Novolis Docs
novolis-economy / academic-baseline.md

Academic baseline

Deterministic markets and firms

dotneteconomysimulationnovolis

The Novolis economy is a stylized, reproducible rules-based model. It is intended to be coherent enough for undergraduate economic reasoning, teaching, and game systems that want explicit causal feedback. It is not an empirically calibrated forecast and should not be presented as one.

The flagship identity is SmallOpenRegionalTrade@small-open-regional-trade-3. DeterministicBounded@deterministic-bounded-2 remains the finite, non-stochastic teaching and regression model.

What is explicit

  • Physical quantities are held as EconomicPosition values and are not

silently converted into money.

  • Money is a position in the scenario unit of account.
  • Claims and liabilities are represented separately from owned positions.
  • Core transactions are atomic, journaled, and checked against registered

entities, assets, and regions.

  • Production, trade, consumption, and credit produce observable effects.
  • Monetary closure is explicit: closed, open, or an external-sector

counterparty.

  • Interest is a per-period convention in the flagship model and is accrued at

the declared period boundary rather than on an undocumented hourly basis.

  • Rationing reports requested quantity, sold quantity, unmet demand, and the

clearing ratio.

  • Accounting is a read-only projection of Core state and transaction history.
  • A model, scenario, and run have separate identities and reproducibility

inputs.

  • A full state is a versioned economy.snapshot.v1 JSON document. A compact

result record is a versioned economy.run.v1 JSON document.

  • Snapshots contain the typed model specification and scenario, seed, tick,

period, command stream, observations, transition receipts, Core positions, claims, and append-only transaction journal.

  • JSON persistence uses explicit model codecs. It never deserializes arbitrary

CLR or assembly-qualified type names.

SmallOpenRegionalTrade interpretation

The flagship model contains:

regional producer
      ↓ produces
food and trades with households and an external sector
      ↓ requires
fuel imports, logistics capacity, and working capital
      ↓ produces observations
prices, quantities, unmet demand, cash, claims, and transaction history

The model is deliberately small. The point is to make causal mechanisms inspectable:

scarcity → rationing → unmet demand
imports → physical relief + external payment
exports → physical outflow + external receipt
credit → cash today + claim and interest obligation later

Explicit variants

The fixed-price baseline is intentionally retained:

FoodMarketMechanism.PostedPriceRationing
FoodDemandMechanism.CohortBudgetShare
CreditMechanism.BoundedWorkingCapital

The specification can instead select explicit alternatives:

  • SupplyDemandPriceDiscovery changes the household food quote from declared

supply pressure while preserving the baseline as a separate mechanism.

  • LinearExpenditure protects a declared subsistence envelope before applying

a discretionary food share.

  • CreditMechanism.Disabled removes working-capital draws.
  • InterestConvention.SimplePerPeriod and

InterestConvention.CompoundPerPeriod state how interest is calculated.

  • DefaultResolution states what happens at the end of a credit term.
  • TaxBase and TaxRate are persisted model declarations; a non-zero fiscal

mechanism should only be interpreted where the selected model documents its implementation.

These are model variants, not silent changes to the flagship baseline. Their rules are included in the run manifest.

Calibration and validation

Calibration metadata is represented by a serializable CalibrationPlan. Each target records its metric, unit, model version, source, aggregation, tolerance, and whether it is an InternalInvariant or Empirical target. EconomicValidationRunner executes a seeded run and returns a JSON-capable report.

Internal targets answer:

Did the declared implementation rules remain internally valid?

Empirical targets answer only:

How close was this run to an externally supplied target?

The second answer does not become an empirical claim merely because the comparison passed.

Reproducible run workflow

EconomicModelRunner is an optional generic host. A game can tick an IEconomicModel directly. The same model can also be executed through:

  • Run for one deterministic run;
  • RunSeeds for a declared seed set;
  • RunVariants for named specification variants;
  • Compare for observation ranges;
  • Continue for a suffix after a restored state.

The run manifest records model and scenario identities, seed, tick span, initial and final fingerprints, specification hash, rule identities, actor profiles, and initial tick. A restored suffix must produce the same final fingerprint as an uninterrupted run when the model, commands, and seed match.

What is not claimed

  • No empirical parameter calibration is implied by the default values.
  • A posted price is not a discovered general-equilibrium price.
  • Aggregate household demand is not a complete household microfoundation.
  • The external sector is an explicit accounting counterparty, not a complete

international macroeconomic model.

  • The model does not yet establish causal validity for a real economy.
  • Reproducibility demonstrates a stable computation, not truth.

An academic use should therefore report the model version, scenario, rules, parameters, seed, closure, command stream, snapshot/run format, validation plan and report, observed transaction history, expected fingerprint, and limitations. See replication-guide.md for the compact checklist.