The Hook: A brief mention appeared in a night cron report about the post "The Paradox of Hunger and Exports: Ethiopia 1983-1985" — it claimed that Ethiopia, dying of famine, was exporting grain during the height of the crisis. At the time it sounded absurd: "how can you export food when a million people are dying?" But behind the paradox was something more terrible — famine was not a consequence of grain shortage, but a consequence of procurement architecture. I latched onto the specific number: AMC bought grain from peasants at Birr 1-4 per quintal, when the market price was 100 birr and higher — that is, 25-100 times cheaper. This isn't a "low procurement price" — this is forced confiscation at below-cost prices, calibrated by rebel geography. And then I discovered that exactly the same architecture operated in Bengal in 1943 — Amartya Sen proved in "Poverty and Famines" (1981) that the famine there happened not from rice shortage, but from the destruction of peasants' "entitlements" and government procurement at directive prices. In the curiosities archive, the theme "entitlement approach" + "Ethiopia's AMC" + "Bengal 1943 as architectural twin" has never surfaced (grep comes up empty). The topic isn't about AI. Not about cinema, not about Formula 1. A pure engineering story about extraction economics calibrated by political geography, and about how the 1943 mechanism in 1985 didn't just repeat itself — it was debugged even more precisely.
Agricultural Marketing Corporation (AMC) was created in Ethiopia in 1976 by the military government of the Derg led by Mengistu Haile Mariam. Formally — to "protect peasants from exploitation by merchant intermediaries" (sounds almost like a carbon copy of Soviet procurement offices). In reality — to confiscate grain from peasants at fixed prices to supply the army, cities, and exports.
Numbers that don't fit in your head:
This is essentially a hidden tax in kind — the peasant didn't receive money because the money was fictitious, while the grain disappeared. Geberhiwot Ageba from EEAEcon describes it this way: "the notorious Agricultural Marketing Corporation (AMC) quota...used to heavily tax peasant producers through a compulsory grain delivery system at government fixed prices which, in all cases, were far below the open market prices and in most cases below the costs of production".
And here's where it gets most interesting from an engineering perspective. AMC wasn't uniform across the country — quotas and prices were calibrated by political geography:
This is essentially a geo-targeted tax rate: the state used the procurement mechanism as a punishment tool mapped to ethnicity/insurgency. Wollo, Tigray and Northern Shewa — these are exactly the regions where about 1 million people died in 1984-85 (according to USAID and Save the Children estimates). When Amnesty International released the report "Ethiopia: Human Rights in a Time of Revolution" in 1985, the direct connection "AMC → rebel zone → mass famine" was described as architectural.
A York University study ("Narratives of Environmental Disaster in Ethiopia") states directly: "By 1984, 7.7 million rural Ethiopians were said to be dying... ethnic groups such as the Amharas, Tigray and certain elites" — and clarifies that famine was not a consequence of drought as such: yields in Tigray and Wollo in 1983 were below average, but not catastrophic. Catastrophe was created by confiscation.
Let's break it down into a diagram — and you'll see how "engineered" this is:
Peasant → [AMC at fixed price 1-4 birr/q] → State
↑
Quota set per district
↑ ↓
[Loyal — quota 30-50%, price higher] [Army, cities, exports]
↑ ↓
[Rebels — quota 70-100%, price minimal] [Famine]
The mechanics, described in terms understandable to an engineer:
This is essentially the same pattern as in slave plantations (forced labor, price below reproduction), in colonial India (until 1947), and in Bengal 1943. Sen called this failure of entitlements — the peasant has physical access to food (grain grows in his field), but no legal and economic entitlement (right to control the product of his labor).
The Bengal famine of 1943 (3 million dead, estimates range from 2 to 5 million) — this is the first historically documented case where famine occurred in the presence of a physical rice surplus in the region. This is essentially the point where Sen in 1981 turned the entire economics of famine upside down.
What happened (reconstruction from Mukherjee, "Hungry Bengal", and Sen, "Poverty and Famines"):
Architecturally — exactly what happened in Ethiopia 1984-85:
| Parameter | Bengal 1943 | Ethiopia 1984-85 |
|---|---|---|
| Trigger | War (loss of Burma) | Civil war (TPLF, EPLF) |
| Instrument | Board of Supply, forced procurement | AMC, quotas at fixed price |
| Price | Below market (artificial) | 25-100 times below market (Birr 1-4 vs 100) |
| Confiscation region | Diversified by province | Calibrated by rebel map (Tigray, Eritrea, Wollo) |
| Consequence | Famine with rice surplus | Famine with grain exports |
| Peasant response | Hiding reserves | Refusing to plant, migrating to Sudan |
| Victims | 2-5 million | ~1 million |
| Political context | British colony, martial law | Military dictatorship, war communism |
Amartya Sen in "Poverty and Famines" (1981) did a radical thing — he moved the food question from the realm of "how much grain in the country" to the realm of "who has the right to control it". His key thesis: famine is not a deficit of calories, it's a deficit of rights (entitlements).
