robbies-razor-benchmarks — Recursive Stability and Compression Efficiency Benchmarks for AI Reasoning Systems

Run a Razor Audit

Evaluate any AI system using Robbie George’s Grand Compression Cosmology:

Reference implementation and benchmarking framework for evaluating Robbie’s Razor compliance, recursive stability, and compression efficiency in reasoning systems operating under constrained compute, memory, and governance bandwidth.

System Architecture Overview

This repository supports Robbie’s Razor™, Naturepedia™, and Plate™ systems as part of a recursive ecological knowledge architecture combining:

• semantic compression
• machine-readable provenance
• recursive relationship mapping
• ecological intelligence architecture
• structured retrieval systems
• low-token semantic traversal

Core reasoning sequence:

compression → expression → memory → recursion

Key concepts: Robbie’s Razor · Grand Compression Cosmology · Recursive Stability · Compression Efficiency · Reasoning Benchmarks

Grand Compression Law of Intelligence

Within the Grand Compression Cosmology, intelligence is modeled through the reuse of preserved compressed structure across recursive cycles under finite resource and stabilization constraints.

The framework proposition can be summarized as:

preserved structure
→ expression
→ memory
→ recursive reuse
→ prediction / adaptation

The core Robbie’s Razor™ sequence is:

compression → expression → memory → recursion

This sequence is a framework architecture.

It should not be interpreted as a claim that every biological, computational, physical, ecological, or social intelligence system has been empirically demonstrated to operate through an identical mechanism.

Accordingly:

Grand Compression framework proposition
≠
universally established law of intelligence

and:

recursive reuse
≠
automatic intelligence

A system must still be evaluated according to its declared task, preserved structure, prediction quality, correctness, resource constraints, and failure conditions.

Recursive Constraint Model

Within the framework, recursive performance may be analyzed using two conceptual constraint classes:

• Energetic Recursion Ceiling — the available energetic budget relative to the cost of coherent transitions;
• Governance Recursion Ceiling — the available stabilization or correction capacity relative to correction demand.

A conceptual energetic ceiling may be written as:

R ≤ E / JCT

A conceptual governance ceiling may be written as:

R × C ≤ S

Combining the two produces the framework-level Safe Recursion Envelope:

R ≤ min(E / JCT, S / C)

Where:

• R = recursion rate;
• E = available energy within the declared system boundary;
• JCT = Joules per Coherent Transition;
• S = available stabilization or governance bandwidth;
• C = correction demand per transition.

These expressions are architectural research relations.

Quantitative application requires operational definitions for every variable, compatible units, measurement procedures, system boundaries, baselines, uncertainty, and falsification conditions.

Without those declarations:

R ≤ E / JCT

R × C ≤ S

and:

R ≤ min(E / JCT, S / C)

must not be represented as universally validated physical laws.

Intelligence Interpretation Boundary

The framework motivates the hypothesis that useful intelligence may depend not only on computation, but on the ability to preserve and recursively reuse structure that remains valid for later tasks.

Possible relevant properties include:

• compression efficiency;
• retained identity;
• preserved relationships;
• provenance;
• memory;
• retrieval fidelity;
• prediction;
• correction;
• adaptation;
• resource cost.

The presence of these properties in an implementation does not independently establish a universal theory of intelligence.

Likewise:

compression
≠
understanding

memory
≠
truth

prediction
≠
causal explanation

recursive stability
≠
factual correctness

Current Governance

Current interpretation is governed by:

The Grand Compression Cosmology — Master Reference Document, MRD v2.0

Canonical identifier:

GC-MRD-v2.0

Repository implementations and benchmark results remain subject to:

• RC-21 — Reference Implementation Distinction
• RC-22 — Domain Transfer Constraint

The repository may test bounded consequences of the Grand Compression Law of Intelligence, but implementation or benchmark success must not be represented as universal empirical confirmation of the law.

Constraint-Bounded Recursive Intelligence

Constraint-Bounded Recursive Intelligence was introduced during the MRD v1.9 development cycle and remains part of the current MRD v2.0 framework.

