Parallax

README excerpt

# Parallax

**An interactive macro research terminal that compares source evidence with market expectations.**

Parallax investigates a research question, builds a world belief from source evidence and a market belief from price data, then measures where they disagree. Every run saves its sources, calculations, model reasoning, risk checks, and paper execution decision in a dark, interactive desk.

Built for a hackathon demo with Next.js, React, TypeScript, PostgreSQL, OpenAI, and Playwright.

## What it does

- Investigates live sources through an AI-controlled browser, with web, PDF, video, and audio inspection when available.
- Compares evidence and market signals across US policy, Japan policy, and risk appetite factors.
- Explains divergence, proposes a thesis, and runs three independent model challenges.
- Applies deterministic risk checks before recording a paper trade or an explicit skip reason.
- Restores saved investigations with source provenance, dated charts, stage events, and expandable reasoning.

```text
SCAN → DETECT → INVESTIGATE → MODEL → COMPARE
  → EXPLAIN → FORM_THESIS → CHALLENGE → RISK_CHECK → EXECUTE
```

Full research continues through explanation and risk evaluation even when scores are low or factors are incomplete; the resulting thesis can abstain. Anomaly scans investigate only after a trigger and can stop at comparison. Runs without usable evidence can stop at investigation.

## Quick start

Requirements: **Node.js 22+**, **PostgreSQL 16** (Dock