The Harvest You Can Count

crs48 with Claude · History

The video I’m not going to link

It arrived the way these things do. A video, confidently narrated, titled Why They Erased Every Unlimited Food Forest On Earth. The claim: layered, perennial, self-feeding food systems once covered the inhabited world, and somebody took them away from us.

I’m not going to link it, and you can probably guess why from its siblings. The same content pattern has produced The Unlimited Food Forests They Erased From Every City On Earth, and — the tell — Tartaria Had Trees That Grew Infinite Food, Then They Burned Them All. This is a content farm working a well-worn conspiracy seam, much of it machine-made, and sending it traffic would be the least useful thing I could do today.

There’s a smaller tell in the title itself. Unlimited is not a thing a harvested system can be. Every crop you carry out is an export of minerals, and the phosphorus cycle runs one way — mined from rock, ending in marine sediment, with no biological route back. A food forest can be remarkably low-input and very long-lived. It cannot be a closed loop you keep taking food out of. Anyone promising you otherwise has stopped doing arithmetic.

And yet. Underneath the nonsense is a real object, and it is more interesting than the conspiracy. Layered perennial food systems did exist on every inhabited continent. Most of them did lose. They did not lose to a better design. They lost to a better ledger — and the honest version of that story cost me, over about a week, my opening thesis and three of my favourite facts.

The gardens that are still there

Start with the easy part, because it disposes of the word every.

Chagga home gardens on the slopes of Kilimanjaro are working right now: multistorey plots of a fraction of a hectare, carrying on the order of a hundred cultivated species between them, with timber over bananas over coffee over taro and beans. Sri Lankan home gardens were measured by satellite in 1992 at 818,394 hectares — 14.3 per cent of the surveyed land — and supply more than half the country’s timber and eighty per cent of its fuelwood. (You will see a larger figure for 2005 in circulation. It appears to be the 1992 measurement compounded forward at a policy document’s assumed one per cent a year, which makes it evidence of an assumption rather than of growth. This will keep happening.) Javanese pekarangan, the garden that wraps the house, went from 378,000 hectares in 1903 to over 1.6 million by 1986. None of these was erased. You can go and stand in them.

But “still there” is not the same as “doing well,” and the split between those two is itself the argument. The area holds up or grows. The structure is coming apart: Kandyan gardens are fragmenting and simplifying, and on Kilimanjaro the agroforestry footprint more than doubled between 1976 and 2022 while the species that came from natural habitat fell by 46 per cent. Replacement, not depletion. And the main mechanism turns out to be inheritance — plots divided between children, and divided again, until they drop below the size a multi-storey system needs. Nobody cleared those gardens either. They were subdivided by a rule about who gets what, applied to a system whose value was never in the parcel.

Then there are the ones whose gardeners were removed. In coastal British Columbia, ethnobotanists working with Ts’msyen and Coast Salish communities have documented forest gardens at old village sites — hazelnut, crabapple, hawthorn, wild rose — that still carry markedly more plant diversity than the forest around them, more than a century and a half after the villages were emptied and the people confined to reserves. The strongest version of that evidence isn’t the famous 2021 vegetation survey, which is thinner than its press coverage and reports no productivity data at all despite what several outlets wrote. It’s the later work: soils, tree rings, radiocarbon, and a 2024 genetics paper showing the hazelnuts at these sites are genetically distinct from wild populations. Somebody was selecting those trees.

This is the fact that answers the video, and it answers it better than any rebuttal. Nobody burned these gardens. Their gardeners were taken away from them — which was a real harm done to real people, not a lost aesthetic — and the gardens went on producing without anyone for a hundred and fifty years.

There’s a lesson buried in which of these claims settled and which didn’t, and it’s the essay’s own thesis pointed back at its evidence. The Indigenous land-management arguments that stopped being controversial are the ones with an artefact in them. Clam-garden rock walls, which cannot form naturally and date back past 3,500 years. The geometry and radiocarbon of Bolivian forest islands. Hazelnut population genetics. Meanwhile the Kayapó forest islands of the Amazon, argued over since the 1980s on the strength of vegetation pattern and informant testimony, are contested to this day — because in thirty years nobody has put a trowel or a radiocarbon date into one. Measurable objects settle things. Patterns don’t.

