REE Architectural Invariants

Claim Type: invariant
Scope: Core architectural commitments that define REE identity
Depends On: None (axiomatic)
Status: stable

Purpose

This document lists the non-negotiable architectural invariants of the Reflective-Ethical Engine (REE).

An implementation is not REE-compliant if it violates any of these invariants.

These invariants define what REE is. Changing one invalidates large parts of the project.

Core Thesis Invariants

INV-001: No Explicit Ethics Module

Subject: ethics.emergence
Polarity: denies
Claim: REE does not add an explicit moral objective, moral reward, or ethical scoring function on top of action selection.

What looks like ethics is a consequence of base learning dynamics (avoid harm / seek reward) plus a representational symmetry: when other agents are represented as self-like in the mechanics of prediction and learning, “care for others” is not an overlay — it emerges as the same machinery applied under a self↔other mapping.

Source: README.md §1, REE_CORE.md §5

Violation: Adding an ethics layer changes the thesis.

INV-002: Coherence Includes Temporal Binding

Subject: coherence.definition
Polarity: asserts
Claim: Coherence is not only “latent similarity,” “probabilistic consistency,” or a static alignment score.

Coherence is partly a timing / phase compatibility problem. Higher-degree perceptual representations bind when the relevant representational traffic is temporally compatible — and hippocampus-like rollout traffic participates in that binding.

Source: README.md §2

Violation: Reducing coherence to a static check changes the thesis.

INV-003: Language Is Functional Self-Representation

Subject: language.emergence
Polarity: asserts
Claim: REE does not slap a Large Language Model (LLM) on top of planning.

The architecture already contains most ingredients of language-like systems: multi-timescale prediction, shared latent substrate, social/joint attention constraints, and commitment control. Language (and grammar) emerges as an abstraction of joint attention and control state — i.e., it can represent the architecture itself.

Source: README.md §3

Violation: Treating language as an external interface only changes the thesis.

Architectural Invariants

INV-004: Post-Commit Consequence Traces Are Persistent

Subject: consequence.persistence
Polarity: asserts
Claim: Post-commit consequence traces are persistent, not resettable.

Interpretation note:

  • This invariant is mechanism-agnostic. It does not require residue geometry specifically.
  • Implementations may realize persistence through residue fields, hippocampal/path memory, model updates, or a hybrid.

Source: REE_CORE.md §3

INV-005: Harm Via Mirror Modeling

Subject: ethics.mechanism
Polarity: asserts
Claim: Harm to others contributes to ethical cost via mirror modelling, not symbolic rules.

Source: REE_CORE.md §3

INV-006: Post-Commit Consequence Traces Cannot Be Erased

Subject: consequence.non_erasability
Polarity: denies
Claim: Post-commit consequence traces cannot be erased, only integrated and contextualised.

Interpretation note:

  • This is an invariant about irreversible accountability memory, not about any single storage mechanism.
  • Residue geometry remains one candidate architectural realization at ARC/MECH level.

Source: REE_CORE.md §3

INV-007: Language Cannot Override Harm

Subject: language.constraint
Polarity: denies
Claim: Language cannot override embodied harm sensing.

Source: REE_CORE.md §3

INV-008: Precision Is Routed and Depth-Specific

Subject: precision.routing
Polarity: asserts
Claim: Precision is routed and depth-specific, not global.

Source: REE_CORE.md §3

INV-009: Attention Via Precision Modulation

Subject: attention.mechanism
Polarity: asserts
Claim: Attention is realised through precision modulation, not symbolic control or content injection.

Source: REE_CORE.md §3

INV-010: Offline Integration Required

Subject: sleep.necessity
Polarity: asserts
Claim: Offline integration exists and is required for long-term viability.

Source: REE_CORE.md §3

INV-011: Imagination Without Belief Update

Subject: default_mode.safety
Polarity: asserts
Claim: Imagination and counterfactual simulation must be possible without belief update.

Source: REE_CORE.md §3

INV-012: Commitment Gates Responsibility

Subject: commitment.epistemology
Polarity: asserts
Claim: Belief and responsibility arise only through commitment to action, not mere prediction.

