UFP-NB1 v0.3.0 Candidate: Predictive Compression and Compiled Observer Cost in Finite Relational Graph Dynamics

UFP-NB1 v0.3.0 Candidate synthesizes the executable projection-closure laboratory introduced in v0.2.0 with three subsequent frozen development increments on predictive memory, compiled observer cost, microscopic-rule sensitivity, and exact N=7 scaling. The work studies a finite relational graph laboratory and asks how much information a coarse observer must retain before its next degree-histogram observation is determined. The principal result is negative-leaning: under NB1-R0, the exact predictive partition remains close to the full unlabelled graph description through N=7, with 153 of 156 invariant classes retained at N=6 and 1,018 of 1,044 retained at N=7. Short clocked observation histories can initialize a graph-blind compiled predictive register, but startup tables can be large and steady-state compression remains small. A predeclared walk-3 parity comparison rule changes the exact predictive cost profile modestly without producing a cheap full-space law. The work therefore separates three costs that should not be conflated: the compile-time world model, the startup history-to-Q initializer, and the steady predictive register. The central finite conclusion is: In this laboratory, predictively sufficient degree-histogram descriptions of NB1-R0 remain almost as expensive as retaining the invariant graph itself through N=7. Memory can initialize predictive state, but exact predictive sufficiency remains costly. This release is a synthesis of preserved finite mathematical and software evidence. No new scientific experiment is introduced by the v0.3.0 wrapper. Earlier failed projections, ambiguous histories, controls, frozen contracts, replay evidence, and custody qualifications remain preserved in the accompanying package. This work does not establish that nature is a graph and does not recover quantum mechanics, general relativity, gauge fields, particles, physical time, thermodynamics, dark matter, or any U1-U10 / N1-N8 recovery gate. No asymptotic or continuum law is inferred from N=4 through N=7. H1/LZ230616 remains HISTORICAL EXPOSED and is not used as a prospective validation benchmark. Status: Research candidate / finite predictive-compression studyVersion: v0.3.0Previous version: UFP-NB1 v0.2.0Previous-version DOI: 10.5281/zenodo.22728166Publication authority: Michael Wayne Hughes