MARGINAL was created by SignalLayer Labs as an independent Apache-2.0 open-source implementation for economically disciplined AI-agent compute allocation.
The project’s economic framing is inspired by Siqi Zhu’s 2026 position paper, “Agentic AI Systems Should Be Designed as Marginal Token Allocators”: https://arxiv.org/abs/2605.01214.
The paper and this software are separate works. The paper proposes a research agenda; MARGINAL provides an independently developed runtime API, transactional accounting model, Decision Ledger, tests, documentation, and universal adapter foundation.
We also acknowledge the broader open-source and research community working on agent budgets, cost observability, model routing, test-time scaling, causal evaluation, and reliable orchestration.