Research

Research claims bounded by
reproducible evidence.

AXIOM investigates agent execution, retrieval durability, and control architectures through source-grounded analysis, reproducible tests, and explicitly bounded theoretical work.

Current agenda

Research questions with operational consequences.

01 / Execution

Effects, recovery, and policy identity

This work examines what can be durably established before and after tool dispatch, and which authority and reconciliation contracts safe compensation would require.

Transactional effects research ↗

02 / Retrieval

Storage ownership and crash consistency

This work evaluates single-owner persistence, replacement journals, embedding-space identity, and the limits of process-crash evidence.

RAG crash-consistency architecture ↗

03 / Control

Loops, optimization, and observability

This work develops falsifiable descriptions of iterative and recursive systems without treating toy formalizations as evidence about hidden model internals.

J-space research program ↗

Research record

Research artifacts with explicit scope and status.

Current source and passing tests in the owning implementation repository take precedence over earlier descriptions.

AI loop architecture taxonomyArchitectural taxonomy; quantitative claims remain workload-specific. RAG HTTP compatibility contractDesign analysis behind the versioned caller/server boundary. Write-file compensation readinessConstraint audit; automatic compensation remains blocked. APEX deterministic execution contractsHistorical June 2026 publication; later source-grounded findings supersede conflicts.

Validated organizational simulation case studies View the research repository

Engineering practice and stack