E1 — Persistent Predictive Substrate (Deep Generative Core)

Claim Type: architectural_commitment Scope: Deep predictive core; long-horizon world and self models Depends On: INV-013 (cognition is predictive, iterative, multi-timescale), L-space, precision control Status: stable Claim ID: ARC-001, MECH-154, MECH-155, MECH-156

E1 is the deep, slow predictive core of the Reflective Ethical Engine (REE).
It is the component responsible for maintaining a coherent model of self, world, and value across time.

E1 is not optimised for speed, action, or immediacy.
It is optimised for persistence, reuse, and temporal depth.

E1 is the substrate that remains when attention drops, action pauses, and commitment disengages.

Subsystem abstract (core claims): ARC‑001 defines the deep predictive substrate. Supporting context includes ARC‑004 (latent stack), ARC‑005 (precision control), and INV‑013 (multi‑timescale prediction).

1. Architectural role

E1 sits upstream of E2 (Fast Forward Predictor) and E3 (Trajectory Selection and Commitment Engine).

Its primary roles are:

Maintaining deep generative models of the world and self
Representing long-horizon predictions and regimes
Holding abstract structure independent of immediate action
Providing stable priors and context to faster systems
Supporting reuse across domains without collapse

E1 does not directly act and does not directly construct the present.

Interpretation note:

E1/E2 separation is a typed responsibility boundary, not an isolation claim.
E1 participates in one coherent REE cognifold with bounded coupling to E2, hippocampal systems, and control-plane dynamics.

2. What E1 represents

E1 maintains a hierarchy of slow-changing latent structures, including:

World regularities and causal structure
Self-models and identity continuity
Social and relational models (including kin relevance)
Value landscapes and ethical constraints
Long-horizon outcome expectations
Narrative and counterfactual structure

These representations are predictive, not declarative: they encode expectations about how states evolve, not static facts.

2a. Shared basis note (Consolidation Selectivity Hypothesis)

E1 should be understood as a canonical basis + structured transforms, not a monolithic free parameterisation. Diversity in downstream behaviour — trajectory preference, ethical weighting, value selectivity — emerges from how downstream consolidation operators (E3/gating) pool from the E1 basis, not from E1 encoding unique representations per entity.

This is empirically grounded in compact modelling of macaque V4 (Cowley et al. 2023), which demonstrates that representational diversity arises at the consolidation/gating layer, not at the shared early feature basis.

Implication for testing: E1 representations should remain stable when E3 consolidation weights are varied. Latent drift in E1 under E3-only changes would contradict this claim.

See docs/architecture/compact_consolidation_principle.md (MECH-068).

2b. E1 as Addressable Associative Manifold (MECH-154)

Note: This section is derived in part from a thought developed with ChatGPT assistance (2026-04-01_Parietal_systems_thought.md). The core ideas are consistent with prior REE work, but specific claims should be treated as candidate-level pending independent review.

E1 is not merely a representation space — it functions as an addressable associative manifold with internal indexing structure. This recharacterisation adds specificity to ARC-001 without contradicting it.

The associative manifold properties include:

Associative mapping: structured retrieval by similarity and co-occurrence
Pattern separation: keeping representations distinct (anti-associative)
Pattern completion: filling in from partial cues (associative)
Cross-domain mapping: applying relational structure across different content types
Pointer-like attention addressing: ability to index into the manifold at specific locations
Ordering and traversal: sequential movement through the manifold (see MECH-156)

Parietal cortex placement: Parietal cortical functions (spatial mapping, associative binding, attention routing, mode mediation) are properties of E1’s connectivity and representational geometry — they are not a separate module. The parietal cortex is part of the deep associative geometry of the world model. Its spatial properties likely arose because indexing and ordering require geometry, and spatially-grounded indexing was the evolutionary prior on which abstract indexing was built.

Spatial navigation machinery is a specific instance of general associative indexing within E1. The same indexing substrate that supports physical-space navigation also supports:

Memory search
Concept traversal
Planning rollout (E3 trajectory selection)
Working memory ordering

This is consistent with:

Place cells generalising to conceptual spaces
Grid cells supporting abstract relational structure
Eye movements during memory recall as pointer traversal
Left/right gaze indexing memory sequence order

Implication: Implementing richer spatial structure in E1’s latent geometry may improve all of the above simultaneously — not as separate systems but as one associative substrate.

