Robert Paine spent the summers from 1963 on prying starfish off the rocks at Mukkaw Bay, Washington, and throwing them into the sea. On the stretch of shore he left alone, fifteen species shared the intertidal zone. On the stretch he cleared of starfish, mussels took the rock within a few years and the count fell to eight. He removed one species and lost seven.
That experiment, published in 1966, is where keystone species begins; Paine gave it the name three years later. A keystone species is a species whose effect on its ecosystem is large out of proportion to its abundance. Remove it and the web does not shrink by one. It reorganizes.
Sixty years on, the mechanism is old and well replicated: a species' effect on its web can be far out of proportion to its numbers. Identifying which species that is, in a system nobody has experimented on, is still contested. Mary Power and colleagues titled their 1996 definition paper Challenges in the quest for keystones for a reason, and the starfish's own effect varies with the shore it sits on. Finance has absorbed neither the mechanism nor the difficulty. The instrument most nature-risk work still starts and ends with is the species list. A species list is a headcount. It tells you who is present. It does not tell you who is holding the place up.
what paine found on the rocks
Paine's shore had a predator, the ochre starfish Pisaster ochraceus, eating mussels, barnacles, limpets, and chitons. The starfish was never the most numerous animal on the rock. But its favorite food, the mussel Mytilus californianus, was the strongest competitor for space, and the starfish was the only thing keeping it in check. Take the starfish away and the mussel bed spread until it crowded out the barnacles, the algae, and the grazers that lived on the algae. Fifteen species became eight. Paine named the pattern in a 1969 note: a keystone, like the wedge-shaped stone at the top of an arch, is small relative to the structure and the structure comes down without it.
Eight years later, James Estes and John Palmisano compared Aleutian islands with and without sea otters. Islands with otters had kelp forests. Islands without had urchin barrens: otters eat urchins, urchins eat kelp, kelp shelters everything else. The sea otter is the rare case where the animal on the poster is also the animal carrying the load. That overlap is useful and it is also a trap. It teaches people to look for keystones among the species they already love.
importance is position, not mass
Power's 1996 paper gave the field its working definition: a keystone is a species whose impact on its community is large, and disproportionately large relative to its abundance. The second clause is the whole point. Abundance is a headcount. Impact is a position: how many other species depend on this one, and whether anyone else can do its job.
That is a graph property. In a food web the starfish sits above the mussel, the mussel sits above the rock, and everything else on the rock depends on the mussel not winning. In a landscape, a small patch of habitat sitting at the only crossing between two larger ones can matter more than either of them; Dean Urban and Timothy Keitt showed in 2001 that connectivity, not area, often decides which patch you cannot afford to lose. Same logic, different scale. What a node is worth to the system is set by where it sits and who leans on it.
Charisma is a different thing. The species on the brochure is chosen for being large, visible, and easy to love. The species carrying the load is chosen by the web, and the web does not care about photographs. Sometimes the two coincide, as with the otter. Often the load sits on a dung beetle, a fungus, an ant, or a pollinator nobody has photographed, and rarely on one species alone. Ecologists call a group of species doing the same job in the web a guild: the nectar-feeders, the decomposers, the seed dispersers. The guild is usually the right unit to think in, because it is the job, not the name, that the rest of the system leans on.
| poster species | load-bearing position | |
|---|---|---|
| what it is | the animal that makes people care | the node whose removal reorganizes the web |
| how you find it | by looking | by reconstructing who eats, pollinates, disperses, and decomposes whom |
| how it is counted | headcount, range, red list category | connections, and what depends on them |
| what its loss does | grief and a headline | a cascade: lost pollination, lost recycling, a different forest, a different watershed |
| overlap with the other column | sometimes (sea otter) | often a species with no constituency |
| what to fund | the individual | the guild doing one job the system cannot do without |
a species list is a headcount
Species lists have never been cheaper. Environmental DNA, camera traps, and bioacoustics now produce inventories at a scale field biologists in Paine's era could not imagine. What those inventories report is presence: who was detected, how many kinds, sometimes how often. Richness, diversity indices, occurrence maps. All headcounts.
Two lists can carry the same two hundred names and describe two different systems, depending on how those names are wired to each other. A list can also report a system close to collapse as healthy, because the count has not moved yet. The mussel bed on Paine's cleared shore was, for a while, a thriving population of mussels.
None of this is news to practitioners. Land trusts have prioritized connectivity over counts for two decades; The Nature Conservancy's Resilient and Connected Network is built on exactly the insight Urban and Keitt formalized. IUCN's Green Status of Species, launched in 2021, scores how much of a species' ecological function remains, not only its odds of extinction. The functional axis exists, and the people closest to the land are usually the first to see it.
