08AUDIT / TRUST IS AN INTERFACE
Every query, teacher and update leaves evidence. Inspect data origins, model changes and the conditions for stopping.
Empirical aggregates in the revised manuscript, traceable to tables and pages.
Local virtual experiments with visible seeds, parameters and round snapshots.
Procedural illustrations of mechanisms, not recordings or measurements.
Future tasks, study designs and theoretical extensions requiring validation.
This website has no device connection, account or raw neural upload. Task observations must not be repurposed for unauthorized health, political, religious or personality inference.
Run the loop to inspect computed propensities, teacher risk and audit branches here.
Kernel matching identifies equivalence classes of task-relevant structure, not unique neural codes or private thoughts.
Rank- and energy-matched interventions are needed to test whether aligned directions enter the decision pathway.
Shared structure and personal preferences must remain distinct; a population mean cannot replace an individual's judgment.
Teacher weight may be zero when conditional innovation is absent, drift is excessive, or evidence is insufficient.