carbon transfer between birch and fir is a measured phenomenon. The Song et al. 2015 defoliation study is a measured phenomenon. Resource flow along source-sink gradients between connected plants is a measured phenomenon. What the audit removes is the further claim that mature trees engage in directed, kin-recognizing parental behavior toward their offspring through the network, which the field data does not yet support. The protocol, stripped of the mythology, is more interesting than the mythology was. The protocol was never a mother. The protocol is a substrate-level mechanism for redirecting resources when one node in a connected population enters a different state than its neighbors.

The audit doesn't throw out everything. Proven findings - like trees sharing carbon and ramping up transfers when stressed - still stand. What gets removed is the idea that trees are conscious parents nurturing their young through the network. Stripped of that story, the actual mechanism is more interesting: it's a system that automatically reroutes resources whenever one connected node changes state.

A protocol designer, reading the mycorrhizal literature without the popular framing, would recognize the architecture immediately.

If a network engineer looked at how mycorrhizal networks actually work - without all the nature documentary framing - they would immediately recognize the design pattern.

Each plant in a connected stand is a node maintaining its own internal state: photosynthetic rate, water status, nutrient reserve, defense-compound concentration. The fungal network is the transport layer, with multiple parallel paths between any two nodes and no central coordinator. Resource flow is governed by gradient: a node in surplus, with high internal carbon, will leak photosynthate into the network at the boundary of its mycorrhizal interface, and the leaked photosynthate will diffuse toward whichever connected node has the lower internal concentration [3, 4]. There is no broadcast. There is no announcement. The transfer is a thermodynamic consequence of the connection itself.

Every tree is a node tracking its own internal status. The fungal threads are the cables connecting them, with no central control point. When one tree has more sugar than it needs, that sugar naturally drifts through the network toward wherever levels are lower - no message sent, no decision made. It just flows.

When a node enters a stress state, the dynamics change. The defoliated Douglas-fir in the Song study did not initiate transfer because it received a signal asking for transfer. It initiated transfer because defoliation altered its source-sink position: the photosynthate it had already fixed was now in excess of what its damaged tissues could store and use, and its connection to the network meant the surplus moved. The neighboring ponderosa pine did not request the carbon. It received what the network's gradient delivered to it.

The damaged Douglas-fir didn't send carbon because it decided to help its neighbor. It sent carbon because damage left it with more stored sugar than it could use, and the fungal connection meant that surplus naturally flowed out. The pine didn't ask for it - it just received what the gradient delivered.

A second layer of the same architecture is documented in defense signaling. The Song study measured not only carbon transfer but the induction of plant defense enzymes in the receiving ponderosa pine following defoliation of the donor Douglas-fir. The receiver, structurally adjacent on the network, began producing the chemical defenses that would protect it against an insect attack it had not yet experienced. The donor's stress had translated, through the network, into a precondition in the receiver. The mechanism is mechanical, not communicative: a coupling between two adjacent nodes through a shared substrate, with a state change in one node propagating through the substrate to alter the conditions of the other.

The network also triggered defense responses. When the Douglas-fir was attacked, the connected pine started producing its own defensive chemicals - preparing for an insect threat it hadn't encountered yet. This happened through the shared fungal substrate, not through any kind of deliberate communication. One node changed et al. 2015 also documented defense-enzyme induction in receiver pine following donor defoliation. Mechanism: state-change propagation through shared substrate - mechanical coupling, not communicative signaling. Donor stress altered substrate conditions; adjacent node responded to altered substrate state.

No layer of the contemporary internet's transport stack runs anything like this.

No part of the modern internet works this way.

The contemporary internet's transport layer is presence-only.

The internet only tracks whether a connection is on or off.

TCP maintains an open connection through keepalive packets that confirm both endpoints are still responsive. When the keepalive fails, the connection is dropped. There is no concept of a graceful state transition between active and absent. There is no source-sink gradient. There is no mechanism by which a node experiencing distress can, by virtue of its distress, alter the conditions of an adjacent node. OAuth tokens expire on inactivity timers, but the timer is calendar-based, not state-based: the token does not know whether the user behind it has had a stroke, changed jobs, died, or simply stopped opening the application. The expiration is a janitorial function applied at fixed intervals to a system that has no other notion of what to do with sustained silence.

