# §6E-D.6(c) — same-fixture x264-medium reference for §6E-A6.5 gate.
#
# Encoded with:
# ffmpeg -y -f rawvideo -pix_fmt yuv420p -s 1920x1072 -r 30 \
# -i /tmp/img4138_1080p_f10.yuv \
# -c:v libx264 -preset medium -crf 23 \
# -x264-params bframes=1:b-pyramid=0:keyint=5:min-keyint=5:scenecut=0:no-deblock=0
#
# Source fixture: SAME 10 frames of IMG_4138 the gate test runs phasm
# against. This is apples-to-apples: any partition-shape / direction
# gap vs phasm reflects encoder behaviour differences only, not
# fixture-content differences.
#
# Histogram extracted via core/scripts/h264_mb_partition_histogram.py
# (PyAV side-data path). Note the libavcodec MV-exporter labels
# Skip/Direct B-MBs with their spatial-direct synthesized MV
# directions, so direction shares partly reflect spatial-direct's
# behaviour, not encoder mode-decision (see
# memory/h264_phase6e_d_trace_breakthrough.md).
#
# Replaces tests/data/x264_reference_mb_histogram.txt (271-frame
# mixed-content batch) — kept only for historical context.
# H.264 mb-partition histogram
# source: /tmp/x264_medium_img4138_f10.h264
# resolution: 1920x1072 (8040 MBs/frame)
## B-frame (4 frames, 32160 MBs)
no-MV (Skip / Direct / Intra) : 579 ( 1.80%)
emitted MVs : 31581 (98.20%)
partition shapes (over 45658 MV records):
16x16 : 43944 ( 96.25%)
8x16 : 755 ( 1.65%)
16x8 : 557 ( 1.22%)
8x8 : 402 ( 0.88%)
direction (over 45658 MV-anchors):
L1 : 25861 ( 56.64%)
L0 : 19295 ( 42.26%)
Bi : 502 ( 1.10%)
## I-frame (2 frames, 16080 MBs)
no-MV (Skip / Direct / Intra) : 16080 (100.00%)
emitted MVs : 0 ( 0.00%)
## P-frame (4 frames, 32160 MBs)
no-MV (Skip / Direct / Intra) : 2477 ( 7.70%)
emitted MVs : 29683 (92.30%)
partition shapes (over 37567 MV records):
16x16 : 26730 ( 71.15%)
8x8 : 5897 ( 15.70%)
8x16 : 2562 ( 6.82%)
16x8 : 2378 ( 6.33%)
direction (over 37567 MV-anchors):
L0 : 37567 (100.00%)