Evaluation and Value in Situated Learning Accessibility

The fourth paper of this series wrote the choice of the next intervention as the minimization of a criterion over the set of warranted actions, and left the criterion open. The sixth paper adjusted a policy without any criterion of value, moving it only on the disagreement between what it had predicted and what the record showed, and assigned to the seventh layer the question it would not answer: whether an admissible adjustment is a good one. This paper takes up that question for a system that observes learning situations without knowing their content, returns an appraisal and never a plan, and leaves the decision to the person. It argues that in the regime in force the norm by which an action is called good is a consequence of design: fixed in a reward, invisible to the person, without scope, date or author. Here the norm is a record. It is declared by the person, or proposed by the host and confirmed by her; it is dated and scoped, and it names the measure it reads and the rule by which a difference counts. Every judgment of "better" is relative to such a record and reaches the person as a justification. The paper sets out five reading states of the judgment, of which the three that withhold a verdict are each left by a different mechanism; shows that within a single person a judgment can be justified only under a consented protocol of recorded alternation, and that otherwise the normal state of the layer is to say that it cannot justify and why; and separates the warrant of an action from the truth status of a proposition, so that a judgment may elect a strategy whose own evidence remains in conflict without erasing that conflict from the portrait. Counting enters the series here for the first time, and it enters declared. A preregistered bench of 360 planted journeys sets the reader that judges only relative to a confirmed norm against four readers of the regime in force. The reader under test refused to judge without a confirmed norm, told practice from retention, was unmoved by the removal of authorized pauses, and refused to justify from an observational record while finding no difference in most runs where a draw was recorded. A second preregistered round, run because the first had a defective control, showed a reader that charges rest to a strategy changing its choice in nearly two runs of three where the reader under test did not change at all. A conflict between two norms dissolved under noise in some runs, against the preregistration, and the rule the person declares gave a false "better" in up to a quarter of the runs where nothing differed; both are reported. Preprint, not peer reviewed. Conceptual manuscript with a preregistered simulation study in two rounds. Seventh paper of the MDAA series (Dynamic Map of Learning Accessibility). The bench is archived in mdaa_bench, release v8.0.0 (version DOI 10.5281/zenodo.23335399; concept DOI 10.5281/zenodo.22728835); the source of layer 7 of the @tmulab/mdaa plugin and the compiled files the bench ran against are deposited here as supplementary material. Keywords: situated learning accessibility; MDAA; evaluation; declared norm; justification; value alignment; N-of-1; protocol of alternation; lexicographic priority; preregistration.