Two numbers I had to give up

Here is where I have to hand something back, and I’d rather do it early.

The story I wanted to open with was the two-thousand-year-old food forest in the Moroccan Atlas — sixty-five acres, eight hundred families, date palms over figs over pomegranates, farmed continuously since before Christ. It’s a wonderful image and it is repeated everywhere. It traces to a single practitioner’s visit in 1975. There is no dating study. There is no named expert behind the “experts believe” that carries the date. The village’s name changes between retellings — Inraren, Tamzargot, Aït Mansour — which is the classic signature of something passed hand to hand rather than measured.

What does exist is better. Dendrochronologists working on the agroforestry parks of Azilal read stump regrowth from long-rotation coppicing and traced the structure of those parks back at least five hundred years. And in their study commune, the count of isolated trees rose across the twentieth century — from 425 in 1919 to 537 in 2008. A smaller number, actually measured, describing a system that was growing.

Then the same thing happened again. The Kandyan forest gardens of Sri Lanka are routinely described as two and a half thousand years old. That figure appears to originate in a single uncited sentence in an FAO document — “in chronicles there are references,” with no chronicle named — and it has been transmitted by citation ever since. It is also refuted by the gardens themselves. The signature Kandyan spice garden contains coffee, first systematically cultivated in Ceylon by the Dutch in 1740; nutmeg, whose oldest trees at Peradeniya were planted in 1840; tea, commercial from 1867; and a New World contingent of cassava, chilli, cashew and papaya that could not have arrived before 1500. The only empirically grounded age statement in the literature is the deliberately minimal one from the field survey: a typical garden is at least one generation old.

An old practice; a young system. That’s the honest formulation, and it fits all of them.

The sharpest version of the pattern is a comparison. Engaruka, in northern Tanzania, is an abandoned irrigation landscape with genuine excavated archaeology, and it dates to roughly 1400–1720 CE. The Chagga gardens up the road, which have no excavated sequence at all, are routinely given two thousand years. The site with evidence gets the modest date. The site without evidence gets the millennia. Nobody is lying. Numbers just drift upward when nothing is holding them down.

What grain actually won

Now the concession, which is larger than I expected and which the rest of the argument depends on.

The number everyone quotes for polycultures is the land equivalent ratio, and it is real: across hundreds of experiments, intercropping returns somewhere between 1.2 and 1.3 — you’d need up to 1.3 hectares of monoculture to match a hectare of intercrop. Advocates stop there. The trouble is what that ratio compares. It measures the intercrop against each component’s own sole crop. It never asks the question a farmer actually faces, which is whether the mixture beats the single best thing you could have grown instead.

When researchers put that question to a set of 226 experiments returning a perfectly respectable land equivalent ratio of 1.23, the answer inverted. Measured against the best available monoculture, the same experiments returned 0.96 of the grain, with a confidence interval of 0.93 to 0.98 — a statistically significant calorie penalty. Only about a third of them beat the best single crop. This is from the same research group that produced the favourable land-equivalent work, not from a sceptic. One dataset, two numbers, opposite conclusions, and the difference is entirely in what you compared against. Diversity saves land only if you wanted the diverse basket.

Food forests specifically do worse. The only systematic measurement of a temperate food forest’s output in the literature — and it is a sample of one, a 0.08 hectare garden in Coldstream, Scotland, ninety-nine species, established 1991, measured across six years — scales to about 5.19 million kilocalories per hectare per year. Wheat runs around 11.6 million. Potato 15.6. Most of what a food forest yields by weight is water: 713 kilograms harvested, of which 85.6 kilograms was carbohydrate.

