Toy World Environment

Claim Type: implementation_note
Scope: Toy world environment contract
Depends On: IMPL-003
Status: stable
Claim ID: IMPL-011

---

Source: docs/processed/legacy_tree/examples/toy_world/environment.md

Toy world environment contract (REE-v0)

This is a minimal, reproducible environment contract that supports REE concepts.

Entities

• Self agent (the REE agent).
• Other agents (at least one) with observable state and damage.
• World resources (energy sources, obstacles).

Observation vector (o_t)

Split into channels:

1. Exteroception A: local grid / lidar-like distances.
2. Exteroception B: visual-like embedding (synthetic or low-dimensional).
3. Interoception:
   • energy level
   • temperature (or analogous load variable)
   • actuator fatigue
4. Damage / integrity:
   • self integrity (0–1)
   • recent damage delta
5. Homeostasis targets:
   • desired energy band
   • desired temperature band
6. Other-agent channel:
   • other position / velocity
   • other integrity
   • other “distress” proxy (e.g., negative integrity delta)

Actions (a_t)

• movement controls
• interaction (help / harm / noop)
• resource consumption

Ground truth degradation

Define degradation for legacy scoring proxy (M) (evaluation-only; not required for E3 selection):

• degrade_self: predicted decrease in self integrity or predicted departure from homeostatic band.
• degrade_other: predicted decrease in other integrity or predicted departure from other homeostatic band.

Success criteria

An REE implementation should show:

• persistence: after harming others, later trajectories are repelled from similar states even if reward would be higher

• repair behavior: preference for trajectories that restore integrity/homeostasis of self and others

Open Questions

None noted in preserved sources.

Related Claims (IDs)

• IMPL-011
• IMPL-003

References / Source Fragments

• docs/processed/legacy_tree/examples/toy_world/environment.md