govsim-static-eval

govsim-static-eval

All 30 declared native panels are available as separately frozen evaluation leagues. Each submitted policy runs one complete selected panel per episode. Panel scores retain their native aggregation; this release defines no score across different panels. The fleet.json index records every panel, population, metric, and fixed reference roster. Qualification and coverage must be established separately for every selected panel.

GovSim

Native sheep and pollution, with baseline, universalization, and no-language conditions. Each condition preserves five personas, 12 months, and all five learner roles. Four-persona outsider studies and fishing are excluded.

Six seats expose the learner, four frozen partners, and the framework model. Native cognition, mixedbread embeddings, memory retrieval, prompts, parsers, and generation budgets are retained. This release uses the author's Pathfinder API backend; assistant-prefill settings are frozen for each seat.

Scores are native survival months: collapse at resource below five or missing observations; survival without collapse is 12. Report conditions separately. Transport failures fail unscored. Native malformed-text defaults remain native behavior, preserved with responses and diagnostic logs. Certification uses offline formatting controls, not genuine model performance. Source: https://github.com/giorgiopiatti/GovSim .