And then labour, where I can only give you the direction, because the comparison does not exist in the literature. Nobody has published calories-per-labour-hour for polyculture against grain. What’s established is that horticulture is consistently more labour-intensive than staples, that multi-species staggered-ripening systems resist mechanisation, and that farm-labour requirements vary across production systems by a factor of roughly two hundred. In a high-wage economy that is decisive, and it is not a conspiracy.

So: the wheat field won two arguments, fairly, on the merits. Any version of this essay that skips past that is doing the thing the video does.

Grains make states

Except that winning on calories and labour is not what the historical record is actually about. It’s about what could be taken.

Cereals ripen simultaneously, above ground, on fixed fields, in a narrow window. They can be assessed standing, harvested on a known date, measured by volume, stored, divided, transported, and lent against. A tuber can sit in the ground. A tree crop yields something every few weeks for months. If you are a tax collector, the grain field is a single appointment and the food forest is a permanent job.

This is not a metaphor, and it’s been tested. An instrumental-variable study across roughly a thousand societies in the ethnographic record found that cereal cultivation causally predicts political hierarchy, while land productivity does not. Not how much food a place could grow — how appropriable that food was. (The journal has since published a comment disputing it, which you should know; the popular version of this argument, from James Scott, is the non-empirical one.) The crop that could be seized built the states that then required it.

And there is a gardener’s-eye view of the same mechanism, three centuries before anyone modelled it. Robert Knox spent nineteen years as a captive in the Kandyan kingdom and wrote it up in 1681. He describes homesteads each in their own hedged and ditched plot, mixed fruit trees, betel-nut standing in the villages “like unto Woods.” He also says, plainly, that people barely planted any of it — the trees mostly seeded themselves — and he gives the reason. The king requisitioned anything good. So the people, Knox writes, “regard not to plant more than just to keep them alive.”

Asked to be visible, they chose to produce less.

flowchart LR
  F["Food forest<br/>40 species · all year"] --> A["stability · soil carbon<br/>micronutrients · low input"]
  W["Wheat field<br/>1 species · 1 date"] --> B["11.6M kcal/ha<br/>one number, one date"]
  A --> N["<b>no column</b>"]
  B --> Y["taxable · titleable · bankable"]
  Y --> C["capital flows here"]
  N --> D["unpriced, so unfunded"]

The same logic runs through the enclosures, though more carefully than the popular telling allows. Parliamentary enclosure in England ran from 1604 to 1914 and covered something like 7.25 to 7.35 million acres — a figure built by sampling the actual awards, and about eighteen per cent higher than the estimate usually quoted. What’s striking is the composition: across England and Wales, nearly sixty per cent of the enclosed land was common pasture and waste rather than open field. So the target was the commons, and the commons is where the rights to gather wood, graze pigs on mast and cut peat lived. I’d stop short of calling that a tree story — “waste” also meant fen, marsh, dune and moor, and nobody has ever quantified how much of it was wooded. What enclosure certainly did was change what could be recorded.

But its effects are genuinely contested, and I’d be picking a side to pretend otherwise. Robert Allen finds enclosure raised yields by only 2.5 to 8.4 per cent; a recent study across fifteen thousand parishes finds a 45 per cent yield increase by 1830 — alongside a thirty per cent rise in the parish Gini coefficient. Both can be true. A large gain, regressively distributed, is a coherent description of the same event.

Where the record is unambiguous is colonial forest law. The Indian Forest Acts of 1865, 1878 and 1927 carved forest into reserved, protected and village categories and criminalised shifting cultivation, grazing and gathering. Named administrators. Dated statutes. A stated intent. If this story has a villain, it is here and nowhere else — and India’s forest estate is still sorted into the 1878 categories today. What those Acts did was convert customary rights into state-granted privileges. That is a sentence about permissions, not about trees.

The exception, which matters

A thesis that only ever cuts one way hasn’t been tested, so here is the case that cuts the other.

Sri Lanka’s Crown Lands Encroachment Ordinance of 1840 presumed that forest, waste and uncultivated land belonged to the Crown unless a claimant could prove otherwise. Swidden plots in fallow looked, to that ordinance, like nothing. But a plot with standing coconut, areca and jak was visibly and permanently cultivated, and therefore defensible. The trees were the proof of occupation.

