Default Mode (Internal Generative Mode)

Claim Type: architectural_commitment
Scope: Internal operating mode; imagination and counterfactual exploration without commitment
Depends On: INV-011 (imagination without belief update), E3, control plane, hippocampal systems, residue geometry
Status: stable
Claim ID: ARC-014

Role in REE

The Default Mode in the Reflective Ethical Engine (REE) is an internal generative operating mode characterised by reduced sensory dominance, suppressed action commitment, and elevated internal simulation.

It enables: • imagination and counterfactual exploration • autobiographical reflection • planning without commitment • continuity of self across time

The Default Mode is not idle and not background noise. It is a necessary mode for any system that: • stores path-based identity • supports replay and recombination • distinguishes imagination from action

Subsystem abstract (core claims): ARC‑014 defines Default Mode as a non‑commitment operating regime, and MECH‑029 specifies reflective replay and ethical simulation within that regime. Supporting context includes ARC‑005 (control plane), ARC‑003 (commitment gating), ARC‑007 (hippocampal replay), ARC‑013 (residue geometry), ARC‑004 (L‑space), and INV‑011 (imagination without belief update).

⸻

Architectural necessity

Given the REE constraints: • perception completes at the shared sensory latent z_S • trajectory commitment occurs only via E3 • ethical residue \phi(z) is path dependent • hippocampal braid stores sparse, segmented trajectories • pattern completion produces hypotheses, not beliefs

there must exist a system mode in which: 1. Sensory prediction error is down-weighted 2. Hippocampal replay can drive internal dynamics 3. Fast forward prediction can explore trajectories 4. Trajectory selection (E3) is suppressed or gated 5. Generated trajectories are treated as non-committal

The Default Mode satisfies these requirements.

⸻

Information flow characteristics

⸻

Temporal displacement and replayed smear

The Default Mode operates over a temporally displaced latent manifold rather than a single-time-slice state.

Hippocampal replay and E1 internal generation sample smeared bundles of future- and past-oriented predictions across multiple horizons (sensory, affordance, narrative, affective). These bundles represent candidate trajectories through time, not present-moment commitments.

Crucially: • The temporal smear is explored without phase-locking to action • Predictions remain unaligned to motor execution timing • The illusion of a unitary “now” is explicitly not constructed in this mode

This preserves a strict separation between imagination and lived experience.

In Default Mode, effective information flow differs from task-engaged mode in specific, constrained ways.

Sensory precision modulation • Precision (gain) on exteroceptive prediction error is reduced • Sensory input is not ignored, but is weakly coupled • Perceptual state z_S remains corrigible and intact

This prevents imagination from becoming hallucination while allowing internal generation to dominate.

⸻

Hippocampal → cortical influence • The hippocampal braid emits replayed or recombined trajectories • These trajectories seed internal dynamics in E1 • Replay may span past, future, and counterfactual paths

The hippocampus drives simulation, not action.

⸻

Elevated internal prediction (E1) • E1 runs with loosened external constraints • Latent trajectories propagate forward and backward in time • Pattern completion fills in unobserved segments

This supports daydreaming, narrative construction, and future imagining.

⸻

Suppression of trajectory commitment (E3) • E3 commitment thresholds are raised • Candidate trajectories remain hypothetical • No policy collapse or residue update occurs

This is the critical guardrail distinguishing imagination from action.

⸻

Relationship to latent spaces • z_S: remains stable; precision is modulated but content is not overwritten • z_A: explored freely via fast forward prediction • \phi(z): shapes traversal cost but is not modified during replay

Default Mode explores the geometry of thought without deforming it.

⸻

Replay semantics

Replay in Default Mode: • samples alternative paths over a fixed residue field • recombines episodic segments via pattern completion • supports regret, planning, and identity maintenance

Replay outputs are hypotheses.

They only become commitments if: • E3 later selects a trajectory during task-engaged mode

Micro-quiescence replay (MECH-092): The same replay mechanism operates at shorter timescales during quiescent E3 heartbeat cycles — brief inter-action gaps within an otherwise active session. This is the within-session, intra-action equivalent of full DMN replay: same hippocampal SWR-equivalent machinery, operating at seconds rather than minutes. Viability map consolidation via micro-quiescence replay is safe because all internally-generated replay carries the hypothesis tag (MECH-094), blocking φ(z) writes regardless of timescale.

