Ziran System: Deficit-Triggered Self-Revision of Observation Relations in Autonomous Science

This paper develops a Ziran System account of deficit-triggered self-revision in autonomous science. Rather than assuming that observation relations, variables, sensors, scales, temporal windows, and inferential interfaces remain fixed throughout inquiry, the framework treats them as revisable components that can be modified when a scientific system detects persistent deficits in prediction, discrimination, intervention, or reconstruction. A deficit is operationally significant when the current observation configuration fails to preserve differences that later prove non-redundant for scientific performance. Deficit-triggered revision therefore does not merely update beliefs within a fixed observational scheme; it can alter which relations are observed, which variables are generated, which boundaries are imposed, which temporal resolutions are used, and which sensor or inferential couplings remain active. The Ziran System provides a distributed architecture for linking deficit detection, external memory, comparison across configurations, intervention, and feedback so that observational failure can modify the conditions that produced it. The paper distinguishes local correction from observation-relation revision: the former restores performance within an inherited configuration, whereas the latter changes the configuration through which scientific differences become observable in the first place. The broader claim is that autonomous science becomes genuinely self-revising only when failures can propagate back into the architecture of observation itself rather than remaining confined to parameter adjustment or theory selection.