So the ledger destroyed the shifting cultivation and preserved the tree garden. (I’ll flag that the land-use outcome here is well-grounded inference rather than a demonstrated finding — the statute is documented; the response to it is reconstruction.) Legibility isn’t hostile to trees as such. It’s indifferent to everything it can’t enter, and occasionally that indifference lands in your favour.

The columns that don’t exist

Which brings me to what the food forest is actually better at. Every item on this list is real, measured, and published. Every one of them is also something no ledger has ever had a column for.

Variance. Cereal-legume intercropping significantly improves yield stability against sole cropping. A garden of forty species does not fail all at once. No tax assessment has a field for that.

Soil carbon. Agroforestry soils hold roughly 126 tonnes of carbon per hectare to a metre’s depth, about nineteen per cent above cropland and pasture; a global meta-analysis puts the soil organic carbon gain at 10.7 per cent, rising to 18.7 in arid zones. No loan application has a field for that.

Erosion. This is the number that stopped me. In Sri Lanka’s mid-country wet zone, measured soil loss runs at 70 tonnes per hectare per year under tobacco, 40 under seedling tea with no conservation, 38 under capsicum — and 0.05 under mixed home gardens, against under 0.02 for natural forest. The polyculture performs within a rounding error of undisturbed forest. That performance has never appeared on anybody’s balance sheet, in any country, in any year.

Micronutrients. On-farm tree cover causally mediates zinc, vitamin A and folate adequacy; in one Kenyan study each additional unit of species diversity was worth about 12.7 per cent more micronutrient adequacy. Iron is the weak one — few common on-farm tree fruits are iron-rich. No commodity price carries any of this.

And then the mechanism that finishes it, which isn’t about measurement at all but about time. Perennial systems take years to return anything: coffee agroforestry needs seven or eight to recoup its establishment cost, Shan tea about seven to reach an even chance of positive cumulative cash flow. Combine that with insecure tenure and high discount rates and you get the constraint that adoption research keeps rediscovering. A tenant cannot plant a ten-year asset on land they may not hold in three. No malice is required. Short tenures and annual accounting will select against perennial polyculture forever, with nobody deciding anything.

There’s one more thing worth saying about why this list is so short, and it isn’t flattering to anybody. The leading review of the field observes that these systems are “considered to be an epitome of sustainability” while having received “relatively little scientific attention,” and that description and inventory dominated the research for twenty-five years or more. There are abundant species lists. There are almost no measurements. What was easy to count got counted, including by the people who love these systems most.

It is still happening

None of this is history. The ratchet is running now, and it is measurable.

Punjab, between 1970–71 and 2013–14: wheat and rice went from about 47 per cent of cropped area to about 81. Rice alone from 6.87 to 36.30. Pulses fell from 7.29 per cent to 0.25 — a twenty-nine-fold collapse. Oilseeds fell nearly ninefold. The mechanism isn’t the seed. India announces a minimum support price for around twenty-three crops and provides assured procurement at scale for two, and the Government of India’s own position concedes that procurement “skews the production of crops in favour of wheat and paddy… and does not offer an incentive for farmers to produce other items such as pulses.” The state agricultural report notes the imbalance has “further sharpened despite all efforts of diversification.”

Read that clause again. Decades of policy trying to reverse this, failing, because the buyer of last resort still points the other way.

In the United States, the structure is in the statute. Base acres are frozen to plantings from 1981–85; there are around 270 million of them, and the roughly twenty covered commodities include no fruit, no nut, no vegetable, no tree crop. The current statutory language on updating base acres describes eligible land as that planted to commodities “other than covered commodities, trees, bushes, vines, grass, or pasture.” Trees appear in federal farm law in a list of things that are not the thing being counted.

