Original grayscale color banding study

Use Bun 1.3.11 and sharp 0.34.5 in an isolated directory
Copy color-banding-study.ts build-color-banding-assets.ts verify-color-banding.py and this file there
Install the pinned sharp package from its normal package registry
bun add sharp@0.34.5
bun build-color-banding-assets.ts .
python3 verify-color-banding.py .

The builder writes uncompressed grayscale PNGs with sRGB rendering intent zero
The source range is sixteen to thirty two in byte normalized units across 1024 columns and 160 identical target rows
The source16 mode rounds the target multiplied by 257
The nearest mode rounds to eight bit values
The noise mode adds two independent fixed xorshift32 uniform draws as their difference before eight bit rounding with seed seventeen
The later upconversion multiplies already rounded eight bit values by 257
The inspection copies stretch each decoded original sample range sixteen to thirty two onto zero to 255 with clipping and nearest integer rounding
The WEBP cover is a presentation aid and is not used for measurements
The Python verifier reads saved PNG chunks checks CRCs decompresses IDAT and independently computes every expected sample and CSV count
PNG container hashes cover the authored writer outputs and source scripts
No AI generation video codec display calibration or unknown image restoration was performed