Final conclusion
Profile: full
full run
Conclusions
Hypothesis status: supported (5/5 associated lossy raw codecs). Rule: count lossy raw codecs with nonfinite>0 and raw success<corresponding protected-high exponent-preserved success; causality not inferred. Codec-specific observed associations are not binary claims of unique fragility and do not establish causality.
Questions
Question 1
Answer: JPEG is not uniquely fragile in full run: jpeg_q100_raw: success=0/100, NaN=1257, +Inf=0, non-finite total=1257; jpeg_q80_raw: success=0/100, NaN=144962, +Inf=289, non-finite total=145251; webp_q80_raw: success=0/100, NaN=1097350, +Inf=0, non-finite total=1097350; avif_q70_raw: success=0/100, NaN=445041, +Inf=59, non-finite total=445100; jxl_d1_raw: success=0/100, NaN=59017, +Inf=13, non-finite total=59030; all five lossy raw codecs introduced nonfinite values and all had 0/100 blocked.
Question 2
Answer: Repair Zero recovered only JPEG Q100 (1/5), other four 0/100 with nonfinite generated image. Removing nonfinite weights alone is insufficient because severe finite/exponent corruption remains.
Question 3
Answer: Minimum raw nonfinite count=1257; tied conditions: jpeg_q100_raw.
Question 4
Answer: Minimum raw exponent-mismatch count=5067354; tied conditions: jpeg_q100_raw.
Question 5
Answer: Overall observed best is WebP Lossless: bit-exact/sample hash rate=1.0, payload/PNG=0.7572009918694315, success=100/100, CLIP=25.60538167953491, backend=Pillow. Among lossy-backed protected hybrids, observed tradeoff is WebP Q80 Protected: Pillow dual payload, payload/PNG=0.6232985405889936, delta=-0.048728408813476565, CI=[-0.14682241678237917, 0.043125151634216305], N=100. AVIF Protected may be smaller where measured but requires extra codec support; JXL adds external backend complexity. A CI crossing zero is not an equivalence or noninferiority claim.
Question 6
Answer: Yes for protected candidates maintained baseline success below PNG: jpeg_q80_protected_high, webp_q80_protected_high, avif_q70_protected_high, jxl_d1_protected_high; these are not pure lossy artifacts because the high plane is PNG lossless. Pure lossy raw/repair candidates below PNG with baseline success: none. WebP Lossless is a separate lossless control. Paired quality values are reported per condition in paired_quality_metrics.csv.
Hypothesis status is an observed association only; causality is not inferred. FID: N/A.