CompressDog

Corrections

Every mistake I have found and fixed on this site, with the date it was found and the article it affected. Nothing here is removed quietly.

2026-10-28

The preset captions on the compression page said the tuned tables were "about 3% smaller" and "5-10% smaller". Both figures came from six images.

Re-ran on 320 images. The tuned tables are 8.3% *larger* at quality 90 and score lower. The captions were rewritten from the new data. the article that documents it

2026-11-02

The metadata scan reported 56 files with GPS coordinates. That was a byte-sequence search, not a real EXIF parse.

Cross-checked with an independent parser: three files had a GPS block, and the coordinates in all three were 0,0 or NaN. The article now says zero usable coordinates. the metadata article

2026-11-02

The same scan reported metadata at 3.57% of file size on average, and 90.7% at most. The segment walker was not skipping the 0xFF padding bytes, so it overstated every size.

Fixed the walker and reported the median alongside the mean: 568 bytes, 0.21%. The mean was being dragged up by a handful of huge APP1 blocks, which is now stated explicitly. the metadata article

2026-11-03

The PNG lossless page said the pixels come out identical, without qualification. Eight of 261 files did not.

Checked all eight: the only bytes that changed were colour values underneath fully transparent pixels. Composited over white, every one was byte-identical. The article now separates "pixels identical" from "bytes identical". the PNG article

2026-11-06

Converting a PNG to JPG in this tool produced a PNG. The worker chose its code path from the source format and never looked at the target, and the result message did not report the output format at all, so the filename stayed .png.

Moved the three format decisions into a tested module. The PNG path now requires the target to be PNG as well, and every success message reports the real output format. the format converter

2026-11-08

The dataset page was about to describe six summary files as "raw per-file results". They were aggregates, about 1 KB each, and the word "raw" would have been false.

Re-ran the measurements to produce real per-file rows, and the page now labels each dataset as either per-file records or aggregate only. the datasets page

This page exists because a site that never corrects anything is either perfect or not looking. I am not perfect, so the second is the honest reading. The date is the day I found the problem, not the day I published the article.