Skip to main content

Blog

Benchmarks, reports, and technical work.

Across CDE, MatterSpace, and ACI, from shorter technical posts to full benchmark reports.

Vareon blog

By System

ACI

Continual Learning After Deployment

ACI

Post-launch model change for shared services, personal devices, and edge systems. Plasticity adds new behavior, stability protects what must not regress, and editability removes learned contributions exactly.

Jump to section
CDE

The Causal Dynamics Engine for Science and Engineering

CDE

Discovery workflows for governing equations, causal graphs, and conservation laws.

Jump to section
MatterSpace

The Universal Generation Engine for Science and Engineering

MatterSpace

Goal-driven inverse generation for Lattice materials workflows and Vital longevity workflows.

Jump to section

ACI

Benchmarks for shared-service updates, local personalization, edge runtime, and typed policy.

ACI posts cover measured behavior across shared-service updates, device-local agents, compiled edge runtime, and policy enforcement surfaces.

ACI Inference Benchmark Results
April 5, 2026-Benchmark-6 min read

ACI Inference Benchmark Results

Shared-service tenants matched a simple linear control while holding tenant isolation and rollback match at 1.0.

Read featured post
ACI Personal Agents Benchmark Results
April 5, 2026-Benchmark

ACI Personal Agents Benchmark Results

Local personalization improved behavior over the base profile, restored exactly from snapshot, and returned cleanly to base after tagged forget.

6 min read

ACI Edge Runtime Benchmark Results
April 5, 2026-Benchmark

ACI Edge Runtime Benchmark Results

Measured local bind and inference paths, signed audit behavior, native packaging, and fixed-task comparison on the compiled runtime.

6 min read

ACI Safety & Policy Benchmark Results
April 5, 2026-Benchmark

ACI Safety & Policy Benchmark Results

Measured typed rule enforcement across personal-agent and edge-runtime deployments with explicit positive-control and spillover rates.

5 min read

CDE

Reports, benchmarks, and discovery workflows.

CDE posts cover discovery workflows, causal structure, governing equations, and benchmark evidence.

CDE's Autonomous Discovery: Validated on Real Physics
March 28, 2026-Benchmark-12 min read

CDE's Autonomous Discovery: Validated on Real Physics

Validated across known systems with typed outputs, negative controls, and governed replay support.

Read featured post
Introducing CDE: The Causal Dynamics Engine for Science and Engineering
March 24, 2026-Announcement

Introducing CDE: The Causal Dynamics Engine for Science and Engineering

Why the unit of work should be a governed discovery engine rather than a loose collection of tools.

8 min read

What a Discovery Engine Actually Does
March 20, 2026-Perspective

What a Discovery Engine Actually Does

Scientific coordination, controls, replay, and memory do not scale when humans remain the glue.

6 min read

Causal Discovery: From Correlation to Mechanism
March 15, 2026-Technical

Causal Discovery: From Correlation to Mechanism

Mechanism-level outputs matter when the real question is intervention, not just prediction.

12 min read

MatterSpace

Reports, rediscovery studies, and generation workflows.

MatterSpace posts cover constrained generation, rediscovery studies, and Lattice or Vital workflows under declared rules.

The Case for a Causal Dynamics Engine
March 25, 2026-MatterSpace-7 min read

The Case for a Causal Dynamics Engine

Different sciences, similar computational structure. How a goal-driven inverse generation engine spans multiple candidate-generation workflows.

Read featured post
MatterSpace Lattice: Blind Rediscovery of Single-Atom Alloy Catalysts
March 2026-Benchmark

MatterSpace Lattice: Blind Rediscovery of Single-Atom Alloy Catalysts

A rediscovery study showing constrained generation and matched-structure recovery rather than vague design-language claims.

10 min read

MatterSpace Vital: Blind Rediscovery of Sealed Longevity Targets
April 2026-Benchmark

MatterSpace Vital: Blind Rediscovery of Sealed Longevity Targets

Exact top-1 blind rediscovery in 3/3 sealed longevity challenge families under the current frozen benchmark package.

8 min read

Reports and Technical Work

Reports and technical work.

Long-form benchmark reports and technical articles across the three systems.

Theory

Theory of Compressive Realism

Foundational theory on stable compressions, auditable revision, and why law-like structure needs disciplined replay.

Read report
Benchmark

CDE on PerturBench Norman19

PerturBench benchmark result and supporting evidence for CDE's biology and perturbation work.

Read report
Benchmark

MatterSpace Lattice Rediscovery Study

Benchmark results from MatterSpace's goal-driven inverse generation engine across constrained candidate workflows.

Read report
Benchmark

MatterSpace Vital Blind Rediscovery

Exact top-1 blind recovery in 3/3 sealed longevity challenge families: senescence clearance, epigenetic reprogramming, and stem-cell rejuvenation.

Read report
Benchmark

ACI Inference Benchmark Results

Accuracy parity on shared features plus tenant isolation, rollback, and tenant delete on the live surface.

Read report
Benchmark

ACI Personal Agents Benchmark Results

Local personalization, exact restore parity, and clean return-to-base after tagged forget.

Read report