Riptide Research

Overview

Every research tool ships a single number where the real answer is a distribution. Riptide takes the live options chain, derives the market's full risk-neutral distribution, and puts my own Bull/Base/Bear view on the same axis, shading the gap between them as expected value and a half-Kelly size. On top of that sits an agentic layer: a Model Lab to author and save my own models, an Edge Radar that scans the whole universe for names where a model's distribution most disagrees with what options are pricing, and a Model Arena that overlays the market, my view, the Street, and an AI analyst on one axis and then grades which of them has actually been right. The quant engine is dependency-free TypeScript with unit tests; the AI analyst is anchored to the implied base rate so it can't free-run overconfident.

What it involved

• Quant engine in pure, dependency-free TypeScript: Black-Scholes, the Breeden-Litzenberger risk-neutral density, expected move, EV, and Kelly, all unit-tested.
• Model Lab: author, save, and reload my own probabilistic models, with the expected value and half-Kelly size recomputing live.
• Edge Radar: an agentic scan that ranks the universe by where a model's distribution diverges most from the options market, the gaps a single price target hides.
• Model Arena: the market, my view, the Street, and an AI analyst overlaid on one axis, with a Brier-score scoreboard that grades which model has actually been calibrated.
• Clearly labeled math: the implied density is labeled risk-neutral (Q), not real-world (P), with the volatility risk premium surfaced instead of mislabeled.

Stack

• Next.js 16
• React 19
• TypeScript
• Tailwind v4
• d3
• Claude (anchored AI analyst)
• Options math