nx_arousal_profile.nx

nx_arousal_profile.nx -- estimate a dual-control profile from OBSERVED sessions, and refuse the part of it that the data cannot support. This closes the loop the lane was missing: nx_arousal_fit recovers tau from a trajectory and nx_arousal_clinical diagnoses from a profile, but nothing turned OBSERVATIONS into a PROFILE. A therapy app watches sessions; nobody hands it SES/SIS1/SIS2. ★★THE IDENTIFIABILITY BOUNDARY IS THE WHOLE POINT. The drive law is drive = (SES*stim)/1000 - (SIS1*stim)/1000 - SIS2 = ((SES - SIS1) * stim)/1000 - SIS2 SES and SIS1 BOTH scale with stimulus, so stimulus-response data can only ever recover their DIFFERENCE. SIS2 is the intercept and separates cleanly. No amount of extra sessions fixes this -- it is structural, not a sampling problem. Separating SES from SIS1 requires the SIS/SES questionnaire, a different instrument measuring a different thing. ⇒ this organ reports net = SES-SIS1 and SIS2, and reports SES/SIS1 as UNIDENTIFIABLE. An estimator that returned three numbers here would be inventing one of them. ★SATURATED OBSERVATIONS CARRY NO SLOPE. drive is clamped to [0,1000]; a session that pinned at either end tells you the drive was "at least" or "at most" something, not what it was, so fitting through it manufactures a slope from a clamp. Those points are refused, not down-weighted. usage: nx_arousal_profile fit <stim1> <drive1> <stim2> <drive2> nx_arousal_profile selftest license_tier: ORIGINAL expect_exit: 0