Projection, solution and diagnostic settings

These optional sections control model-wide projection inputs, numerical solution settings and diagnostic detail.

Choose terminal conditions

The original terminal calculation remains the default. To explicitly enable the new equilibrium continuation after years.last_projection, add this to the build’s model_configuration.yaml:

projections:
  terminal_closure:
    method: equilibrium
    exogenous_tail: constant

To explicitly restore the original calculation, replace that mapping with the following, removing any existing annual-change overrides:

projections:
  terminal_closure:
    method: legacy
    exogenous_tail: constant

Merge these examples into an existing projections section; do not duplicate the key. The terminal_closure mapping and all its settings are optional. method defaults to legacy, exogenous_tail to constant, and annual_changes to {} (no overrides). Omit annual_changes when there are no overrides, as in these examples. Only equilibrium permits a linear tail or nonempty annual changes. The existing sibling setting projections.constants_decay_rate is also optional, defaults to 0.0 and accepts a decimal rate at least zero and less than one; the equilibrium tail continues that decay beyond the endpoint.

See Terminal conditions for the economic meaning of both methods, all settings, trend and override examples, adjustment units, and instructions for rebuilding projections after changing configuration. A constant growth-rate input still allows economic growth; the new closure does not force terminal capital or asset positions to zero.

Preserve maintained projection inputs

Generated baseline inputs are normally replaced when model data is updated. Set a preservation flag only when the named file is maintained deliberately:

projections:
  preserve_labor_augmenting_technical_change_file: true
  preserve_baseline_exogenous_projections_file: true

Preservation does not bypass the advancement input contract. G-Cubed still validates the source deviations, including exact zeros in the final 50 projection years, and checks preserved LATC and baseline growth rows against the configured assumptions. If a file is stale or inconsistent, correct the source and regenerate it with the corresponding preservation flag disabled; then review the result before restoring preservation. Preserve unrelated custom baseline adjustments while reconciling SHL, ROGY and ROGYEUR.

Saved solved models and baseline results can also become stale after a change to common growth or its source inputs. When the experiment runtime loads a solution, it also checks the configuration and inputs selected by the current run. Copying a saved solution into another build does not bypass that check.

A stale-input error requires fresh model preparation, calibration and solution, followed by baseline and dependent experiment regeneration. Use the maintained run script’s force-resolve option; do not rely on rerunning a script that merely reloads its cached solution. See the growth regeneration sequence.

Adjust stable-manifold settings

solution:
  stable_manifold:
    tolerance: 1.0e-7
    maximum_iterations: 1500
  unit_root_tolerance: 1.001
  enforce_aggregate_consistency: true

Use a decimal point in scientific notation. 1e-7 may be read as text; 1.0e-7 is a number.

Request detailed diagnostics

diagnostics:
  detailed: true

Omitted booleans use their documented defaults. Do not quote true or false.

Baseline tuning