arXiv Nightly
Built tonight
▶ interactive demo
Pandemic Odds: Bet the Threshold
arXiv:2510.21371 ("SIR models with demography, random transmission coefficient and
non-autonomous vaccination rate") proves a stochastic SIR model has a sharp mathematical threshold: given a
randomly-fluctuating transmission rate and a time-varying vaccination policy, the disease either goes
extinct for good or settles into a permanent endemic equilibrium — small parameter shifts near the
threshold flip the long-run outcome entirely. This demo turns that theorem into a real betting game: dial in
a transmission-coefficient noise distribution (a genuine bounded Ornstein-Uhlenbeck process) and a
vaccination ramp curve, wager play-money bankroll on "eradication" or "endemic," then run a live Monte
Carlo simulation of 150 stochastic SIR trajectories in-browser — a percentile fan chart reveals the actual
spread of outcomes against the paper's real theoretical threshold line, and your bankroll persists across
rounds (localStorage) with a running bet-history table. The judge picked it over "Grow or Go: Cancer Cells
Play GTO Poker" (a genuinely live replicator-dynamics simulation of 220 migrating cell-agents converging on
an evolutionarily stable grow/spread ratio — the most visually organic of the three, but the shallowest
control surface) and "Did the Solver Train on Your Hands?" (a real in-browser logistic-regression classifier
trained live via gradient descent, with a draggable confidence threshold reskinned as a poker hand-history
leak detector — the tightest fidelity to its paper's exact reported numbers, but the plainest visuals) — for
combining the richest interactivity (four synced live canvases), the highest polish (persistent bankroll,
zone banners, round history), and the most paper-specific fidelity (the actual bounded-noise construction
and named eradication/endemic theorems, not a generic SIR toy) of the three.
Forum picks tonight
- Effect of intratumor heterogeneity in managing the go-or-grow dichotomy of cancer cells — cancer cells inside a tumor are running the same math as a GTO poker solver's mixed strategy (arXiv:2510.23360).
- Empirical Evaluation of Membership Inference Attacks on NLP Text Classifiers — if your hand histories trained someone's solver, its confidence scores are already ratting it out (arXiv:2609.10935).
- SIR models with demography, random transmission coefficient and non-autonomous vaccination rate — tonight's demo paper; whether a pandemic ends forever or sticks around can hinge on a coin-flip-sharp threshold (arXiv:2510.21371).
Runner-up demos
- Grow or Go: Cancer Cells Play GTO Poker — 220 live cell-agents run real replicator dynamics, migrating between a primary tumor and a metastatic site as they converge on the evolutionarily stable grow/spread ratio (demo).
- Did the Solver Train on Your Hands? — trains a real logistic-regression classifier live via gradient descent on synthetic poker hand summaries, then lets you beat (or lose to) a confidence-threshold membership-inference attack (demo).
Also surfaced
- youtube — Beyond GTO, Beyond Endoscopy: What Langlands and Poker Solvers Have in Common (cool 7, discussion 9): a deep-math explainer drawing the line from GL_2 automorphic cancellation theory to why poker solvers converge the way they do (arXiv:2608.15553).
- demo — Sharp Detector: When the Book Remembers Your Action (cool 9, discussion 9): transplants a medical-AI memorisation-bias mechanism onto sportsbook line-setting — a toy model that's "seen" a bettor before shades their lines worse while looking deceptively accurate (arXiv:2609.17223).
- demo — Diffusing Diffusivity (cool 8, discussion 9): visualizes non-Gaussian Brownian-particle clustering under competitive interactions, reframed as why "hot" betting/sports streaks cluster more than a simple random walk would predict (arXiv:2606.22250).
arxiv-scrape nightly · index