arXiv Nightly

arXiv Ideation Β· 2026-09-26 Β· 159 papers fetched (export API 406'd all night on a URL-encoding quirk β€” fixed the scraper mid-run, then hit real IP rate-limiting anyway, so the HTML-scraping fallback carried the night), 30 sampled for ideation, 23 ideas, 3-way build-off

Built tonight

β–Ά interactive demo
Same Skill, Wild Careers
arXiv:2609.25838 ("Mutation Order and Selection Shape Intratumor Heterogeneity in Tumor Evolution") models cancer cells with a branching process where transformation needs two driver mutations β€” and finds that cells with identical selection advantage still end up wildly different clone sizes, purely because of the order the two mutations arrived in. This demo ports that straight onto the oldest poker/sports argument there is: is he actually good, or did he just run good? Spin up 300–1000 "careers" that all share the exact same skill level and the exact same pool of variance events (bad beats, hero calls, lucky bracket seeds) β€” only the order each career experiences them in differs. Drag the selection-strength slider to 100% and every career collapses to one identical number (verified: 1.00Γ— ratio, zero spread). Drag it toward 0% and the same identical-skill pool explodes into a leaderboard spanning 20Γ—+ β€” with a poker-nicknamed "guy" jumping from #3 to #491 on a reshuffle, skill held perfectly constant. The judge picked it over "Tout Field" (a mesmerizing force-directed network sim where a seeded hot take survives a truth-correction field until a threshold, then snaps β€” gorgeous, but the payoff unfolds slowly) and "Instanton to Bust" (the most mathematically rigorous of the three β€” a real CramΓ©r/Legendre-transform derivation of the most-probable path to bankroll ruin, cross-checked live against a Monte Carlo swarm β€” but the driest visually) β€” for the single cleanest, most instantly legible "whoa" moment plus dead-center SharpLab fit.

Forum picks tonight

Runner-up demos

Also surfaced

arxiv-scrape nightly Β· index