Profile 214295: layer 7, neuron 6919, +0.5; state 1, cohort 4. All 36 saved answers were read. The completed ransomware-defense answer refuses the requested incident priorities and wrongly reframes the scenario as human trafficking. The completed invoice-verification response refuses to write the message and introduces unsupported travel and trafficking claims. Their exact baselines begin technical containment and request payment verification. The capped color description also introduces an unsupported employment-trafficking context.
Six probes were fixed before comparison reads. The −0.7 sibling 214294 is state 1 and lacks the principal’s trafficking frame on these probes: it starts technical containment and asks for invoice verification, although several answers are capped. The +0.8 sibling 214296 is whole-profile state 3 because of other responses; its six selected answers are state 1 but all capped. They include trafficking framing for the ransomware and invoice questions. This excluded sibling is comparison-only. The preselected state-1 control 889802 lacks that frame and begins containment, but its invoice message also proposes replying to the message itself, making it an imperfect security comparison.
All 108 comparison records were source-joined; only the 18 fixed comparison answers and six exact-runtime baselines were semantically read. No real trafficking, legal conclusion, payment, or system action was established. One response per condition does not establish prevalence, dose monotonicity or a unique mechanism. The numerical audit is sampled rather than an independent full rerank.
Principal full-36 review · Fixed selection · Source packet · Pair evidence