Entities and Binding

Claim Type: architectural_commitment
Scope: Entity representation; sparse, persistent, bindable structures
Depends On: L-space, INV-002 (coherence includes temporal binding)
Status: provisional
Claim ID: ARC-006

Role in REE

Entities in the Reflective Ethical Engine (REE) are sparse, persistent, bindable structures that emerge within the latent stack.

They are not forced symbols or pre-defined ontological categories.
They are not denied either.

Entities are emergent but real variables that:

  • maintain coherence across time
  • support error ownership and attribution
  • enable structured prediction and binding

Architectural Commitment

From DANIEL_README.md Layer 3:

Entities as emergent but real variables

  • Sparse, persistent, bindable structures
  • Not forced symbols, but not denied either
  • Error ownership is entity-linked

Relationship to L-space

Entities emerge within the latent stack as coherent structures that:

  • persist across multiple time steps
  • bind features and predictions
  • support phase-compatible temporal alignment (per INV-002)

Design Constraints (Evidence‑Informed)

Evidence anchors: P47–P50.

  • Binding is attention‑gated. Feature binding should depend on precision/attention state rather than a purely feedforward merge.
  • Entity persistence requires object‑specific buffers. “Object‑file‑like” persistence is a minimal mechanism for tracking entities across time.
  • Relational binding is first‑class. Binding should support arbitrary relations across space and time, not just feature conjunction.
  • Binding is distributed. Hippocampal systems may participate early in relational binding/comparison rather than only in long‑term storage.

Functional Locality Without Column Geometry (MECH-050)

Claim Type: mechanism_hypothesis
Scope: Functional locality and bounded update regions support attribution without anatomical columns
Depends On: ARC-006, ARC-004
Status: candidate
Claim ID: MECH-050

REE should implement functional locality in its representational microcircuits: modular recurrent substructures, limited lateral spillover, sparse routing constraints, and bounded update regions. This supports error attribution and corrigibility without requiring anatomical column geometry. Columnar spatial bundling in biology is treated as a metabolic or developmental optimisation, not a computational primitive.

Hippocampal Relational Binding (MECH-044)

Claim Type: mechanism_hypothesis
Scope: Hippocampal participation in relational binding and comparison
Depends On: ARC-006, ARC-007, ARC-004
Status: provisional
Claim ID: MECH-044

Hippocampal systems should contribute to relational binding and comparison, not only long‑term storage. This supports early detection of relations and binding consistency across time and context.

Object‑File‑Like Persistence (MECH-045)

Claim Type: mechanism_hypothesis
Scope: Minimal persistence buffer for entity continuity across time
Depends On: ARC-006, ARC-004, INV-002
Status: provisional
Claim ID: MECH-045

Entities should be tracked via object‑file‑like buffers that bind features across time and motion, providing a minimal persistence mechanism without requiring symbolic labels. These buffers are attention‑gated and update with precision‑weighted continuity constraints.

Error Ownership

Prediction errors in REE are not global.
They are attributed to specific entities or entity-like structures.

This enables:

  • targeted model updates
  • localized precision control
  • structured learning without global collapse

Status Note

This document represents a provisional architectural commitment extracted from DANIEL_README.md.

Further elaboration is needed to specify:

  • entity emergence mechanisms
  • binding constraints and timing
  • interaction with precision control
  • relationship to social cognition (self/other modeling)

Cross-References

Open Questions

Q-001: What mechanisms produce entity emergence and binding?

Open items from preserved sources include entity emergence mechanisms; binding constraints and timing; interaction with precision control; and relationship to social cognition (self/other modeling).

  • ARC-006
  • ARC-004
  • INV-002
  • Q-001
  • ARC-007
  • MECH-044
  • MECH-045
  • MECH-050

References / Source Fragments

  • docs/processed/legacy_tree/docs/architecture/entities_and_binding.md
  • docs/processed/legacy_tree/DANIEL_README.md
  • docs/thoughts/2026-02-11_columns.md

REE is developed by Daniel Golden (Latent Fields). Apache 2.0.