◄ WORLD IV · SONIAROOM · the Russian world

THE OGAS — THE ALL-STATE AUTOMATED SYSTEM

Viktor Glushkov’s ОГАС (1962–70s): a nationwide computer network to run the Soviet planned economy in real time — approved in principle, never built. The centre is genuinely runnable: a proportional control loop that settles a plan target, plus the hierarchical network that would have carried it. It shows exactly when the plan converges and, pushed too far, when it flies apart.

source V. M. Glushkov, OGAS proposal (1962–1970s); B. Peters, How Not to Network a Nation (2016); cybernetics rehabilitated in the USSR after Stalin. Room THE STATE. Figures cited inline; rendered, not quoted.

◧ roots · lineage · witness
3

THE ROOTS

Under Stalin, cybernetics was denounced as a “bourgeois pseudoscience.” After his death it was rehabilitated (mid-1950s), and control theory became respectable. In 1962, mathematician Viktor Glushkov proposed OGAS — a single automated system to manage the whole economy in real time, closing the loop between what was produced and what the plan asked for.

The dream was cybernetic: the economy as a controllable system with feedback, not a stack of paper reports arriving a year late. AMBER historical framing.

5

THE NETWORK ~approx

Glushkov’s architecture was a hierarchy: a top tier of about ~200 main computer centres feeding down to roughly ~20,000 local centres in factories and regions — the plan as one live system rather than a filing cabinet.

Those counts are proposal-era estimates AMBER. The shape is exact and testable: a branching tree of depth D and fan-out b holds (bD+1−1)/(b−1) nodes — computed live in the centre.

7

THE LINEAGE AMBER

OGAS has a twin in Chile: Cybersyn (Stafford Beer, 1971–73), a smaller real-time control room for a socialist economy — also killed by politics. And a ghost: the internet the USSR did not build. OGAS predates a civilian internet, yet died while ARPANET (1969) grew.

It ties directly to GOBERNET and the socialist calculation debate: a loop can be made to settle a quantity, but settling is not the same as knowing the right target.

▼ the loop · the plan, controlled ▼
4

DATA IN — the plan & the network in ↓

A controller nudges the current output x toward a plan target by a fraction of the gap. The gain k sets how hard it pulls. The network is a tree of fan-out b, depth D.

200
40
0.50
10
2
▼   xₙ₊₁ = xₙ + k·(target − xₙ)   ▼
0

▣ THE PANEL — the control loop LIT

Each step the error obeys errorₙ₊₁ = (1−k)·errorₙ. It dies iff |1−k| < 1, i.e. 0 < k < 2; at k=1 the error is gone in one step; at k ≥ 2 it never settles.

▼   does the error die?   ▼
8

DATA OUT — settling & scale out ↓

Stability band: the plan settles for 0 < k < 2 only — the one invariant OGAS’s controllers would have had to respect. LIT

the audit · the wall ◨
1

WHY IT DIED

WALL The machine could close the loop; the ministries would not. A live, transparent system threatened every official whose power lived in the gap between the real numbers and the reported ones. Add the enormous cost, inter-agency turf, and no single patron willing to own it — and OGAS was approved in principle and never built.

The barrier was not mathematics. The control law works (centre, LIT). The wall was political: transparency is a cost to whoever benefits from opacity.

2

THE GRAVEYARD

“The USSR never imagined networked computing.” Cut. OGAS (1962) out-ambitioned the early civilian internet — a nationwide real-time network years before ARPANET (1969) carried its first packets.

“A big enough computer would have made the plan work.” Examined. The loop settles a chosen target; it never tells you the target is right. See the calculation debate.

“It failed because the math was wrong.” Cut. The feedback law is correct and stable in-band. It died in committee, not in the equations.

6

THE TAMPER — break stability

Push the gain to k = 2.4 and claim the plan still settles. The overshooting controller now amplifies its own error each step — the numbers grow without bound and the stability witness turns red.

Stability needs 0 < k < 2. Claiming a settle at k≥2 does not make it so — the honest witness disagrees, and the error runs away.