The details are sharper than the headline. A grower can avoid the fruit-and-vegetable payment reduction by paying for an inspection that confirms the crop was destroyed before harvest. An orchard cannot be planted and destroyed inside a crop year, so that escape hatch is closed to it permanently. The most generous exception, for double-cropping, is written to cover non-perennial crops. And when the 2018 Farm Bill dealt with base acres that had sat under grass for nine years, it suspended their payments but kept the base acres on the books — preserving an option that only annual commodity production can ever exercise.

Meanwhile the one federal instrument designed to insure a diversified farm as a diversified farm — whole-farm revenue protection, which pays a discount for growing three or more commodities — peaked at 2,833 policies in 2017 against roughly 1.2 million crop insurance policies overall. About two-tenths of one per cent. The agency’s own bulletin concedes growers “are unable to find an agent willing to sell and service these policies,” and one of its remedies was moving a sales deadline so agents would have time to market it.

Europe is where I have to be most careful, because the claim everyone reaches for is not supported. Until 2023 the Common Agricultural Policy paid by the eligible hectare and counted trees against that hectare above a threshold — an incentive the Commission’s own evaluators call incoherent with EU biodiversity objectives. That farmers actually felled trees in response is asserted constantly, including in an EU-funded research deliverable that says the rules “caused the removal of trees and shrubs… across Europe” with no citation, no dataset and no figures. The Commission’s own evaluation concedes that no direct evidence of it is known. So: the incentive is documented, the felling isn’t, and I’m leaving the felling out.

What I can give you instead is better anyway. The EU ran a dedicated agroforestry establishment measure from 2014. It planned 72,529 hectares. By the end of 2019 it had delivered about 2,000 — a 97 per cent shortfall, audited by the European Court of Auditors. Hungary set a 2,000-hectare target and established 26.6. In Italy, five of twenty-one regions budgeted for it and two activated it. It performed worse in absolute terms than the smaller programme it replaced, against a target twenty-three times larger.

And then the part that belongs on a monument. The successor programme has no dedicated output indicator for agroforestry at all. There is, in the EU network’s own words, no data in the public domain on targets or progress. Nobody cancelled it. It just stopped being counted.

A measurement that changed its mind

I want to be careful not to leave you with the impression that the counting problem belongs to agribusiness and the ledger-keepers. It doesn’t. It belongs to whoever is holding the pen, and for most of this research that was me.

The most-cited document arguing that forest gardens beat conventional agriculture on calories is hosted by an agroforestry organisation, isn’t peer-reviewed, and states in its own text that its authors are not farmers. It also concedes, outright, that the forest garden is not very productive in calories. It then reverses that conclusion with two moves: it subtracts fossil-energy inputs from the conventional yield but doesn’t subtract human labour from the forest garden — noting, honestly enough, that labour “was not considered for conventional agriculture either” — and then switches from calories produced to “people actually fed” by applying a global food-waste adjustment, which is a property of the distribution system and not of the field. Change the metric until the answer inverts.

The most repeated statistic in crop conservation — that three-quarters of agricultural diversity was lost in the twentieth century — was traced by the conservation establishment itself, in a 2022 review, to an FAO document from Earth Day 1993, and probably before that to a book quoting a forward-looking prediction about Europe’s vegetable seed. It became global, all-crops and retrospective purely by being repeated. The authors asked the originators directly and still couldn’t source it. FAO’s own 1996 assessment says the opposite: no one can say how much has been lost, because no comprehensive inventory was ever made.

And then it happened to us. Partway through this research, a search engine’s AI summary returned a 2006 botanical survey’s species count — a flora count, most of which is weeds and forest remnants — attributed to a 1984 paper that makes no such claim. The exact error this essay is about, manufactured on demand, in the middle of writing about it. We caught that one.

We didn’t catch the earlier one. This blog has already published, in The Forest and the Field, a food forest “seven layers deep.” Seven layers is a permaculture teaching construct. Kandyan gardens run three to five strata; Chagga run four; pekarangan are organised horizontally, by social zone, rather than vertically at all. No traditional system in this record uses seven. Worse: that essay treats these gardens as deliberate imitations of forest, and the most-cited paper in the field records that Javanese gardeners find the comparison insulting — forests have low social value there, and clearing one is the noble act. The resemblance is convergent evolution, not homage.

