• Chart appearance panel: switch journal presets (Nature / PLOS ONE / Cell / eLife) and live-edit point size, line colour, CI band opacity, and residual colours
• Download the regression fit and residual plots as PNG (96 / 150 / 300 dpi) or vector SVG on a white publication background

Anthropic, the UK AI Safety Institute, and the Alan Turing Institute have demonstrated that inserting just 250 carefully crafted documents into training data can plant a permanent backdoor in LLMs ranging from 600M to 13B parameters. Bigger models and more data do not dilute the poison.

• Added an environment-gated Google AdSense loader. No ad script is emitted unless PUBLIC_GOOGLE_ADSENSE_CLIENT is configured
• Added a reusable ContentAdSlot component and one blog-post ad placement after the article body, before support CTAs and related articles
• Added /ads.txt endpoint driven by PUBLIC_GOOGLE_ADSENSE_PUBLISHER_ID for AdSense ownership verification
• Updated CSP allowlists for AdSense script, connection, and ad frame origins without relaxing the global security policy
• Updated English and Traditional Chinese privacy policy pages to disclose optional AdSense advertising and third-party cookie behavior

• Site-wide "Academy Alchemist" grimoire redesign — brass-gold palette, deep-wood and warm-parchment surfaces, and Cinzel + Lora serif typography, consistent across dark and light themes
• Cards now carry classical double-rule frames with a warm brass glow on hover; primary actions use solid gold buttons
• Illuminated drop-cap capital opens blog articles and spellbook entries

• Stroop colour-word test: congruent / incongruent / neutral trials with reaction-time timing
• On-screen colour buttons plus D/F/J/K keyboard input; controlled neutral fullscreen surface and 3-2-1 countdown
• Editable protocol parameters with named templates (built-in telomere-study template); export results as CSV, no data leaves the browser