# The Steward — Simulation Definitions

## Purpose

Stable IDs and immutable definitions now form the vocabulary shared by
simulation state, systems, world adapters, scenes, tests, and future save
migrations.

`SimulationIds` is the canonical source for code-facing `StringName` IDs.
`ActionDefinition` and `ProfessionDefinition` are editor-readable custom
resources. `SimulationDefinitions` loads, resolves, and validates the complete
prototype set.

## Current action contract

Each executable action definition contains:

- stable `action_id`;
- display name;
- default duration;
- optional preferred profession ID;
- target type: resource, activity, or free movement;
- resource action ID when the action targets a ResourceNode.

The current actions are gather food, gather wood, deposit food, withdraw food,
patrol, study, eat, rest, and wander. Idle and dead are stable state sentinels,
not executable action definitions.

SimNPC reads default duration and preferred-profession metadata from these
definitions. WorldViewManager reads target type and resource-action metadata
instead of maintaining a separate gather-action map.

## Current profession contract

Each profession definition contains a stable `profession_id` and display name.
The current professions are farmer, woodcutter, guard, scholar, and wanderer.

NPC generation chooses from the registry's stable IDs. NPC state stores and
restores those IDs as `StringName`, and rejects records that reference an
unknown profession or executable action.

## Validation

The registry rejects:

- missing or duplicate IDs;
- empty display names;
- non-positive action durations;
- invalid target types;
- resource actions without a resource-action ID;
- references to unknown professions or resource actions;
- definition resources that fail to load.

`tests/simulation_definitions_test.gd` verifies registry integrity,
definition-backed behavior, unknown-ID rejection, and serialization
round-tripping.

## Deliberate boundary

Definitions currently own identity and the static metadata already proven by
the prototype. They do not yet own:

- utility thresholds and consideration curves;
- preconditions or costs;
- completion effects;
- interruption and failure policy;
- reservation strategy;
- detailed target-selection policy beyond current resource/activity/free
  metadata;
- presentation hints beyond target type.

Selection, execution, target resolution, and visual travel are now separate
responsibilities. The remaining concerns should move into definitions only
when the action-system implementations prove a stable, reusable contract.