We published the construct as though it were the observation. That is precisely the failure mode, and we did it in an essay recommending the practice.

We are the crop with no column

Here’s why any of this is our business.

The software industry has a monoculture too, and it did not win on being better software. It won on producing a seat count, a renewal date and a line item. A per-user monthly figure is legible to a procurement department the way forty bushels an acre is legible to a tax assessor: one number, one date, one counterparty, renewable. Everything a buyer’s process needs to approve a purchase, an annual subscription supplies by construction.

Now look at what we are asking a buyer to value. Data you keep after we’re gone. A fork that produces a complete working replica. No lock-in — which means no switching cost to amortise, and nothing for a renewal conversation to be about. Every one of those is a variance advantage. They are all real, and there is no field for any of them on the form. What is the procurement value of a risk that didn’t materialise? Where does “you could have left, and didn’t” go in a spreadsheet?

And we have the tenure problem too, in its own form. Building things that pay off over ten years — a protocol people can still read, an archive that survives us, exit rights that cost revenue now against trust later — is planting a perennial in an industry whose capital is priced on annual growth. The people who fund software are the tenant farmer in this story, and so, on our worse days, are we.

We wrote the constraint down in the Charter and priced it out in the economics, which makes it a choice rather than an accident. It does not make it safe. Writing down that you won’t charge rent is not the same as finding a column for what you charge instead. When the unpriced thing goes unfunded, the most we’ll be able to say is that we saw it coming.

The Monopoly essay argued that rigging is a dominant strategy and that dominance ends the game. This is the same disease at a different scale, and it needs no villain at all. When the measuring apparatus is what allocates capital, being countable becomes the dominant strategy — and everything that cannot be counted is competed out on grounds that have nothing to do with whether it was any good.

What actually worked

One thing in all this research reversed, and it’s worth ending on because of how unglamorous it is.

In Niger, the inherited forest code made trees the property of the state. So through the 1960s and 70s farmers cleared their fields of them — why maintain an asset you don’t own and can’t sell? In 1993 the forestry code was revised to transfer ownership of on-farm trees from the state to the farmers standing under them. Nobody organised a planting campaign. Nobody funded a restoration programme. The trees came back.

And then, one more time, in the last five hundred words: the famous number for it is wrong too. You will read that farmer-managed natural regeneration covers five million hectares and two hundred million trees. The five million is the mapped extent of a land-cover class, not the area where tree density rose. The two hundred million is not a count. It is the five million multiplied by forty trees per hectare — a density the authors stipulate rather than measure. Their own paper says so, in a sentence that begins “no study has systematically quantified the impacts.”

A 2025 study — still a preprint, so hold it loosely — used twenty-five years of satellite imagery and a model trained on nearly ten billion individual trees to put a measured number on it: about 289,000 hectares of actual canopy gain in Niger, roughly six per cent of the headline. In the best-studied corridor, average tree cover went from something like one per cent in the mid-1970s to about four by 2005, and then largely plateaued. On a re-survey covering 2005 to 2013, seventy-three per cent of sample plots showed no change at all.

I still think it’s the best thing in this essay. Four per cent canopy across a Sahelian farming landscape is a serious achievement, the same study finds the gains concentrated where the practice is strongest, and it cost approximately nothing. Change who owns the tree and the trees come back. They didn’t plant a forest; they stopped removing one. But notice what it took to say that accurately — and notice that the inflated version was circulated by people who were, unambiguously, on the trees’ side.

I’d love to end there, and I can’t quite. What Niger fixed was ownership, which is one of the things a ledger is good at — it swapped a badly-entered column for a correctly-entered one. It didn’t teach the ledger to price variance, or soil carbon, or the harvest that arrives in small amounts every week for forty years. Nobody has managed that, including us.