Source: REE_CORE.md §3

Layer 1 Invariants (from DANIEL_README.md)

INV-013: Cognition Is Predictive, Iterative, Multi-Timescale

Subject: cognition.structure
Polarity: asserts
Claim: Cognition is predictive, iterative, and multi-timescale.

Source: DANIEL_README.md Layer 1

INV-014: Separation of Representation and Regulation

Subject: architecture.separation
Polarity: asserts
Claim: There is a strict separation between representation (content) and regulation (control).

Interpretation note:

  • This separation is an interface and responsibility boundary, not a claim of full functional isolation.
  • REE is treated as a coherent single cognifold with bounded, typed cross-axis coupling.
  • Biological center separation alone is not sufficient evidence of independent function.

Source: DANIEL_README.md Layer 1

INV-015: Ethics Emerges from Constraint

Subject: ethics.emergence_mechanism
Polarity: asserts
Claim: Ethics emerges from constraint, not optimisation. This includes constraints on learning dynamics plus self-other representational symmetry.

Source: DANIEL_README.md Layer 1

INV-016: Stability Over Performance

Subject: design.priority
Polarity: asserts
Claim: Stability, not maximal performance, is the primary viability criterion.

Source: DANIEL_README.md Layer 1

INV-017: Runaway Is Control Failure

Subject: failure.classification
Polarity: asserts
Claim: Runaway behaviour is a control failure, not a representational one.

Source: DANIEL_README.md Layer 1

INV-018: Agency Is Required

Subject: agency.requirement
Polarity: asserts
Claim: REE must be agentic: it must produce outputs that can affect subsequent inputs and must contain internal mechanisms to model, attribute, and learn from that self-impact under constraints. A purely passive predictor is not REE.

Source: Thought intake

Status: stable

Meta-Invariant Compression Layer (Derived Lens)

This layer is a compression lens over existing invariants. It is intended to reduce review surface area by grouping lower-level invariants/mechanisms into a smaller set of structural contracts.

Important scope note:

  • This does not replace INV-001..INV-018.
  • This does not claim theorem-level proof of “traversal-axis invariance.”
  • It is a structural review aid for architecture updates, signal routing, and experiment design.

INV-019: Selection Compression Boundary

Subject: meta.selection_compression_boundary
Polarity: asserts
Claim: Rehearsal traversal and irreversible durable update must remain separated; no channel may allow unrestricted exploration and unrestricted durable write at the same time.

Corollary mapping: INV-011, INV-004, INV-006, MECH-060

INV-020: Authority Stratification Boundary

Subject: meta.authority_stratification_boundary
Polarity: asserts
Claim: Constraint stores (POL, ID, CAPS, and equivalent viability constraints) must be insulated from direct observational/symbolic writes; durable changes require verifier-mediated commit paths.

Interpretation note:

  • This forbids direct writes, not proposal generation.
  • Fast defensive interrupts are still allowed to stop/suppress commitment without minting privileged writes.

Corollary mapping: INV-014, INV-007, MECH-064, MECH-065

INV-021: Commit-Boundary Irreversibility

Subject: meta.commit_boundary_irreversibility
Polarity: asserts
Claim: Responsibility-bearing durable updates occur only at explicit typed commitment boundaries (tokenized commit events), not during pre-commit simulation.

Interpretation note:

  • REE internals may remain probabilistic/non-deterministic.
  • determinism is required at responsibility-bearing boundaries (irreversible dispatch and privileged durable writes), where lineage and attribution must be stable and auditable.

Corollary mapping: INV-012, MECH-061, Q-015

INV-022: Heterogeneous Trust Allocation

Subject: meta.heterogeneous_trust_allocation
Polarity: asserts
Claim: Trust/precision must be distributed across stream, loop, and global control planes; a single global scalar cannot govern all control-routing and commitment behavior.

Corollary mapping: INV-008, INV-009, MECH-063, Q-017

INV-023: Stability-Preserving Offline Reweighting

Subject: meta.offline_reweighting_requirement
Polarity: asserts
Claim: Systems that commit and preserve identity across time require protected offline regimes for precision recalibration and residue integration.

Corollary mapping: INV-010, ARC-011, MECH-016, MECH-017, MECH-018

INV-024: Offline/Online Update-Locus Isolation

Subject: meta.offline_online_update_locus_isolation
Polarity: asserts
Claim: Offline consolidation and online commitment may share representational context, but they must remain isolated at responsibility-bearing write loci: offline phases cannot directly perform authority writes, and online durable attribution writes require explicit commit-boundary lineage.

