Skip to main content

The web dashboard

ot web                   # → http://127.0.0.1:8787

The dashboard is the desk's front end. It runs locally, reads your watchlist, and degrades gracefully to keyless data panels when no LLM engine is configured.

OpenTrading web dashboard

Reading a ticker page (top to bottom)​

  1. Consensus strip — one chip per analyst (engine, quant, TimesFM, AI, debate), colored by its tilt. The verdict is computed from the votes: CONSENSUS LONG/SHORT when 2+ agree and none oppose, NEUTRAL with no majority, THIN when too few have reported, and ⚠ STAND ASIDE the moment any two analysts point in opposite directions.
  2. Chart + walls + cones — price with the dealer call/put walls and the forecast cones overlaid. The wall DTE window is tunable from the Confluence-ladder header (0DTE / 5 / 7 / 14 / 30 / 45 / 60 / 90) — pick 0DTE for intraday pin risk, longer windows for swing structure. The ladder shows the net GEX at that window, which shifts as you widen it.
  3. Bull vs Bear — the committed call: a 5-tier verdict with entry, invalidation and a time stop. See the debate.
  4. Confluence ladder — every price level the desk emits on one axis; the rows named by 2+ independent methods (marked ×2) are the ones that matter.
  5. AI analysis — a grounded second opinion: summary, action, sentiment gauge, sniper levels and risks.

Switching the engine​

The header has a model dropdown. It controls the AI Analysis and News Analysis panels; results are cached per (ticker, engine, model).

The debate has its own engines pill — click it to open a small picker with three modes:

  • diverse (default) — bull and bear run on different engines for perspective diversity; the judge on Claude.
  • single — all three roles run on the header engine·model you selected (the pill mirrors it live as you change the header).
  • custom roles — three dropdowns appear: pick an engine·model for the bull, the bear and the judge independently (leave any on auto for the desk default). Great for A/B-ing judges or pitting two models against each other.

Switching costs no tokens; the new assignment applies on the next run. Note that Claude Code · default (and Codex default) means your CLI's own default model — change it with /model in the CLI, or pin one here.

Your verdict is cached one-per-(ticker, language) and mirrored to data/debate-cache.json, so it never disappears when you change the engine picker and it survives an ot web restart — hit ↻ to re-run with the current engines.

English / 中文​

A 中文 / EN toggle in the header (or ?lang=zh) flips the entire UI and the model's output between English and Simplified Chinese — the same desk for a US book or an A-share / HK book.

The Learn tab​

#/learn is the built-in textbook. Every module on a ticker page has a ? help chip that jumps to the matching explainer — the confluence ladder, forecast cones, dealer gamma, smart money, calibration — each with a worked real-world example. Start there when a number isn't obvious.

Macro & Flow​

The dashboard home shows the global picture: the macro score (rates & liquidity), Fear & Greed, BTC funding, SPY dealer gamma, Hyperliquid perps, on-chain whale flow, engine diagnostics, and Polymarket crowd odds.