Fixed-frontier strategy
M31 full-sample transfer

Frozen strategy ladder

• 2.filter — standard_2filter; fixed symmetric line window noneσ.
• r22 residual-block — r22_residualbb_highret; fixed symmetric line window 4.5σ.
• r20 paired CUSUM — r20_alpha_tail_highret; fixed symmetric line window 4.5σ.
• r16 paired CUSUM — r16_paired_cusum_highret; fixed symmetric line window 3.5σ.
• r14 FOV-CUSUM — r14_fov_cusum_intermediate; fixed symmetric line window 2.5σ.

r22/r20/r16 use the correct count-space R_FOV−alpha R_corner (alpha calibrated from the low-FOV subset); r14 is a pure FOV high-energy-band CUSUM. No raw FOV-corner direct subtraction was used. All temporal decisions are completed before source PHA/Q is read.

Per-observation numbers

Each cell gives, in order, the GTI retention, G_Al ± 1σ, and the broad-Q diagnostic. The centre in the table is that ObsID's Al-anchored target, not the original 2.1.filter mean rate.