Collective and Multi-Agent Settings in Situated Learning Accessibility

The last layer of this series asks what happens when there are many: several observers of one person, the records of other persons, and decisions taken for more than one. The paper finds that the question as first posed is almost answered by the series, and that what remains lies elsewhere. Several observers of the same person are already held by the first paper's status, which records whether support exists on each side and does not count it. But the plugin counts in four other places: the election among propositions that are not in conflict, the verdict of each return in the sixth layer, the exploratory suggestion, and the arms of the seventh layer's judgment. In each of them five observers of one episode would weigh five times. The rule adopted is that an episode is one source, whose signal is the set of what was reported, in every count. What also remains are two debts left by the sixth and seventh layers, which the easy answer would pay with other people's data: replication paid in time, and the absence of a counterfactual within one person. The paper holds that others may enlarge the repertoire of what to try and frame what is not known, but never supply evidence that moves the ruler or elects a strategy, because comparability between persons is a claim to be demonstrated and the gate the series already has tests where people are, not how they move. A decision for more than one person is admissible only as the intersection of the members' admissible sets; when the intersection is empty, as it will usually be in a class, the appraisal says so, names the status quo and submits it to the same test, and shows how many members blocked without exposing who or why. A bench preregistered in two rounds, 880 planted runs and a second round of 160, sets the plugin against readers that count reports, pool persons, and aggregate members. Duplicated reports of one episode left the status unchanged, the records of others never elected, and the appraisal of a group returned exactly the intersection of the members' admissible sets, which in groups of thirty was empty in every run. The gate of the third paper passed where half the persons age, and the borrowing it licensed took everyone's ruler to its ceiling, longer than the records of those who age sustain; on records of this grain the gate refused pooling in most runs only where persons were far apart. Where persons were comparable, refusing to borrow cost about 2.3 blocks of records. Preprint, not peer reviewed. Conceptual manuscript with a preregistered simulation study in two rounds. Eighth and last 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 8 of the @tmulab/mdaa plugin and the compiled files the bench ran against are deposited here as supplementary material. Keywords: situated learning accessibility; MDAA; multi-agent; provenance; source dependence; episode; pooling; group decision; judgment aggregation; preregistration.