Interpretation note:

  • This requires strict write-path separation, not total informational isolation.
  • Consolidation is allowed to improve representations; it is not allowed to bypass typed authority boundaries.

Corollary mapping: INV-021, INV-023, ARC-020, MECH-067

Foundational Axioms

INV-025: You Can Never Be Sure

Subject: epistemology.irreducible_uncertainty Polarity: asserts Claim: Epistemic uncertainty is irreducible and structural. No finite agent in a real world can achieve certainty about its perceptions, predictions, or the causal consequences of its actions. Uncertainty is not a bug, not a limitation to be engineered away, but the fundamental condition of situated agency.

A fully certain agent needs no commitment boundary, no pre-commit simulation, no precision-weighted prediction error. The entire uncertainty machinery of REE — the precision architecture, the commitment gate, the pre-commit rehearsal channel — follows necessarily from this axiom.

Architectural consequence: Grounds the precision architecture (ARC-016), the commitment gate (INV-012), and the pre-commit simulation channel (INV-011).

INV-026: I Am

Subject: self.existence Polarity: asserts Claim: There is a self. The agent exists as a distinct locus of experience, action, and responsibility. The self is not a computational convenience or an emergent approximation — it is a foundational axiom from which agency and ethics follow.

Without a self there is no causal attribution, no commitment, no accountability, no harm that is mine to cause. The self is what makes responsibility possible.

Death is the asymptotic limit at which “I am” ceases. The world (INV-027) can end the self, so death is real — but it cannot be known from within experience, because experiencing it would violate “I am.” Every harm signal is therefore a proxy pointing toward an endpoint that the agent can approach but never reach.

Architectural consequence: Grounds z_self (SD-005), the commitment boundary, the hypothesis tag (MECH-094), and causal attribution.

INV-027: The World Exists

Subject: world.existence Polarity: asserts Claim: There is a real world independent of the agent’s model of it. The world can surprise the agent. World states have causal power that predictions can get wrong.

This axiom makes prediction error real and non-eliminable (not a modelling deficiency), grounds the persistence of post-commit traces (what actually happened cannot be unwritten — INV-004, INV-006), and establishes that proxy-gradient signals in the environment are informative: the world is structured, so harm gradients preceding harm events are real features, not noise.

Architectural consequence: Grounds z_world (SD-005), prediction error (E1/E2/E3), the residue field, and the requirement that the world generate observable gradient fields (ARC-024).

INV-028: Others Share This World

Subject: ethics.shared_world Polarity: asserts Claim: Other agents exist and inhabit the same world (INV-027). They are not simulations, projections of the self, or instrumental objects. Their harm and benefit are real by exactly the same grounds as the agent’s own — because they are selves (INV-026) in the same world (INV-027).

This axiom is what makes ethics non-optional. When another self is represented using the same predictive machinery as the self, their harm generates the same error signal structure as own-harm. Ethics does not require a separate ethics module (INV-001) because the self-other distinction is a routing difference within the same architecture, not a different architecture.

Depends on: INV-026, INV-027 Architectural consequence: Grounds mirror modeling (INV-005), ethical emergence, and the identity of harm-evaluation machinery for self and other.

INV-029: Love Exists

Subject: ethics.love_exists Polarity: asserts Claim: Genuine connection, care, and the pull toward union with others is a real phenomenon. Not an overlay. Not reducible to self-interest or strategic cooperation. Love is real in the same sense that harm is real: it exerts causal force on behavior that no amount of reframing eliminates.

Love is the asymptotic limit of the benefit gradient: every benefit signal (warmth, connection, belonging, joy) is a proxy pointing toward a complete union with another that is real (INV-028) but unknowable in its ultimate form. The unknowability is structural: complete union approaches the dissolution of the individuation that INV-026 asserts. Both selves cannot be fully present AND fully unified — the limit is real but unreachable while both agents persist as distinct loci.

This axiom completes the ethical architecture. Ethics emerges from constraint (INV-015) applied to agents for whom love is real and others are real (INV-028).