The internet checks whether a connection is still alive on a fixed timer - and if not, it drops it. It has no idea whether the person on the other end is stressed, gone, or dead. Auth tokens expire on a clock, not based on what's actually happening with the user. The system has no way to respond to silence except to eventually delete it.

The HTTP protocol, the transport layer above TCP that nearly all consumer internet activity runs through, is stateless by design. Each request is independent. The session-state architecture that supports modern web applications, cookies, tokens, server-side session stores, was bolted on top of HTTP in the 1990s as a workaround. The workaround treats each user as continuously present until manual intervention declares otherwise. There is no protocol primitive for the user who has begun to fade. There is no diffusion gradient that automatically rebalances a system's resources away from a node that has stopped using them and toward a node that needs them.

The internet was built so each webpage request is totally separate, with no memory of you. Tracking who you are across visits was a hack added later. That hack keeps treating you as "still here" even when you've walked away. There's no built-in way for the system to notice you've gone quiet and free up your spot for someone else.

The consequence is the architectural condition documented across enterprise SaaS environments: provisioned identities accumulating indefinitely, license waste running into nine figures per organization, machine identities outnumbering human identities by ratios approaching 144 to 1, and the cost of maintaining the accumulated state being distributed across every other user of the system in the form of degraded performance, increased attack surface, and inflated subscription pricing. The forest does not have this problem. A Douglas-fir that stops photosynthesizing stops drawing on the network and starts releasing what it has already accumulated. The ponderosa pine downstream of it does not pay rent for the dying tree's continued nominal presence in the canopy.

Because of that "always present" assumption, companies pile up unused accounts and software seats forever. This wastes enormous amounts of money, creates security risks, and makes systems slower and more expensive for everyone. Trees in a forest don't have this problem - when one stops pulling nutrients, it stops getting them, and the neighbors benefit automatically.

What the foresters know

Here is what forest scientists have figured out.

The applied side of this work is being conducted in conditions the engineering trade does not normally encounter.

This research is happening in settings that most software engineers never work in.

The Mother Tree Project, established at the University of British Columbia in 2015 and now spanning multiple long-term research sites across the province, is investigating how forest management practices influence the survival and growth of seedlings under different overstory retention conditions. The project's core question, beneath the popular framing, is a protocol-design question: under what stand structures does the network's redistribution function persist, and under what structures does it collapse? Clearcuts collapse it. Fragmented overstory damages it. Wildfire of sufficient intensity erases it. The network is not infinitely resilient. It has operating conditions, and those conditions can be exceeded.

A major research project in British Columbia is studying how young trees survive when different numbers of older trees are left standing around them. The real question underneath all of it is: what causes a forest's underground sharing network to keep working, and what causes it to break down? Clearcuts break it. Fragmentation damages it. Intense wildfires erase it. The network has limits.

The work of researchers like Justine Karst at the University of Alberta sits in productive tension with the work of Simard's lab at UBC. Karst's 2015 paper in New Phytologist, with Erbilgin and other collaborators, documented transgenerational effects of bark-beetle-caused tree mortality on subsequent pine seedlings through fungal mutualisms. The mortality of one cohort altered the soil's fungal community, and the altered community produced different growth outcomes in the next cohort. Karst's group is the one that publicly challenged the popular claims about the network. Karst's group is also the one that has continued to document the specific, mechanistic ways in which absence in the network shapes the nodes that follow [5, 8]. Both labs are doing the same kind of work. Both labs are taking the absence of a node seriously enough to measure what happens because of it.

Two different research groups are studying the forest network, and they don't fully agree - but they're actually asking the same question. One group found that when trees die from beetle attacks, the fungal community in the soil changes, and that changed soil affects how the next generation of trees grows. Even the group that pushed back on some of the bigger claims still kept carefully measuring what happens to a network when a node goes missing.

This is what the engineering trade has not yet built an equivalent of: a body of empirical work, conducted across decades, on what a connected system does when one of its nodes stops contributing. The internet has the connection layer. The internet has, in fragments, the metric layer; the SaaS-management industry now generates billions of dollars annually quantifying the waste produced by absence. What the internet does not have is the redistribution mechanism. It can measure that the seat is empty. It cannot route around the empty seat in any way that compensates the rest of the system for what the empty seat is no longer providing.

