nx_rigfloor.nx

nx_rigfloor.nx -- DERIVE THE ASSET FLOOR FROM THE REFERENCE CORPUS INSTEAD OF DECLARING IT. WHY THIS EXISTS. nx_asset_floor_gate is THE ruler for "is this character game-ready", and it carried five HAND-PICKED bounds: AF_VERT_MIN 4000, AF_TRI_MIN 8000, AF_TRI_MAX 120000, AF_JOINT_MIN 40, AF_MIDLINE 3000. A picked bound structurally cannot answer the question the operator actually asks -- can we emit something that MATCHES THE MODELS WE HAVE -- because it is not a function of the models we have. This organ makes the bound a MEASUREMENT of the reference population, so the gate stops ASSERTING a floor and starts REPORTING one. THIS IS NOT A SECOND RULER. nx_asset_floor_gate remains the one ruler; this is the PRODUCER that derives its bounds into a conf the gate reads. Producer and consumer cannot disagree because there is exactly one of each. nx_spendgate ranked nx_asset_floor_gate as the top incumbent for this intent -- extending it is the entire point, and a rival floor organ would be the duplicate-ruler defect. ROSTER-DRIVEN, NOT DIRECTORY-WALKING, ON PURPOSE. sys_getdents64 exists and a bounded walker is proven in nx_wiki_content_loader. It is still the wrong primitive HERE: the corpus IS the thing under measurement, so a silently-truncated listing would not merely lose a file, it would corrupt the derived bound itself and then publish that bound as a fact. A roster is complete by construction and its coverage is DECIDABLE -- rows declared vs rows read, both printed, and if they do not sum then no floor is written at all. METHOD: OBSERVED RANGE OVER THE FULL POPULATION. No percentile index is chosen, because choosing one would reintroduce exactly the picked constant this organ exists to remove. The band is [min, max] of what the reference assets actually are. WHY n>=2 IS NOT A MAGIC NUMBER. A range needs two observations to HAVE WIDTH. At n=1 the band is a point and would refuse every asset except that one file -- a bar that is not a bar. So the organ abstains at n<2. That is the arity of the operation, not a threshold anyone selected. EXITS: 0 OK | 1 OUT-OF-BAND | 2 usage | 3 UNOBSERVABLE (no corpus / degenerate / partial read) A partial roster read NEVER degrades to a floor: it abstains, because a floor derived from an unknown fraction of the corpus is worse than no floor -- it looks authoritative and is not. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0