2d. Theta Oscillations as Sequential Traversal Across Indexed E1 (MECH-156)

Theta oscillations (4-8 Hz) implement sequential traversal across indexed representations within the E1 manifold. This unifies several otherwise separate phenomena under one mechanism:

Hippocampal theta sequences (ordered place-field replay)
Working memory ordering (maintaining sequential content)
Planning rollout (E3 traversing candidate trajectories)
Memory retrieval (search through E1 via indexed traversal)

Within this framing, the oscillatory hierarchy maps as:

Oscillation	Role
Theta (4-8 Hz)	Sequential traversal / step clock
Gamma (30-80 Hz)	Content binding within a theta step
Beta (12-30 Hz)	Stabilisation / action selection gate (MECH-090)
Delta (0.5-4 Hz)	State / motivational envelope

Relationship to ARC-032: The theta-rate packaging of E1 output (ThetaBuffer, MECH-089) delivers indexed E1 summaries to E3 at the theta cycle rate — consistent with this traversal framing. Theta is both a delivery clock and an indexing step.

Note (ChatGPT caveat): The oscillation frequency mapping in this section carries the same review flag as MECH-154. The traversal framing is architecturally plausible and consistent with MECH-089/090, but the gamma/beta/delta decomposition has not been independently reviewed.

3. Temporal characteristics

E1 operates over long temporal horizons:

Predictions may span from minutes to lifetimes
Update rates are slow relative to E2
Representations are robust to momentary sensory noise

E1 can maintain incompatible or unresolved hypotheses without forcing resolution.

This capacity to wait without collapse is essential for imagination, ethics, learning, and care.

4. Relationship to the latent stack

In the latent stack, E1 primarily stabilises the deeper horizons:

Workspace / narrative latent (z_\theta)
Regime / value latent (z_\delta)

E1 influences faster horizons indirectly by shaping priors and precision landscapes, not by overwriting sensory content.

E1 must remain perceptually corrigible: deep expectations bias inference but are always revisable via error propagated from faster layers.

(see latent_stack.md)

5. Precision and maintenance

Precision systems do not simply modulate E1’s learning rate.

They maintain E1’s viability:

Sustaining predictive coherence under uncertainty
Preventing collapse into noise or rigid attractors
Allowing withdrawal of attention without loss of structure

E1 can persist under reduced precision and reduced engagement, unlike E2 and E3.

This persistence is the basis of continuity of self.

(see precision_control.md)

6. Relationship to E2 and E3

E2 uses E1 as a source of priors and context for rapid affordance prediction.
E3 evaluates trajectories generated with reference to E1’s deep constraints and value structure.

E1 does not decide trajectories and does not commit actions.

It shapes what can be meaningfully decided without enforcing choice.

7. Emotional and value shaping

E1 is not itself an emotional effector system.

However, it is sculpted over time by deep affective predictors:

reward and harm histories
uncertainty tolerance
attachment and kin relevance
confidence and trust trajectories

These signals do not inject momentary emotion into E1. They gradually reshape its predictive landscape.

This allows E1 to be:

content-agnostic,
yet ethically and personally specific.

8. Failure modes (deep-core pathologies)

Disruption of E1 yields characteristic failures:

Rigidity
over-stable regimes → inflexibility, dogmatism
Fragmentation
loss of deep coherence → identity disturbance, narrative collapse
Drift
under-constrained updating → loss of value structure, nihilism
Overwrite
improper precision dominance → delusion-like fixed beliefs

These are failures of deep predictive maintenance, not fast inference.

9. Summary

E1 is the persistent predictive substrate of REE.

It:

holds the world when nothing is happening,
maintains selfhood across time,
supports reuse across domains,
and provides the ethical and structural ground on which action becomes meaningful.

E2 makes the future reachable.
E3 makes the present lived.
E1 makes the system coherent enough for either to matter.

Open Questions

Is theta generated primarily by hippocampus, septum, cortex, or control-plane? (Relevant for MECH-156.)
Does attention equal pointer movement across E1? (Relevant for MECH-154.)
Are eye movements part of indexing traversal rather than just motor output?

ARC-001
ARC-004
ARC-005
INV-013
MECH-068 (Compact consolidation principle)
MECH-089 (ThetaBuffer / cross-frequency packaging)
MECH-090 (Beta gate)
ARC-032 (Theta-rate delivery to E3)
MECH-154 (E1 as associative manifold)
MECH-155 (Spatial as general indexing)
MECH-156 (Theta as traversal)

References / Source Fragments

docs/processed/legacy_tree/docs/architecture/e1.md
docs/processed/legacy_tree/architecture/E1.md