Testbed-calibration traps in small-scale emergence science: two pre-registered negatives

Code and result logs for two pre-registered emergence negatives and the calibration diagnostics that prevent them. Code is MIT; result logs CC-BY-4.0. Figure/manuscript sources not included; large standard datasets not bundled. v1.2 adds the gamma-difficulty ladder, detector validation on real induction emergences, and scan-statistics analysis code and logs reported in the submitted manuscript. v1.3 adds the 2026-07-05 bolstering campaign reported in the submitted manuscript: gamma-ladder rungs 0.5/0.7, the 30-seed gamma=0.9 calibration set, and the M-axis (M=401/801) and N-axis (eval set 16/64) cells including the gamma=0 late emergence at four times the budget. v1.4 adds the red-team-driven validations reported in the submitted manuscript: the sixty-seed gamma=0.9 calibration set, the M=801 and horizon-axis cells, the gamma=0 late-emergence set at four times the budget, the gradual-emergence latency control, and the BIG-Bench scan-statistic audit (script and adjudication table). v1.4.1 replaces four gradual-emergence logs that v1.4 had archived mid-training with their completed versions (lr 1.5e-4 seeds 1-4; all ten reduced-lr seeds emerge, onsets 6525-6850, matching the manuscript); all other content identical to v1.4. v1.5 adds the gamma=0.5 budget grid (60 logs), archives the 26-cell rescue-control raw data, and includes the rescue-statistics builder script.