egsim confirmatory materials: preregistered 6x2 factorial data, frozen analysis and code for "Organizational Independence of Oversight Roles in Communication-Delayed Autonomous Missions"

Confirmatory materials for the preregistered study registered at osf.io/y5wc6 (2026-09-02, public, no embargo). This is the confirmatory version of the record whose previous version holds the pilot materials (10.5281/zenodo.22267880). Design. A 6 x 2 factorial simulation of Mars surface operations under dust-storm hazard, paired within environment seed: six oversight configurations (MONO, NEGOT, CREW-SUB, CREW-PLACEBO, CREW-IND, OCL-EXT) crossed with one-way communication delay of 0 or 24 minutes, 25 seeds (53-77) disjoint from the pilot. One self-hosted open-weights model at temperature 0, fixed by checkpoint hash, with an identical per-episode token budget across configurations. Contents. 270 raw decision-episode cells with full provenance; the environment, runners and role prompts; the analysis script frozen before the confirmatory data existed; the dataset-gate output; per-cell logs; the preregistration text; and the Methods and Results of the empirical paper. reproduce.sh re-runs the analysis against the deposited cells and diffs it against the recorded result. Integrity as delivered. Zero parse failures (the registration's hard stop), one model checkpoint across every cell, no duplicated cell, and no delay pair split across machines. Five cells superseded by a mid-run node reassignment are shipped in a quarantine directory outside the analysed paths rather than deleted. Stated limitations. The equivalence leg of H1 is undecided rather than refuted - certifying equivalence at the registered margin would need roughly 95 seeds instead of 25 - and 100 of the 270 cells carry no code-revision stamp because the boards that produced the descriptive arms received the harness by file copy rather than a git checkout. Both are documented in the archive README.