Thought leadership

Blog

Essays on research discovery, causal dynamics, and governed science—written for teams who need claims that survive replay, not just screenshots.

Featured Research

Explore research

Theory · Featured Work

Theory of Compressive Realism: Open Systems, Viability, and Regime-Indexed Laws as Stable Compressions

An operational framework unifying epistemology and ontology. Scientific laws are the most stable compressions of boundary-accessible information. Thermodynamics, microphysics, geometry, and spacetime are all earned descriptions — revisable under strict audit discipline. 18 operational commitments. 13 theorems. 5 empirical programs. No metaphysical privilege.

Read the theory paper

March 28, 2026·Benchmark·12 min read

ARDA's Autonomous Discovery: Validated on Real Physics

Three real-world systems with known ground truth. Zero human input. Governing equations, causal graphs, and cyclic regulatory topologies — all discovered autonomously and verified against established science.

Read featured post

More Research Papers

Benchmark · March 2026

Artifact-Complete Benchmarking of ARDA on PerturBench Norman19

ARDA predicted cellular responses to gene perturbations never run in the lab. Achieved the highest score on the PerturBench Norman19 frozen test split with Cosine LogFC 0.9109 — benchmarked against GEARS and baselines under identical scoring rules.

Read the full research

Benchmark · March 2026

MatterSpace: Blind Rediscovery of Single-Atom Alloy Catalysts

600 candidates across 23 dopant elements. Zero target knowledge. MatterSpace independently rediscovered Re₁@Ni and Ir₁@Ni catalysts to sub-angstrom accuracy (0.408 Å RMSD) — valid by construction, $15 cloud cost.

Read the full research

CDE · March 2026

Causal Structure Invariance Under Distribution Shift

CDE produces structurally invariant representations. Graph entropy varies by 0.22% across regimes where neural networks degrade by 639–1,127%. Validated on NASA CMAPSS and Tennessee Eastman.

Read the full research

CDE · March 2026

CDE Industrial Case Studies: Where Neural Networks Collapse

How the Causal Dynamics Engine holds where neural networks collapse. 0.997 path fidelity on Tennessee Eastman. Causal graph invariant across all four CMAPSS operating regimes.

Read the full research

CDE · March 2026

Multi-Modal Scientific Discovery via the Causal Dynamics Engine

Systematic ablation study: spatial + graph input reduces causal ambiguity by 2.7× on spring-mass networks. Scientific integrity validated — no false improvement when extra modalities carry no information.

Read the full research

Latest

March 25, 2026·MatterSpace

The Case for a Universal Discovery Engine

Materials, drugs, algorithms, chips, biology — different science, same computational structure. Why one engine with domain packs beats fragmented per-domain tools.

7 min read

March 24, 2026·Announcement

Introducing ARDA: The Research Discovery Engine

What changes when discovery is not a menu of tools but an engine that profiles data, routes modes, runs controls, and promotes only what the evidence ledger can replay? A grounded introduction to ARDA and how it works.

8 min read

March 20, 2026·Perspective

Why Engines Beat Assistants in Scientific Discovery

Assisted AI accelerates tasks. A discovery engine absorbs the coordination cost—method selection, control batteries, promotion gates—so teams scale the scientific method instead of headcount.

6 min read

March 15, 2026·Technical

Why Causal Discovery Changes Everything

Mechanism-level claims, intervention reasoning, and governed causal structure—how the Causal Dynamics Engine (CDE) differs from correlation-rich models that only predict.

12 min read

March 10, 2026·Perspective

Why Governing Equations Matter More Than Predictions

Predictions age; laws compound. A case for typed LawClaim outputs, conservation checks, and interpretable structure as the durable asset—especially when regulators and partners ask how you know.

7 min read

March 5, 2026·Enterprise

Enterprise Governance for Research Discovery

Truth Dial tiers, autonomy policies, evidence ledgers, and publish bundles—how enterprises keep exploration fast without letting sketches masquerade as external-grade claims.

9 min read