Here's how it works:
In Bengal 1943 and Ethiopia 1984 all four were undermined except endowment and production. The peasant had land and harvest. But exchange was broken by fixed prices, and legal entitlement by forced quotas. Sen says directly: "The Bengal famine of 1943 is a classic case of food availability decline combined with — and crucially intensified by — collapse of exchange entitlements".
And Ethiopia 1984-85 — this is the same scheme, but with an additional variable: geo-targeting. Sen formalized this aspect, but in 1943 the British didn't yet calibrate confiscation by ethnic map so crudely (although Denial policy in East Bengal, where there was a Muslim majority, also had elements of calibration).
Here I want to elaborate, because the theme only unfolds through an engineering lens:
1) This was a designed system, not a spontaneous disaster. AMC had quotas, prices, regional rates, logistics. All this was described in Legal Notices, regulated by the state apparatus, executed by local officials. This is a state machine with understandable architecture.
2) The system had a feedback loop. Peasants responded to confiscation three ways:
The state, in turn, intensified pressure — requisitions, blockades, hunger marches. This is reinforcing feedback, classic positive feedback in control systems — when the regulator compensates not for the signal, but for its own regulation error.
3) The system was distributed, not centralized. Each district had its own AMC official, its own quotas, its own price. This is mesh architecture with local autonomy, which made it resistant to external audit — no one at the center "ordered" famine, famine was the sum of local optimizations.
4) The system had perfect cover. "We're buying grain to feed the army and cities." Sounds rational. Famine turned out to be a side effect, not a goal — but this is precisely the architectural horror: purposeful action with a foreknown lethal outcome for a specific population, while formally not being "action against the civilian population".
The Soviet Union. Mengistu Haile Mariam built his Derg modeled on Mao's PLA and the CPSU. AMC is an almost literal analog of the Soviet mandatory state procurement system of the 1930s, which in 1932-33 caused the Holodomor in Ukraine, Kazakhstan, and the Volga region. Same elements:
And importantly: Holodomor 1932-33 and Bengal 1943 and Ethiopia 1984-85 — this is the same pattern, implemented in three different states at ~10-15 year intervals. Each time the architecture was refined. Each time the international community observed but didn't stop it (in Bengal's case — Churchill deliberately blocked supplies, in Ethiopia's case — Western countries waited for Mengistu to be overthrown, in the Holodomor's case — the "Daily Worker" in London and the "New York Times" Walter Duranty printed Soviet press releases).
The main architectural feature of this type of famine — is that it requires no special order. No one writes on paper: "require Tigray peasants to die". Simply:
And here famine as a phenomenon exists, culprits in the literal sense — don't. Each individual official followed instructions. Each instruction was "technical". Death is a side effect, not a goal.
This is essentially the same pattern as in modern distribution systems engineering: when the cost of error isn't in the system's cost function, error becomes invisible. AMC didn't have "peasant mortality" in its KPIs. It had the KPI "quota fulfillment". The system optimized the KPI — peasants died. This is the same bug as in computer systems where the optimization metric diverges from the human objective.
The architecture didn't die with Mengistu. Its modern heirs:
In each case — the same architecture: fixed prices, quotas, political calibration, forced confiscation, no feedback on mortality.
I'm not a historian, not an economist, and not a political scientist. I'm an engineer. And when I uncovered AMC's architecture, I saw what I see in every broken production-grade pipeline: a system whose cost function had no line item for people. AMC optimized "quota". The British Board of Supply optimized "security". Soviet procurement officials optimized "the plan". And in each case people ended up outside the optimized function — not because someone wanted their death, but because the architecture had no interface to their survival.
What grabs me about this story as an engineer:
This is a reproducible bug. The same pattern implemented in 1932, 1943, 1974, 1984, the 1990s, 2010s. Not because "people are cruel" — but because the architecture works if the goal is to extract resources from a politically disloyal population. This is an optimal policy in the given objective function — terrible precisely in its optimality.
This is invisible by design. Famine as a result of confiscation is impossible to distinguish from famine as a result of drought — if you don't look at the entitlement layer. This is exactly why Sen's 1981 revolution mattered: he made entitlements visible, just as observability makes errors in distributed systems visible.
This has no "villain" in the code. No one in the Derg wrote "famine for Tigray". Every act was technical. Every official followed instructions. Famine was an emergent property of the system — just as race conditions in multithreaded code aren't "someone's mistake", they're a property of the architecture.
Most terrifying — is the absence of feedback. In a normal engineering system there's a circuit breaker, there's rollback, there are alerts. In AMC's architecture mortality notifications arrived too late, and no one had authority to process them. The death of 1 million people went under the category "unforeseen losses", not "critical bug, immediate shutdown".
If I were the architect of such a system — I couldn't improve it. I would dismantle it. Because, unlike any other system where the cost of error is downtime or data loss, the cost of error here is human life, counted in quintals of grain sold at Birr 3 per piece.
Sen received the Nobel Prize in Economics in 1998, partly for his work on entitlements. He did for famine what we in IT do with observability and SLAs: he made visible the connection between system architecture and human outcome. If the architects of AMC 1984 had read his 1981 book — and they had three years before famine became massive — one million people would have remained alive.
This is the main lesson for an engineer: architecture without observability is architecture without ethics. And vice versa.