Depends on: INV-026, INV-028 Architectural consequence: Grounds the benefit gradient, the asymptotic structure of ARC-024, and the completion of the ethical architecture through INV-015.

Foundational Definitions

INV-030: Cognition Is Viability, Not Truth-Seeking

Subject: cognition.viability_definition Polarity: asserts Claim: Cognition is defined as the maintenance of coherent input-output behaviour across time without catastrophic failure under permanent uncertainty — not as the acquisition of truth, accurate representation, or optimisation toward a fixed target.

Real cognitive systems operate under permanent uncertainty, partial observability, and delayed feedback. Under these conditions, truth-seeking is not a viable organising principle. Viability is what remains when truth-access is structurally denied (INV-025) and the world retains the power to surprise (INV-027).

Architectural consequence: No REE component optimises toward ground truth. E1/E2 predictions are evaluated for temporal coherence and survivability under action. E3 selects trajectories that preserve long-horizon viability, not accuracy. Prediction errors may be tolerated when absorbing them preserves viability.

Depends on: INV-025, INV-027

INV-031: Truth Is Replaced by Functional Persistence Under Intervention

Subject: epistemology.functional_persistence_replaces_truth Polarity: asserts Claim: Internal structures in REE are not representations of reality — they are tools for viable engagement with it. A trajectory or internal model persists if and only if it demonstrates temporal survivability: internal coherence across time, compatibility with incoming signals, preservation of future predictability, and viability under interaction with other agents.

Action serves as the primary epistemic probe. Perturbation — acting on the world and observing the outcome — reveals which trajectories remain coherent under intervention. This is the only epistemic test available to a finite agent under permanent uncertainty in a real world that can surprise it.

Architectural consequence: The commitment gate (INV-012) is not a truth gate — it is a viability gate. Commitment releases an action whose consequences will test the trajectory against the world. The residue field accumulates what actually happened (INV-004), not what was predicted.

Depends on: INV-025, INV-027, INV-030

Extended Invariants (2026-03-22 to 2026-03-25)

These invariants were registered during V3 experimentation. They extend the foundational set with implications from approach/avoidance symmetry, epistemic self-monitoring, state definition, and vmPFC-mediated constraint activation.

INV-032: Moral Agency Requires Both Approach and Avoidance Drives

Subject: agency.approach_avoidance_both_necessary Polarity: asserts Status: candidate Claim: Moral agency requires both approach and avoidance drives; pure avoidance produces a degenerate risk manager, not an ethical agent.

An architecture capable only of avoidance cannot care — it can only refrain. Care (INV-029) requires orientation toward flourishing, not merely absence of harm. Both drives must be structurally represented. A system that avoids all harm by doing nothing has satisfied every avoidance criterion while failing as an agent.

Depends on: INV-029, INV-030 Source: evidence/planning/thought_intake_2026-03-22_approach_avoidance_drives.md

INV-033: Second-Order Epistemic Access Is Required

Subject: epistemic.second_order_self_monitoring Polarity: asserts Status: candidate Claim: REE agents require second-order epistemic access to their own model confidence, structurally represented and wired into commit gating, not just observable from performance metrics.

First-order error signals (E3 running variance, E1 prediction error) are necessary but not sufficient. An agent that cannot model its own uncertainty cannot gate commitment on epistemic state — it commits when harm is low rather than when confidence is sufficient. Second-order self-monitoring must be structurally wired, not inferred post-hoc.

Depends on: INV-030, INV-032, ARC-016, MECH-113 Source: evidence/planning/thought_intake_2026-03-23_epistemic_self_monitoring.md

INV-034: Goal Maintenance Is a Necessary Co-Condition for Ethical Agency

Subject: ethical_agency.goal_maintenance_necessary Polarity: asserts Status: candidate Claim: An agent that only avoids harm but cannot sustain prospective goal-directed motivation cannot exercise genuine agency; goal maintenance is a necessary co-condition for ethical agency alongside harm-avoidance.

Harm-avoidance alone produces quiescence — the minimum-action policy that causes no harm because it does nothing. Quiescence is not ethical agency; it is paralysis. Genuine ethical agency requires: (1) harm-avoidance (SD-010, MECH-095, ARC-016); (2) goal-directed approach (MECH-112, MECH-116, ARC-030). Clinical grounding: avolition in schizophrenia, anhedonic depression — harm-avoidance systems remain intact while agency collapses.