⸻

Reflective / DMN Mode and Ethical Replay (MECH-029)

In DMN-like cognition:

  • external precision is lowered,
  • internal prediction and replay dominate,
  • hippocampal systems inject episodic material freely,
  • long-horizon models are explored,
  • learning is tentative and low-commitment.

This is not “offline” cognition in a trivial sense. It is where:

  • identity is rehearsed,
  • narratives are integrated,
  • moral simulations occur,
  • regret and counterfactuals are explored without immediate consequence.

Ethically, this mode allows learning without harm. Control-plane gating of what can be replayed here becomes crucial: excessive replay can destabilise, insufficient replay can freeze growth.

Source: docs/thoughts/2026-02-08_modes_of_cognition_control_plane_regimes.md

⸻

Entry and exit conditions

Entry into Default Mode may be triggered by: • absence of immediate task demands • lowered external salience • intentional reflection or planning • sleep onset or quiet wakefulness

Exit from Default Mode occurs when: • sensory precision is up-regulated • task-positive networks engage • E3 commitment thresholds are lowered

These transitions are gradual, not binary.

⸻

Failure modes (for clarity)

Misconfiguration of Default Mode explains known pathologies: • Psychosis • imagined trajectories become committed • pattern completion escapes “hypothesis-only” constraint • Rumination / depression • replay loops excessively within narrow regions of curvature • Distractibility / ADHD-like dynamics • premature Default Mode intrusion into task-engaged operation

These are architectural, not moral, failures.

⸻

Design constraints

The Default Mode must satisfy: 1. No perceptual overwrite It may not directly modify z_S. 2. No forced commitment It may not trigger E3 collapse. 3. No residue update Ethical curvature φ(z) must remain unchanged. Implemented via the hypothesis tag (MECH-094): internally generated trajectories are categorically routed away from the φ(z) write pathway. During DMN, only the pre-commit simulation error channel is active (MECH-060); the post-commit realized-outcome channel — which drives residue accumulation — is suppressed. The qualitatively different affective texture of imagination vs. real experience is the phenomenological signature of this suppression: the absence of the full somatic post-commit signal. φ(z) is read during imagination (producing attenuated aversive affect), but not written to. Pathological failure: loss of the hypothesis tag (PTSD flashback, psychosis) opens the post-commit channel during replay, accumulating residue from simulated content (MECH-076). The damage in rumination is tag loss, not excessive vividness. 4. Hypothesis tagging Generated trajectories must be explicitly non-committal (implements INV-011, INV-019). The tag is categorical, not a precision parameter — route separation is primary, attenuation secondary.

These constraints preserve safety, corrigibility, and identity.

⸻

Interpretation

The Default Mode allows the REE agent to explore:

“Who I have been, who I might be, and what paths exist — without yet deciding.”

It is the architectural basis of: • imagination • mind-wandering • autobiographical selfhood • ethical reflection without action

⸻

Cross-references • Hippocampal systems: hippocampal_systems.md • Latent stack: l_space.md • Trajectory selection (E3): trajectory_selection.md • Residue geometry: residue_geometry.md • Offline reprojection / sleep: sleep.md

⸻

This Default Mode analogue aligns closely with empirical properties of the human default mode network, including reduced sensory precision, increased hippocampal–cortical coupling, and suppression of action-oriented networks. In REE, these properties arise from architectural necessity rather than biological imitation.

Open Questions

None noted in preserved sources.

  • ARC-014
  • ARC-003
  • ARC-005
  • ARC-007
  • ARC-013
  • INV-011
  • INV-019
  • MECH-029
  • MECH-060
  • MECH-076
  • MECH-092
  • MECH-094

References / Source Fragments

  • docs/processed/legacy_tree/docs/architecture/default_mode.md
  • docs/processed/legacy_tree/architecture/Default_mode.md
  • docs/thoughts/2026-02-08_modes_of_cognition_control_plane_regimes.md

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