Remove the pilot_full, pilot_full_rerun and pilot_full_relaunch splits and eval_results__pilot_full from the head: gathered as control_set (the rows each launch requested at 50d26c4c, the reference rows at a251840f, the evaluation in eval_results__control_set at ceba09db); they remain in the history up to ceba09db
control_set: the rows each launch requested (pilot_full@54bf0be, pilot_full_rerun@47abca9, pilot_full_relaunch@e6677a1; figure-JSON columns, dropped since 30 Oct 2026, left out)
Rebuild pilot_full with the PDF-backed figure question (does the figure present results of the paper's own study? paper PDF + caption + image, google/gemini-3.7-flash, 3 tries, majority): lasapp Fig. 2 and pmsat Fig. 2(a)/(b) left as worked examples; tables unchanged. Answers replayed from a paid run of the identical request.
pilot_full_relaunch: rows of crossover and pmsat copied unchanged from pilot_full@133755b9, to rerun the two papers whose new figures (crossover Fig. 3, pmsat Fig. 2(b)) no run was asked for