arXiv Nightly
Built tonight
▶ interactive demo
The Cyclic Judgment Trap
arXiv:2609.23977 ("Binding-Motivated Contextuality: A Cross-Domain Cyclic Test in
Perception and Judgment") designs quantum-foundations-style cyclic tests (Bell/Leggett-Garg/KCBS-family
noncontextuality inequalities) for human perception and judgment — the paper's own instrument is
preregistered but unrun, so this demo runs an honest, independent version of the same idea rather than
reproducing results that don't exist yet. You get 8 conditional-probability questions about a same-game
parlay ("Team A covers AND the under hits AND the star scores 25+"), asked in a shuffled order every time.
If your beliefs are classically coherent, chaining Bayes' rule forward and backward around the 4-prop
cycle must telescope to the same number — most people's answers don't, and the demo scores exactly how
much they don't, then hands you a shareable "contextuality score" (verified: perfectly consistent answers
score ~1%, deliberately contradictory ones score ~99%, realistic noisy answers land around 45%). The judge
picked it over E-Power Play (a mathematically flawless race between an anytime-valid e-process and a
fixed-sample Neyman-Pearson test, watching one hit "PROVEN" hundreds of bets before the other resolves) and
Stale Line Arbitrage (a real D3 force-network sim with a hand-rolled Jacobi eigenvalue solver
computing live algebraic connectivity — deep, but a stray CSS overlap bug clips the panel heading on load)
— for being bug-free, the most shareable ("I scored 92%, I'm a sportsbook's dream customer"), and still
honestly grounded in the paper's real cyclic-inequality math rather than a generic reskin.
Forum picks tonight
- The Type-II Error of Test Supermartingales: e-Power versus the Chernoff-Stein Exponent — how many bets until you can PROVE your handicapper has an edge, without picking your sample size in advance — and the real rate penalty you pay for that flexibility (arXiv:2609.27765).
- Resilient Monitoring of Social Dynamical Systems through Collaborative Multi-Agent Networks under Latency — sportsbooks as a resilient-consensus network with bad latency guarantees; the paper gives the formula for how long your line-shopping arbitrage window stays open (arXiv:2609.27902).
- Binding-Motivated Contextuality: A Cross-Domain Cyclic Test in Perception and Judgment — tonight's demo paper; your own SGP judgments are provably order-dependent, which is exactly the soft spot parlay pricing exploits (arXiv:2609.23977).
Runner-up demos
- E-Power Play — an anytime-valid test supermartingale races a fixed-sample Neyman-Pearson test to "prove" a betting edge, with sliders for true effect size and null parameters (demo).
- Stale Line Arbitrage — a 12-node D3 sportsbook network converging on a fair price under latency and adversarial noise, with a live "arbitrage window open" badge (demo).
Also surfaced
- project — Prop Aggregation Audit for nba-props-terminal (discussion 8): tree-based forecasters (GBMs, random forests) that look great at the leaf level can systematically collapse into biased, miscalibrated forecasts once you aggregate up to the level a prop actually settles at (arXiv:2609.27912).
- project — ChronosAttack Red-Team Harness for Sage/Sentinel (discussion 8): an LLM agent can be steered into wrong decisions purely by adversarially reordering or delaying tool calls, with every individual tool output left untouched (arXiv:2609.27857).
- demo — Rebound Roulette (discussion 8): Euler-integrates a 5-compartment within-host HIV ODE day by day — same viral-dynamics math, repurposed as a rebound-timing gambling toy (arXiv:2609.29972).
arxiv-scrape nightly · index