nx_perceptual_judge.nx

nx_perceptual_judge.nx -- signal-vs-profile honesty judge. module: nishi-core.perception.judge depends: nishi-core.perception.profile + nishi-core.io.syscalls disk_kb: 4 capability: PERCEPTION wired_status: PARTIAL_WIRED MISSING_CAPABILITIES: - JUDGE_BEHAVIORAL_VALIDATION_DATASET (needs nx_perceptual_dataset entries from Customer Zero behavioral A/B per [[NISHI_DOG_PERCEPTUAL_TEST_HARDWARE]]) - JUDGE_PER_AXIS_BREAKDOWN (current stub returns coarse score only; per-axis frequency/spatial/dynamic/periodicity/reverb decomposition queued) license_tier: PUBLIC_NISHI_SUBSTRATE genealogy_id: feedback-multi-species-perceptual-substrate-not-anthropocentric_2026 + feedback-honest-perf-verdict-no-aspirational-claims + feedback-no-false-ok-substrate-honesty-audit Per cardinal [[feedback-multi-species-perceptual-substrate-not-anthropocentric]]: Produces a MEASURABLE verdict score (0..100) of how likely a target species' perceptual system would classify a given signal as "real prey / real conspecific / real environmental" vs "artificial rendered media." Substrate uses this to gate honest claims of "this codec is dog-real" without anecdote. THIS IS A PARTIAL_WIRED STUB. Implementation requires behavioral validation dataset; signatures + per-axis decomposition shape + honesty gate are wired so downstream callers can compose against the stable interface while the model trains.