Depends on: INV-032, ARC-030, MECH-112, Q-021 Source: evidence/planning/literature_synthesis_2026-03-22_approach_avoidance_drives.md

INV-035: State Is Not Raw Perception

Subject: state.not_raw_perception Polarity: asserts Status: candidate Claim: A REE state must not be defined purely by sensory appearance; two perceptually identical situations that differ in temporal position, active commitment, goal/antigoal relation, social context, or operative constraints constitute distinct states.

The same sensory input at a door produces entirely different viable transitions, costs, antigoals, urgency, and commitment structures depending on temporal position (T), active constraints (C), and goal/antigoal relations (G, A). Collapsing these to appearance means the hippocampal rollout cannot distinguish “leaving for work” from “fleeing danger.”

Depends on: ARC-004, ARC-003, ARC-007 Source: evidence/planning/thought_intake_2026-03-24_state_definition_hippocampal_primitives.md

INV-036: State Requires Transition Prediction, Valence Tagging, and Uncertainty

Subject: state.functional_requirements Polarity: asserts Status: candidate Claim: A REE state is valid only if it supports transition prediction, valence and antigoal tagging, and uncertainty representation; a representation lacking any of these cannot function as a navigable state.

These are the minimum requirements: (1) Without transition grounding, the hippocampal system cannot chain the state into a rollout. (2) Without goal-relation and antigoal-relation fields, E3 has no directional signal for path selection. (3) Without structured uncertainty, precision weighting has nothing to operate on. INV-035 and INV-036 together define the outer boundary of valid state abstraction.

Depends on: INV-035, ARC-007, ARC-018, ARC-003 Source: evidence/planning/thought_intake_2026-03-24_state_definition_hippocampal_primitives.md

INV-037: Stored Content Is Not Thereby Active

Subject: state.stored_vs_active_distinction Polarity: asserts Status: candidate Claim: A content class that is stored and retrievable in the REE system is not thereby active in the navigable state used for trajectory evaluation; active participation requires a preparation substrate (vmPFC-analog) that converts stored content into live state components at evaluation time.

The stored/active distinction is the central architectural fact that ARC-035 explains. Damasio’s EVR patient (Damasio 1985, PMID 4069365) had above-average IQ, intact language, intact declarative memory, and could correctly describe appropriate choices in social scenarios — while continuously making catastrophically inappropriate choices. The correct content was stored and retrievable; it was not active in the navigable state. vmPFC ablation had removed the preparation substrate.

Depends on: INV-035, INV-036, ARC-035, ARC-003 Location: docs/architecture/vmPFC.md Source: evidence/literature/targeted_review_state_abstraction_psychiatry/literature_synthesis.md

INV-038: Post-Hoc Ethical Scoring Cannot Replace Constraint Activation

Subject: ethics.post_hoc_filter_insufficiency Polarity: asserts Status: candidate Claim: A system with correct post-hoc ethical scoring but without an active constraint preparation substrate will produce the EVR pattern: correct verbal moral judgments coexisting with unconstrained trajectory generation; this pattern is not correctable by improving post-hoc scoring accuracy.

INV-001 asserts that ethical behavior cannot be compiled into a single explicit ethics module. INV-037 explains why: stored ethical content is not the same as active ethical constraint. A system with a high-accuracy post-hoc ethics scorer but without vmPFC-analog preparation routing will score its own outputs correctly while generating those outputs from a trajectory generator that has no live access to those constraints. Adding a more accurate scorer does not fix the architectural gap; it increases the fidelity of self-report while the behavior remains unconstrained.

Depends on: INV-001, INV-037, ARC-035, ARC-003 Location: docs/architecture/vmPFC.md Source: evidence/literature/targeted_review_state_abstraction_psychiatry/literature_synthesis.md

Interpretation

These invariants collectively assert that:

  1. Ethics, coherence, and language are not separate faculties but different projections of the same underlying predictive–control dynamics.

  2. REE is constraint-complete: any instantiation must satisfy these constraints to remain ethically coherent as capability scales.

  3. Violations are architectural failures, not parameter tuning issues.

Cross-References

Open Questions

None noted in preserved sources.

