Versioned
Every note carries a publication date, update date and stable version.
Public research / V1.0
A versioned public record of the questions, definitions and evidence boundaries shaping minimum-state intelligence.
These documents make the public research surface precise without presenting ambition as evidence or exposing protected mechanisms.
Every note carries a publication date, update date and stable version.
Permanent URLs and primary references make each position independently addressable.
Every document states what it establishes—and what it does not.
Start here
Foundation / 001Minimum-state intelligence begins before the choice of an optimization technique. It asks how little physical memory a specific intelligent system can occupy, under a specified workload and runtime, while remaining inside an explicit quality boundary.
The public record
A precise introduction to minimum-state intelligence and the search for the smallest evidence-valid physical memory state of an AI system.
A public framework for distinguishing candidate AI memory reductions from results that are supported by a defined denominator and quality boundary.
Why credible AI memory-reduction claims must distinguish storage, host RAM, accelerator memory, peak allocation and complete runtime state.
A framework for defining task, behavioural and system-level quality boundaries before searching for AI memory reductions.
Why a universal AI memory-reduction system needs one physical objective and architecture-specific models of structure, state and execution.
How the physical memory footprint of AI interacts with accelerators, serving density, power, cooling and infrastructure planning.
Publication rule
Research positions, methods, measurements and results are labelled separately. No reduction figure becomes a finding without its target, denominator, workload, quality definition and authority.
Explore the evidence standard