• Use a three-step checklist to inspect blank predictions, sample-size uncertainty, and sensitive-task routing before publishing model rankings.
• Download the companion blank_prediction_audit.py script from the resource page.

• Automatic thresholding could be destroyed by a few non-stain pixels. A red scale bar drawn by a slide viewer deconvolves to a DAB optical density far above any real tissue, and the histogram was scaled by that maximum, squeezing the real signal into the lowest bins. On 14 real COX2 and iNOS slides every one of them read as no expression, with no warning shown. The histogram now uses the fixed ImageJ 8-bit transmittance scale, so outliers cannot stretch it
• Results from before this fix are not comparable with results after it. The algorithm version moved from 2 to 3 and is recorded in the CSV export. If you still have the original images, please re-run them
• The Triangle threshold method is unreliable on screenshots taken from slide viewers. It is left mathematically unchanged because it is a faithful port, but a warning now appears when an automatic threshold lands at the edge of the gray range. Otsu and IsoData are the safer choices for these images
• The IHC analyzer accepts several images at once. A thumbnail strip lets you switch between them and tune the threshold on each, while stain method and scoring settings stay shared so images remain comparable
• New warnings flag images whose tissue contains an unusual share of very dark pixels, or whose staining range is too narrow for automatic thresholding to be trusted
• Thumbnails in both IHC tools were blank because the content security policy blocked the image source they used. They now render correctly

• Automatic thresholding rebuilt on the fixed ImageJ 8-bit transmittance scale; batch results from earlier versions are not comparable and the algorithm version is recorded in the CSV export
• Warnings for degenerate histograms and dark pixel outliers now reach the results table, and the warning banner states how many images are affected instead of only listing the first three
• CSV export gained histogram domain, threshold degenerate, low contrast, and darkest bin fraction columns
• Thumbnails in the upload list render again after switching away from an image source the content security policy blocked

• Automatic thresholding rebuilt on the fixed ImageJ 8-bit transmittance scale, so a red scale bar or other non-stain pixels can no longer stretch the histogram and push every result to zero
• Results are not comparable with earlier versions; the algorithm version is now 3 and appears in the CSV export
• Load several images at once and switch between them with the thumbnail strip; threshold and reference white are kept per image while stain and scoring settings stay shared
• New warnings for degenerate histograms, unusually dark pixel fractions, and automatic thresholds that land at the edge of the gray range
• Thumbnails render again after switching away from an image source the content security policy blocked

Osteoarthritis is more than wear and tear. An Advanced Science study in mice and human synovium finds that HMGCR, the statin-targeted enzyme of cholesterol synthesis, runs too hot in synovial fibroblasts, overproduces cholesterol, drives cellular senescence, and fuels synovitis and cartilage breakdown; local joint knockdown of HMGCR in mice reverses senescence and eases pain and cartilage loss.