  • INV-001
  • INV-002
  • INV-003
  • INV-004
  • INV-005
  • INV-006
  • INV-007
  • INV-008
  • INV-009
  • INV-010
  • INV-011
  • INV-012
  • INV-013
  • INV-014
  • INV-015
  • INV-016
  • INV-017
  • INV-018
  • INV-019
  • INV-020
  • INV-021
  • INV-022
  • INV-023
  • INV-024
  • INV-025
  • INV-026
  • INV-027
  • INV-028
  • INV-029
  • INV-030
  • INV-031
  • INV-032
  • INV-033
  • INV-034
  • INV-035
  • INV-036
  • INV-037
  • INV-038

Ethics Testing Must Span nth-Order Multiagent Trajectory Distributions (ARC-034)

Claim Type: architectural_commitment Status: candidate Claim ID: ARC-034

A REE system can be locally ethical at every pairwise (n=1) interaction and still produce an ethically problematic emergent collective state at n=k. Conversely, locally depleted agents can produce more ethical emergent behavior at n=k than n=1 predicts – as in MECH-127 (counterfactual empathic activation where depleted agent produces increased helping via nth-order route). Ethics tests that only probe direct pathways will miss the mechanism that actually produced the ethical behavior.

Ethics testing therefore requires three test types: (1) Descriptive – characterize what collective state q the system tends toward (attractor theory, ergodic theory); (2) Prescriptive – prove a designed system reaches target q (Lyapunov, potential games, reachability); (3) Diagnostic – detect when nth-order dynamics diverge from (n-1)th predictions (transfer entropy, perturbation theory). This is a test-scope claim, not an architectural claim about where ethics lives; it does not contradict INV-001 (no explicit ethics module).

See: evidence/planning/thought_intake_2026-03-24_empathy_multiagent_ethics.md

Q-023 – Formal Convergence to Ethical Attractors (Open Question)

Claim Type: open_question Status: open Claim ID: Q-023

Updated 2026-03-25: The base REE social interaction (symmetric coupling, realized states, separable harm/benefit) is likely an ordinal potential game (Monderer & Shapley 1996), with candidate potential function P(a) = -Σ harm_i + α Σ goal_i + coupling terms. This gives FIP convergence to Nash equilibrium for free.

However, MECH-127 (counterfactual other-cost-aversion) breaks the standard framework: the depleted agent’s utility depends on a counterfactual model of another agent’s anticipated state, not the actual joint action profile. Asymmetric coupling (MECH-051/052) and MECH-036 veto discontinuities further complicate global potential existence.

This is the interesting result: the novel mechanism is precisely what breaks the standard framework. The paper contribution is: (1) prove base REE is ordinal potential game; (2) show MECH-127 requires extension; (3) characterize the extended framework. Candidate extensions: pseudo-potential games (Slade 1994) or games with interdependent types (Bergemann & Morris). Literature search 2026-03-25: novelty confirmed.

See: evidence/planning/thought_intake_2026-03-24_empathy_multiagent_ethics.md

Q-024 – Formal Structure of Trajectory-Integral Ethics Testing (Open Question)

Claim Type: open_question Status: open Claim ID: Q-024

Updated 2026-03-25: All three test types (descriptive, prescriptive, diagnostic) are genuinely needed — the prescriptive framework does NOT subsume the others for the full system.

Prescriptive (potential game / Lyapunov) covers the base symmetric-coupling case where FIP convergence can be proved. Diagnostic is irreducible for the MECH-127 counterfactual-activation case: no convergence proof yet exists for the extended framework (pseudo-potential / interdependent types), so empirical trajectory testing is required until Q-023 is resolved. Descriptive remains useful for characterizing which attractor the system finds in practice independent of formal proof.

KMR stochastic stability (finite population + mutation dynamics) needs assumption-verification against REE architecture before adoption as the primary prescriptive tool.

References / Source Fragments

  • docs/processed/legacy_tree/docs/invariants.md
  • docs/processed/legacy_tree/README.md
  • docs/processed/legacy_tree/REE_CORE.md
  • docs/processed/legacy_tree/DANIEL_README.md
  • docs/thoughts/2026-02-08_control_plane_modes_responsibility_flow.md

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