The Germ of a Second Nature: Conditions for the Emergence of Primary Assemblers and Requirements on the Sensor–Actuator Endowment of an AI Verification independence as a condition of material self-improvement — a TRIZ-based analysis and a preregistrable test protocol (F1)

This article asks one narrow question. A distributed planetary technosphere — here called Digital Life (DL): a network of artificial agents able to perceive, act on and rebuild matter — would need a material germ, a minimal self-reproducing manufacturing core of the kind Drexler called an assembler. Under what structural conditions can such a germ exist, and what does it require of the sensors, actuators and verification arrangements of an artificial intelligence? The method is TRIZ (ideal final result, resolution of technical and physical contradictions, substance–field analysis, the laws of technical-system evolution, analogy) combined with functional analysis. The load-bearing result concerns verification. A material germ requires an agent that closes the loop perceive → predict → act → verify → reproduce, and the verify step is operative only if the verification channel carries information about the result beyond the agent’s own self-assessment — that is, only if its verdict is not inherited from the agent. This transfers the Non-Redesignable Core (NRC) condition from self-modifying software to matter. Three levels are kept apart: a conditional-information lemma, proved in full in Appendix A; a bridge to the NRC definitions, restated with explicit definitions and proved in Appendix A; and an empirical transfer to real systems, which is a hypothesis. Whether a particular real system satisfies the premise of the bridge — that its access to external quality passes only through its own outputs — is not tested here; it must be audited case by case, and a checklist for that audit is given. The transfer is made testable by F1, a two-phase protocol. Causally linked generational lines with and without an independent quality evaluation are compared on a hidden external quality functional Φ, first on a computational carrier and then on a material one; the statistical design, decision gates and decision rule are fixed in advance. Illustrative power calculations show the price of the design: with five lines per branch, only contrasts of about 1.75 standard deviations of the line-level statistic are detected with 80% power. Requirements on the sensor–actuator stack of a future germ are derived from the same condition and given indicative orders of magnitude. Wet-path priority, control by conditions rather than coordinates, the phase boundary as the true limit and controllable replication are stated as hypotheses F2–F5 with conditions of refutation. No test has yet been run: the article is a conceptual analysis and a preregistrable test protocol, not an empirical result.