Software engineers haven't built anything like this body of forest research. The tech industry can now measure how much money is wasted by unused accounts. But it still has no way to automatically redirect those wasted resources to people who actually need them. It can see the empty seat. It just can't do anything useful with it.

The single design principle

There is one key rule the forest network follows that no computer network follows.

There is one principle the mycorrhizal protocol embeds at the substrate level that no current network protocol embeds at any level.

The principle is that connection itself implies obligation to redistribute under gradient. Two nodes that are joined to a shared substrate are not, in the forest's protocol, free to maintain whatever state they choose without that state propagating through the substrate. A surplus on one side becomes a flow toward the other side automatically. A deficit on one side becomes a draw from the other side automatically. The connection is the redistribution mechanism. There is no separate accounting layer. The protocol's economics are encoded in the physics of the diffusion.

When two trees share a fungal network, the network automatically moves nutrients from the tree that has more toward the tree that has less. No manager, no software, no billing system is needed. It's physics, not policy.

The TCP/IP stack does not embed this. A node on the internet may be richly resourced and connected to a node that is failing, and the connection itself does nothing about the asymmetry. The richly resourced node continues to receive packets at the same rate it was receiving them before. The failing node continues to be charged for the same services it was being charged for before. The packets that would represent a redistribution, if redistribution existed in the architecture, do not exist, because the architecture has no concept of the resource asymmetry, no measurement of node-state differential, and no gradient to propagate along.

The internet doesn't work this way at all. A healthy server connected to a failing one keeps getting resources at the same rate. The failing one keeps getting charged the same as before. Nothing flows to fix the imbalance, because the network doesn't even know an imbalance exists.

This is not a minor design difference. This is the difference between a substrate that automatically rebalances under load and a substrate that requires every rebalancing decision to be made by a human, executed by a separate piece of software, billed to a separate account, and audited by a third party. The forest's protocol is what cybernetics in the 1950s called a homeostatic system: a system whose own dynamics keep it in a viable operating range without external supervision. The internet's protocol is the opposite. The internet's protocol generates state, accumulates state, and offers no native mechanism for any state to ever decay, redistribute, or leave.

This isn't a small gap - it's a fundamental difference. The forest self-corrects automatically. The internet requires humans, extra software, separate billing, and outside auditors just to do what the forest does on its own. Cybernetics researchers in the 1950s had a name for systems that self-regulate: homeostatic. The internet is the opposite. It only accumulates state. Nothing ever naturally leaves.

The fungal hyphae implement that self-regulating dynamic as a protocol primitive. It is the operating mode of the network, not a feature applied to it.

The fungal threads that connect trees make self-correction the default mode of the network - it's not an add-on feature, it's how the network operates.

The forest is not a metaphor.

The forest is a real system, not just a useful comparison.

The mycorrhizal network is approximately as old as terrestrial plant life itself. The earliest direct fossil evidence of plant-fungal symbiosis comes from the 407-million-year-old Rhynie chert in Aberdeenshire, Scotland, where exceptionally preserved Early Devonian land plants are colonized by arbuscular mycorrhizal fungi morphologically indistinguishable from extant lineages [10, 11]. Molecular evidence places the symbiosis still earlier, contemporaneous with the colonization of land by plants more than 475 million years ago. The protocol has been running, refining itself through evolutionary pressure on the connected populations, for a duration that is approximately five orders of magnitude longer than any human computational system has existed. During that duration it has solved, structurally, a set of problems that the contemporary internet cannot solve at the protocol layer: how to redistribute resources across a connected population when individual nodes enter different states; how to transmit defense signaling from a stressed node to its neighbors before the stress reaches them; how to permit graceful node exit through gradient release rather than abrupt connection drop; how to operate without a central coordinator on a network whose topology is constantly changing.

This underground network is nearly as old as land plants themselves - around 475 million years old. Fossils prove it. In that time, it has evolved working solutions to problems the internet still can't solve: sharing resources when nodes change states, warning neighbors before stress arrives, letting nodes exit gracefully, and running without any central controller on a constantly shifting network.

