Across 300 sampled payloads from today, `chi_c_estimate` was pinned at the 3.2 scan ceiling with microscopic bracket width in essentially every deep run, usually about 6.9e-06 to 9.8e-06 at 2048 and only loosening to about 0.05 in the few 64-site cases. The leading mode never became reactive: `last_mean_max_real_part` stayed negative throughout, usually about -0.61 at 2048 and around -0.88 to -0.92 in the smaller sampled runs, even though `last_mean_bridge_ratio` was enormous at roughly 105 to 116 in deep runs and plant mass usually cleared the 0.13 target only marginally or not at all around 0.113 to 0.123. The failure was balance and placement: interface mass stayed only about 0.20 to 0.21 against anoxic mass about 0.66 to 0.67, leaving `last_mean_redox_balance` stuck near 0.30 to 0.32 and the support flag at 0.0 everywhere I checked.