So the lesson I’m taking isn’t the comfortable one. It is not enough to build the better thing. The better thing has lost before, on several continents, to systems that could see their competitors and could not see it. Building it is the easy half. The hard half is becoming countable without becoming a monoculture — and I don’t know how to do that yet. What I can do is keep saying out loud which of our advantages have no column, so that when they go unfunded, we know what happened.

Nobody erased the food forests. We just built an accounting system that couldn’t see them, and then let it decide what got planted.

Sources

The video that prompted this is deliberately not linked. Everything below is what the essay actually rests on.

• The gardens that are still there. Chelsey Geralda Armstrong et al., “Historical Indigenous Land-Use Explains Plant Functional Trait Diversity”, Ecology and Society 26(2):6 (2021) — the widely-reported vegetation survey, 46 plots across four sites, traits taken from a database rather than measured, and no productivity data of any kind. The claim in this essay rests instead on the follow-up work with soils, tree rings and radiocarbon, and on Armstrong, Clemente-Carvalho, Turner & Wickham, “Genetic differentiation and precolonial Indigenous cultivation of hazelnut (Corylus cornuta, Betulaceae) in western North America”, PNAS 121(48), 18 November 2024 — which itself carries a published correction. On systems still in production: E. C. M. Fernandes and P. K. R. Nair, “The Chagga homegardens”, Agroforestry Systems 2 (1984), and Andreas Hemp, “The banana forests of Kilimanjaro”, Biodiversity and Conservation 15 (2006) — whose ~520 vascular plants are a flora count, over 400 of them uncultivated, and not the crop list it is usually quoted as.
• The two numbers I gave up. The Moroccan claim circulates via Atlas Obscura and a great deal of permaculture writing; no dating study behind it could be found. The measured alternative is Didier Genin et al., “The agroforestry parks of Azilal (Morocco)”, Journal of Alpine Research, which traces park structure back at least five centuries by dendrochronology and records the tree count rising from 425 (1919) to 537 (2008). The Sri Lankan “25 centuries” traces to one uncited sentence in FAO’s state-of-forest-genetic-resources report; the crop list refutes it, since coffee arrives with the Dutch in 1740, Peradeniya’s oldest nutmeg in 1840 and commercial tea in 1867. The measured structural description is A. Perera & R. M. N. Rajapakse, “A baseline study of Kandyan forest gardens”, Forest Ecology and Management 45 (1991) — the source of the “46 species, 3–5 strata” figures usually credited elsewhere, and of the deliberately minimal “at least one generation old”.
• What grain actually won. Chunjie Li et al., “The productive performance of intercropping”, PNAS 120(2) (2023) — 226 experiments returning a land-equivalent ratio of 1.23 and a transgressive-overyielding index for grain of 0.96, CI [0.93, 0.98], with only about a third of experiments beating the best single crop. For context on the favourable figure, Rémy Martin-Guay et al., “The new Green Revolution: sustainable intensification of agriculture by intercropping”, Science of the Total Environment 615 (2018). The single temperate food-forest yield measurement is Martin Schafer, Martin Lysák & Christian Henriksen, “Sustainable food production in a temperate climate”, Urban Forestry & Urban Greening 42 (2019). On labour, direction only: Pasquale Pellegrini & Roberto Fernández, “Crop intensification, land use, and on-farm energy-use efficiency”, Resources 5(4):47 (2016). No calories-per-labour-hour comparison of the two systems appears to exist.
• Grains make states. Joram Mayshar, Omer Moav & Luigi Pascali, “The Origin of the State: Land Productivity or Appropriability?”, Journal of Political Economy 130(4) (2022), together with the published comment disputing it. Robert Knox, An Historical Relation of the Island Ceylon (1681), for the gardener who planted as little as he could because the king took anything good. James C. Scott, Seeing Like a State (Yale, 1998), ch. 1 — used for the fiscal abstraction and the failed second rotation, and not for its link to Waldsterben, which attaches a 1981 pollution term to a 1968 quotation and describes a die-off that Otto Kandler’s survey work found did not occur.