It models recursive intelligence as an implemented process operating within finite substrate and governance constraints rather than as an unconstrained abstraction.

Relevant constraints may include:

• energy;
• compute;
• memory;
• bandwidth;
• thermal capacity;
• cooling;
• material infrastructure;
• fabrication;
• networking;
• coordination;
• correction capacity;
• governance bandwidth.

The framework expresses a candidate substrate-alignment condition as:

Gᵣ ≤ Eₛ

Where:

• Gᵣ = recursive gain per iteration;
• Eₛ = substrate expansion capacity.

This relation should be interpreted as a framework-level architectural condition.

It is not automatically a dimensionally complete or universally validated physical law.

A quantitative application must define:

• what constitutes recursive gain;
• what constitutes substrate expansion capacity;
• the units used for both quantities;
• the relevant time interval;
• system boundaries;
• normalization;
• baseline;
• uncertainty;
• measurement procedure;
• competing explanations;
• and failure conditions.

Without those declarations:

Gᵣ ≤ Eₛ

should remain a conceptual substrate-alignment relation.

Efficiency vs Expansion

Constraint-Bounded Recursive Intelligence motivates an important architectural distinction:

capability growth through improved efficiency
≠
capability growth through substrate expansion

A recursive system may potentially increase useful work through:

• better compression;
• preserved reusable structure;
• memory reuse;
• reduced recomputation;
• improved retrieval;
• lower correction burden;
• better algorithms;
• better utilization;
• improved hardware.

Physical infrastructure expansion may also increase available capacity.

The framework therefore does not require the claim that compression efficiency is always the primary driver of long-term capability growth.

A safer relation is:

useful recursive capability
may increase through
internal efficiency improvements
and/or
external substrate expansion

The relative contribution of each must be measured for the system being evaluated.

Constraint Response

If recursive demand exceeds an active substrate constraint, possible outcomes may include:

constraint activation
→ adaptation
→ efficiency improvement
→ substrate expansion
→ plateau
→ degradation
→ failure

Different systems may follow different paths.

Therefore:

Gᵣ > Eₛ
≠
automatic collapse

and:

Gᵣ ≤ Eₛ
≠
guaranteed stability

The relation identifies a framework concern about alignment between recursive growth and supporting capacity.

It does not, by itself, determine every system outcome.

Relationship to the Physical Substrate Constraint Field

The repository contains a dedicated engineering orientation for this concept:

docs/physical-substrate-constraint-field.md

That document should be interpreted alongside the current MRD v2.0 architecture.

The relevant distinction is:

internal recursive organization
+
external substrate capacity
→
bounded operating regime

This is a conceptual relationship, not a complete physical equation.

Current Authority

Current governing authority:

The Grand Compression Cosmology — Master Reference Document, MRD v2.0

The historical v1.9 introduction remains part of the development record.

Current interpretation is governed by MRD v2.0.

Canonical authority for the broader recursive engineering architecture spans the current MRD sections governing:

• recursive stability and physical substrate constraints;
• structural intelligence engineering;
• preserved reusable structure;
• predictive evaluation;
• reference implementation;
• and domain transfer.

Repository implementations remain subject to:

• RC-21 — Reference Implementation Distinction
• RC-22 — Domain Transfer Constraint

Accordingly:

framework constraint relation
≠
empirical physical law

and:

reference implementation
≠
universal confirmation

Robbie’s Razor Architecture

Within the Grand Compression Cosmology, Robbie’s Razor™ provides a reference architecture for organizing recursive information processing around:

compression → expression → memory → recursion

A broader implementation-oriented loop may be represented as:

Environment
    │
    ▼
Observation
    │
    ▼
Compression
    │
    ▼
Expression
    │
    ▼
Memory
    │
    ▼
Recursion
    │
    ▼
Prediction
    │
    ▼
Action
    │
    ▼
Feedback
    │
    ▼
Memory Update
    │
    ▼
Recompression
    └──────────────→ renewed observation / processing

This is a Grand Compression reference architecture.

Read the full README on GitHub →