What that axis lacks is a financial channel. There is no standing way to turn "this guild holds the canopy over that reservoir" into a named source a beneficiary can fund. That is the gap the pillar of this series names: stock metrics tell you how much nature remains; the graph tells you whether the service keeps arriving. A species list is a stock metric. Condition is which nodes carry the load. If your nature-risk work stops at exposure, a list is where it stops too.
from a list to a map
Jay Gutierrez, a systems biologist by training, builds graph intelligence for living systems. His public engine for this question is AI Ecologist, and his one-line description of it is the ask this post has been building toward: turn a species list into a map of what holds an ecosystem together, and where it breaks first.
His public brief describes three moves. Reconstruct the interaction network from the species inventory alone, with no time series required, benchmarked against known food webs at a link-prediction accuracy near 0.95. Locate where that network is structurally fragile: the keystones and single points of failure a count cannot reveal. Then route the result to a decision, whether restoration priority, watershed protection, or a nature-finance screen. He also profiles structural fingerprints across a library of reference ecosystems, currently twenty across six continents, and labels the cross-biome comparison indicative: a structural similarity, not a calibration.
The example he leads with is a cloud-forest basin that feeds the drinking water of a city of millions. On the species list, the apex predators are the obvious keystones. In the reconstructed web, they are not in the load-bearing core. Fourteen nectar-feeders sit deeper in the structure than the puma, eleven of them hummingbirds, and the species with the highest structural impact are the soil's recyclers: a dung beetle and the leaf-cutter ants. The canopy, and the watershed it stabilizes, rests on line items most inventories would scroll past.
It also rests on people. That basin is inhabited and governed. The communities who live in and hold the forest decide what happens to it, and they often already know which flowers and which soil animals matter. A reconstruction is a targeting layer over that knowledge, not a replacement for it, and a model output is not a mandate.
He is careful about what that means, and so should anyone quoting it. The flagship network is fully reconstructed, with no observed-interaction tier yet, so he labels its structural reading exploratory by construction. The reconstruction and fragility layer is benchmarked and reproducible; the financial outputs are prototype, screening-grade, and not market-calibrated. It is a targeting layer, not a field census.
We collaborate with Jay. The engine is his. What we take from the work is the question it makes answerable: not how many, but who carries the load, and where it breaks first.
what this means if you steward land, write policy, or move capital
If you steward land, the inventory you already hold is worth more than you are using it for. The next question to ask of it is structural: which guild does the job you are managing for? Pollination, decomposition, seed dispersal, water retention. The species doing that job are the ones whose loss would change your land, whether or not any of them is rare. Protect the guild, not only the individual.
If you write policy, listings and recovery plans are organized by species and by extinction risk, which is correct for their purpose. Keystone position is a separate axis. An abundant dung beetle will never be listed, and it may be the one thing holding a canopy over a reservoir. IUCN's Green Status already gives you that second axis; State Wildlife Action Plans and habitat-condition baselines are where it can be written in.
If you move capital, the poster species is a marketing asset and the load-bearing guild is the risk asset. A water utility's treatment bill leans on a forest; the forest leans on its pollinators and recyclers. That is where a dollar of protection buys the most avoided loss, and it is rarely where the dollar currently goes. It buys it only with the consent of, and payment to, the people who hold and steward that forest.
On ensurance, a place, a people, or a purpose can be held by an agent, an onchain account with its own wallet, and funded through a certificate tied to that agent. The money funds the present condition of a named source; it is not a payout after loss. The guild is not a new instrument. It is how you decide what the funding is for: which place, and which job within it, the money protects. Our stage is honest: live agents, live certificates, small volumes. Jay's engines are not wired into our pricing. What his public work does is make the job nameable, and a named job on a named place is something a beneficiary can fund.
frequently asked questions
what is a keystone species?
A keystone species is a species whose effect on its ecosystem is large out of proportion to its abundance. Robert Paine coined the term in 1969 after showing that removing one starfish species from a Washington shoreline cut the community from fifteen species to eight. Keystones are defined by position in the food web, not by size, charisma, or numbers.
how is a keystone different from an endangered species?
Endangered measures a species' own odds: population size, range, rate of decline, as assessed under the IUCN Red List criteria. Keystone measures what happens to everyone else if the species goes. A species can be both (the sea otter), one without the other (a rare orchid few other species depend on; an abundant dung beetle the canopy depends on), or neither. Extinction risk is about the node. Keystone status is about the edges. For what an endangered listing does and does not tell you, see the fish that has to keep floating.
can you tell what holds an ecosystem together from a species list?
Not from the list alone. A list reports presence, richness, and status; it does not report who depends on whom. You can, however, reconstruct the interaction network behind a list from established ecological evidence, and then read the structure for which species are load-bearing and where the web is thin. That is the public method Jay Gutierrez's AI Ecologist describes. It is a benchmarked targeting layer, not a census, and its financial outputs are screening-grade.
taking action
The species on the poster got you to care. The guild carrying the load is what your protection has to reach.
- Find what is already held — natural assets on ensurance, the places and systems with an account behind them.
- See who depends on whom — explore places by their linkages, not their headcount.
- Fund a named source — specific ensurance, certificates each tied to one agent holding a place, people, or purpose.