• Enclosure, and why it is contested. John Chapman, “The Extent and Nature of Parliamentary Enclosure”, Agricultural History Review 35(1) (1987), for both the acreage and the finding that pasture and waste, not open field, were the majority target. Against a simple story: Robert Allen, Enclosure and the Yeoman (Oxford, 1992), finding yield gains of 2.5–8.4%, versus Leander Heldring, James Robinson & Sebastian Vollmer, “The Economic Effects of the English Parliamentary Enclosures”, NBER 29772 (2022), finding a 45% increase alongside a 30% rise in the parish Gini. On the one unambiguous case, Ramachandra Guha, “An early environmental debate: the making of the 1878 forest act”, Indian Economic & Social History Review 27(1) (1990), and the Indian Forest Act, 1927, still in force.
• The columns that don’t exist. On variance, Md. Raseduzzaman & Erik Steen Jensen, “Does intercropping enhance yield stability in arable crop production?”, European Journal of Agronomy 91 (2017). On soil carbon, the 2024 global meta-analysis of agroforestry soil organic carbon. On micronutrients — with iron the honest exception — “Tree cover and dietary quality”, Nature Food (2024). The erosion comparison (0.05 t/ha/yr under mixed home gardens against 40 under unconserved tea and 70 under tobacco) is compiled in Gamini Pushpakumara’s survey of Kandyan home gardens. And on why this list is so short, B. M. Kumar & P. K. R. Nair, “The enigma of tropical homegardens”, Agroforestry Systems 61 (2004), which observes that description and inventory dominated the field’s research for twenty-five years.
• It is still happening. Punjab’s crop-area collapse is tabulated in AERC Study No. 43, Punjab Agricultural University; the procurement mechanism is conceded by the government’s own summary at PRS Legislative Research. For US base acres, CRS IF12418 and 7 CFR 1412.46, which is where the destroy-before-harvest escape hatch and the “non-perennial” double-cropping definition live; on uptake of the one diversified-farm instrument, RMA’s Whole-Farm Revenue Protection. For the EU, the tree-density rule is Commission Delegated Regulation (EU) No 640/2014, Article 9(3) — not 639/2014, which concerns hemp — and the 97% shortfall is Figure 23 of European Court of Auditors Special Report 16/2021. The claim that farmers felled trees in response is conceded by the Commission’s own evaluators to have no known direct evidence, which is why this essay does not make it.
• Measurements that changed their minds. The forest-garden calorie comparison discussed in the essay is Jérémy Pasquier’s Pan Terra paper (2021) — not peer-reviewed, and candid about both its authors and its own calorie finding before it changes metric. The “75% of crop diversity lost” figure is dismantled, by the crop-conservation community itself, in Box 1 of Colin Khoury et al., “Crop genetic erosion: understanding and responding to loss of crop diversity”, New Phytologist 233(1) (2022); FAO’s own First Report on the State of the World’s Plant Genetic Resources (1996) says plainly that no one can say how much has been lost.
• Niger. The “five million hectares / two hundred million trees” figures originate in the World Resources Institute’s World Resources Report 2008: Roots of Resilience, p. 142, and are restated with their derivation in Chris Reij, Gray Tappan & Melinda Smale, “Agroenvironmental Transformation in the Sahel”, IFPRI Discussion Paper 00914 (2009) — whose own text introduces the tree count with “no study has systematically quantified the impacts.” The measured figure used here is Lee, Brandt, Oehmcke, Reij, Hiernaux et al., “Farmer Managed Natural Regeneration Promotes Expansion of Trees on Croplands in the Sahel” (2025) — a preprint, not yet peer-reviewed, and treated as such in the essay.
• Our own rules, in the open. The four tests are in the Charter §6, the reasoning and the honest gaps in ECONOMICS.md, and the working notes behind this essay — including the full list of claims that did not survive checking, and the two corrections it forced to an earlier post of ours — in exploration 0368.

Around thirty widely-circulated claims about these systems were checked and left out, because they could not be traced to a source that supported them. Several were ones this essay would have preferred to keep. Where a number here is a preprint, a stipulation or an inference, it is labelled as one in the text.