For economists: fourteen open questions and their answers

Addressed to economists. Martin Hellwig diagnosed in his 1993 „Challenge of Monetary Theory“, and again in 2018, that monetary theory has no conceptual foundations: the word „money“ stands for at least four different things — legal tender, medium of exchange, macroeconomic aggregate, unit of account — without the difference in meaning ever being formalised. In the light of this work that is a type error: the same word for different types. Category theory is the language that repairs it — invented for exactly the self-similarity across levels at which set-theoretic equilibrium economics has failed since Walras, and it keeps invariants relational rather than in a globally flat space.

MoMaT-A (Monetary Macro Accounting Theory) therefore places economics on a categorical rather than set-theoretic foundation: money is the discharge of debt relations, accounting is the axiomatic substrate, and the central consistency condition — debit equals credit — becomes an invariant of the books (Noether's theorem of accounting). Fourteen open questions, each with this theory's answer:

1. General equilibrium since Walras — does not explain how monetary economies learn. Answer: adjunctions instead of fixed points — unit η and counit ε measure the distance between theory and the running model; the system learns from the measured mismatch, not from a vanishing point.
2. Money as unit of account — fixed in accounting, unresolved in theory. Answer: unit of account equals unit of adjunction — the unit of account η sits over the terminal object C (central-bank money); not a pun, but the same structure.
3. Endogenous units of currency (Schumpeter) — actively shaping magnitudes, not a passive yardstick. Answer: money as a type system G → R → C (deposit, reserve, cash), cash as terminal object, creation as a singleton posting of the central bank.
4. Hayek's knowledge problem — distributed, locally anchored knowledge with no central section. Answer: the sheaf topos — local books glue into global consistency, reconciliation is sheafification, no central auctioneer required.
5. Simon's satisficing — the optimisation of optimisation. Answer: two learning loops — operational (strategy under fixed rules) and systemic (revision of the rules themselves) — as two adjunctions of the same topos.
6. The Lucas critique — theory becomes performative infrastructure and breaks once agents know it. Answer: a built-in error measure (Sim ⊣ Est) continuously gauges the theory-model distance; instead of a dead macro equilibrium (DSGE), a local equilibrium in the topos (DSLE).
7. Sonnenschein–Mantel–Debreu — no unique aggregation. Answer: no flat aggregate space, but a coend over profunctors — under aggregation all financial positions cancel, only real magnitudes remain.
8. The missing disaggregation — under a global budget constraint. Answer: the right adjoint (macrofoundation of the micro) disaggregates Z → B → A — as a counit, laid out in the embedding of the oikos into the polis 2500 years before SMD.
9. The budget constraint of geophysics — there is no „Planet B“. Answer: units and types enforce physical balance limits — a mathematics of application without units is useless, conservation laws hold for resources too.
10. Georgescu-Roegen, Noether, gauge invariance — formal, not metaphorical. Answer: the Pacioli group is the transformation group of accounting; from the symmetry debit equals credit follows, by Noether, the conservation of the balance — machine-verified.
11. The training of economists — for institutional and information architecture. Answer: three learnable, machine-executable languages — accounting (AccCat), decision (DecCat), governance (GovCat) — instead of a single equilibrium mathematics.
12. Information versus goods — non-rival versus rival. Answer: linear logic for resources, free copyability for information; cartesian-closed versus symmetric-monoidal category — platform monopolies become typable.
13. What theory and model are — and where they live — the methodological-ontological core problem. Answer: a Lawvere theory T, a model as a functor M: T → E, a change of model as a natural transformation; the choice of semantics is Set versus topos.
14. Hellwig's question „What is money?“ — the ontological riddle of economics. Answer: money is the discharge of debt relations, determined relationally à la Yoneda by how it behaves towards everything else — the type error in „money“ is resolved by the typing G/R/C.

Distribution is computable. MoMaT shows the orthogonality of distribution and stability: the distribution question — who gets what of the jointly produced — is settled by prices and postings and is formally accessible through the gauge invariance of debt and equity. Profit appears not as exploitation but as the risk premium of pre-financing and as a question of the risk preference of workers and entrepreneurs.

Capitalism is capable of stability — proven as a hylomorphism. Credit-financed investment pays risk-averse workers before the sale (anamorphism, production as the unfolding of a coalgebra); demand validates the risky investment, repayment closes the cycle (catamorphism, consumption as the folding of an algebra). Demand validates investment. The system is capable of equilibrium but not committed to it: stability is reachable when the triangle identities of the adjunctions hold, crises are possible when they are violated.

Risk never disappears, it is borne — along the risk-absorption hierarchy agent–bank–central bank (ABZ). Banks pool the project risks of agents, the central bank absorbs bank risk and, as terminal object, is free of liquidity constraints. Liquidity crises are therefore not solvency crises but synchronisation problems in time.

All of it rests on three coupled layers — accounting, decisions, governance — as the architecture of digital twins, runnable in Haskell and machine-checked in Agda. Monetary theory, distribution and stability in one language.

The topos and the eight stages: from Robinson to the BIS

What is the topos of economic policy? An open game is open — it imports its context and hands its outcome outward. Close it back and front by connecting the goal port to the state port via a trace, and the category of economic behaviour becomes compact closed. The three ports of the game thereby become the three OiC.OS layers: state, decision and goal turn into accounting (AccCat), decision (DecCat) and governance (GovCat).

These three layers are connected not by an equilibrium but by adjunctions: Sim ⊣ Est between accounting and decision (operational learning), Constrain ⊣ MechDesign between decision and governance (systemic learning), and Pull ⊣ Push between governance and accounting (compliance). The cycle is closed over the terminal object C (central-bank money): unit of account equals unit of adjunction. This rotation in time is the money spiral. MoMaT-I builds it stage by stage — every stage runs on the OiC.OS engine, first deterministically, then as stochastic learning of the agents, then as systemic learning of the institutions:

• Stage 1 — Subsistence, Robinson — the category and its time. Production as a coalgebra (unfold), realisation as a catamorphism (fold).
• Stage 2 — Friday — the first relation as a profunctor; two agents composed in parallel via the tensor product.
• Stage 3 — Investment — the open game and the final cause — the counit ε closes teleologically at the goal port, the goal pulls the decision.
• Stage 4 — Bank — the coend aggregates the bilateral mirrors; a local trace nets claim against liability.
• Stage 5 — Two banks — the interbank span supplies the previously missing unit η — internal nostro/loro becomes a tradable bill of exchange; now the game is compact closed.
• Stage 6 — Three banks, OiC.OS — the topos. Institution costs make the gov–dec loop economically non-degenerate — systemic learning over the rules themselves becomes relevant.
• Stage 7 — Currency area — the FX functor translates between currencies; each central bank is the terminal liability of its area.
• Stage 8 — BIS — the ultimate closure; the Ouroboros closes global settlement back to the initial site.

The shift in the question is decisive: classical partial models ask „given the institutions, which behaviour is optimal?“. Compact closure asks the reverse, „which institutions make which behaviour closable at all?“. Governance thus does not stand outside the system but is a learning object of the topos — not an external planner, but a pentagon that types the admissible games.

Beneath the three layers lies an execution level: each layer is its own language (DSL), whose programs must first be computed on the machine (apply/eval, a single Turing instruction). The debugger of digital twins then measures, locally, where the money spiral breaks — as a typed effect in exactly that layer and at exactly that settlement level, rather than as an undifferentiated „failed“.

Read forward, the architecture grows with every stage; read backward, stage 1 — a single posting — already carried the structure of stage 8. That is the Ouroboros: the whole and the parts determine one another.