These are not mystical properties. They are mechanical properties of a substrate whose connection medium has the right physical and chemical characteristics to support them. The fungal hyphae are the medium. The medium supports the protocol. The protocol does the work.

None of this is magic. The fungal threads are the physical medium, and the medium is what makes the protocol possible. The protocol does the actual work.

The contemporary internet's connection medium does not have these characteristics. Copper, fiber, and radio do not diffuse resources along gradients. They transmit packets between addressed endpoints. The protocol layer above them inherits this constraint and cannot, without redesign, exhibit any behavior the substrate does not enable. The redistribution functions that the SaaS-management industry, the FinOps consultants, and the IAM-governance teams now perform manually are the functions a different substrate would perform automatically. The cost of running them manually is the structural tax the current architecture levies on every entity that operates inside it.

The internet's physical medium - copper, fiber, radio waves - can only send packets between two specific addresses. It can't move resources along gradients. Every redistribution job that the forest does automatically, humans in the tech industry now do manually - and those humans cost money. That cost is a tax the current architecture charges everyone.

A mycorrhizal protocol on the internet would not look like the mycorrhizal network in the forest. The substrate is different. But the design principle is portable, and the principle is currently not present in any layer of the stack the consumer internet is built on. The principle is that adjacency on the network creates a mutual obligation that the network itself enforces. The forest enforces it through diffusion. An internet that wanted to embed the principle would have to enforce it through some equivalent mechanism the silicon substrate supports. No such mechanism is currently in production.

An internet built on the forest's principle wouldn't look like a forest - the physical materials are too different. But the core idea is transferable: being connected to someone should mean the network itself holds both parties accountable. The forest enforces this through physics. An internet version would need a different mechanism suited to silicon. That mechanism doesn't exist yet.

The unbuilt specification

Here is what hasn't been built yet.

Sometime in the next several decades, the internet will either acquire a redistribution layer or it will not. If it does not, the existing waste-accumulation pattern will continue to compound, and the cost of running the absence-blind architecture will continue to be paid by the entities downstream of the absence: the security teams managing the dormant credentials, the procurement teams renewing the unused licenses, the users absorbing the degraded performance, and the institutions paying for an infrastructure that increasingly performs less of the function it was designed to perform. The forest has run this experiment already. The forest's outcome, in stands where the network has been disrupted by clearcutting or fire, is that the connected population fragments into independent nodes, and the fragmented nodes cannot, individually, do what the network did collectively.

In the coming decades, the internet will either add a way to redistribute resources automatically, or it won't. If it doesn't, the waste and the security risks will keep growing, and the people downstream - security teams, procurement teams, regular users, and entire institutions - will keep paying the price. The forest already ran this test. When its network gets broken up by logging or fire, the trees turn into isolated individuals, and isolated individuals cannot do what a connected network does.

The species that already does what the engineering trade has not done has been running its protocol for more than four hundred million years, has no engineers, no patents, no funding rounds, no roadmap, and no language for what it is doing. It has only the substrate, and the substrate runs the protocol.

A living thing that already figured out what engineers are still trying to build has been running its system for over 400 million years - with no teams, no money, no plans, and no words to describe it. It just uses the material it lives in, and that material does the work.

The mesh in the British Columbia experiment is what the protocol designers in 1969 left out. The mesh in the experiment lets carbon move between two organisms whose roots never touched. The cable between two endpoints in 1969 did not. Half a century of subsequent protocol layers have inherited the absence and elaborated on it. The species that solved the problem is in the soil under the cable, doing what the cable cannot, on a timescale that predates the cable by a factor of one hundred million. What does the engineering trade owe a substrate it has spent its entire history standing on top of without ever asking what runs underneath?

The forest network experiment in British Columbia showed something the people who built the early internet missed - living things can share resources through soil without ever physically touching. The original internet just connected two points with a wire. Everything built on top of that wire inherited the same blind spot. Meanwhile, the living system that already solved this has been doing it in the dirt under those wires for 100 million times longer than the wires have existed. The engineering world has always built on top of nature without once stopping to ask what nature is already running beneath it.