Study finds open-source memory components fail 19x more under single-vendor control

LLM agent memory is now a first-class architectural component in five major surveys published in early 2026, yet none treat governance, capital structure, or license posture as architectural variables. Dmitrii Dmitrenko's paper argues they are critical. Over eight years (2018-2026), the author cataloged 38 license-and-sustainability events across 105 production-relevant open-source data-infrastructure and AI-tooling projects. About 24% of projects experienced at least one adverse event. The conditional rates split sharply: 46% for single-vendor venture-backed projects, but only 2.5% for foundation-governed projects funded outside the venture cycle. This 19-fold differential is robust to coding choices. Foundation governance with a venture-backed steward (n=3) contained one adverse event, suggesting foundation rules may block unilateral relicensing while leaving distribution decisions to the steward.

Annualized incidence rose from 2.7 to 4.2 events per year across the window. Counterfactuals like PostgreSQL, pgvector, SQLite, Apache Kafka, and Caddy each showed stability from different structural sources: distributed copyright, absence of monetisation pressure, foundation governance with non-venture stewardship. The paper proposes a six-field decision instrument for architects: governance, capital structure, license, foundation membership, fork-or-migration availability, and steward concentration. As open-source memory components become critical for LLM agents, these findings provide a framework to avoid sudden license changes, project abandonment, or sustainability crises.