nx_vcodec_bframe_ceiling.nx

nx_vcodec_bframe_ceiling.nx -- measure the B-FRAME ceiling, decoder-transparent. B-frames (bidirectional prediction) are x264's single biggest lever and the residual decomposition points here: our P-frames are ~5-9x behind x264 while intra is only 2.6x, and sub-pel (6-tap re-audit) is dead -> the gap is STRUCTURAL prediction. A B-frame predicts each block as the AVERAGE of a best-match in a PAST ref and a best-match in a FUTURE ref: averaging cancels noise, and (the big win) reveal/occlusion regions that one direction cannot predict are covered by the other. This probe, on the aptly named bframe_test, integer-searches each interior 16x16 block against both the previous AND next CLEAN frame, and compares the residual of uni-directional (past-only, what our codec does today) vs bi-directional (past+future avg). A large bi<uni cut = B-frames are THE lever and the (pervasive, VOD-only) enc+dec+reorder build is justified; a small cut = look elsewhere (multi-ref, partitions). Pure measurement on clean originals => no bitstream, no parity risk. license_tier: ORIGINAL