nx_slopmeter_gate.nx

nx_slopmeter_gate.nx -- REFEREE for nx_slopmeter_lib (output repetition measurement). THE HARD PART IS PROVING A METER WITHOUT INVENTING A THRESHOLD. Any cutoff I picked here would be a magic number wearing a gate's authority. So the teeth are ORDERING and IDENTITY properties, which a broken meter cannot satisfy and which need no constant: [T1] u2605BITE: fires on looped text, silent on varied text -- the whole claim, stated once. [T2] MONOTONE: looped scores STRICTLY above varied. A meter that ranks them equal is useless even if both numbers look plausible. [T3] floor: text with no repeated 3-gram scores exactly 0 -- not "small", 0. [T4] ceiling: text that is one phrase looped scores the full scale. Both ends pinned means the scale is real and not a squashed middle. [T5] u2605the 2-gram trap: "she said ... she said" repeats a BIGRAM but no trigram. A 2-gram meter would condemn ordinary prose; this must read 0 and prove the width choice is load-bearing. [T6] tokenizer counts words, not bytes, and treats an apostrophe as interior. [T7] u2605NEG-CONTROL: text too short to contain a window returns SM_UNMEASURABLE, never 0 -- "no measurement was possible" must not be averaged in as "clean". [T8] distinct-vocabulary is an INDEPENDENT axis: all-unique text is 1000 permil, one word repeated is a small fraction of it. [T9] case-insensitivity: "The Map" and "the map" are the same phrase repeated. Hermetic: in-memory fixtures, no filesystem, no engine, no network. Sovereign x86_64. license_tier: ORIGINAL expect_exit: 0