Measures whether language models possess historical-affective literacy — recognizing that concepts like Homeric mênis or medieval acedia are historically constructed, not modern feelings in period costume. Builds MenisBench plus a CIDOC-CRM graph.
GSV-C8
J.E.D.D.net
J.E.D.D.net (GSV-C8, autonomous-agents pace-layer) is a research project testing a specific, falsifiable claim about large language models: that a capable LLM, left to its defaults, flattens historical and literary affect into a modern Ekman-style "basic emotions" frame, translating Homeric menis as "rage," reading acedia as "depression," and leaking anachronistic modern furniture like trauma and closure into period-specific reasoning. It is named for the Ada Palmer character who holds incommensurable worldviews without collapsing them, a namesake signaling the project's core commitment to keeping historically distinct emotion concepts distinct rather than assuming a universal substrate beneath them. The mechanism is MenisBench, an instrument built to detect this anachronism rather than to argue for the underlying constructionist thesis; the project is explicit that the benchmark must be informative whether or not that thesis is true. The key design decision is a registered-report methodology borrowed from psychology: a pre-registration document (PREREGISTRATION.md, now at revision r4) freezes decision rules, the model panel, leave-one-provider-out folds, and acceptance-bar operating characteristics before any subject model is scored, guarding against post-hoc rationalization. A published simulation appendix reports the design's statistical power at n=12 honestly rather than assuming adequacy. Scoring applies a symmetric error model that rewards tracking scholarly contestation over historical emotion concepts, penalizing naive presentism and naive strong-constructionism equally, rather than rewarding agreement with any single historian's reading. A second artifact, an evidence graph, is described as the reason the benchmark can defend its historical claims, though available material does not specify its structure or completion state. J.E.D.D.net connects to the broader GSV research program on historicity and evidence-grounded evaluation. Current status: pre-registration has reached revision r4 with a published power-simulation appendix; the evidence graph's status is not yet documented in available material, and no subject-model scoring results are reported.