A council of three AI agents votes on every buy, sell, and hold. The reasoning is in the trail. Backtest it, paper-trade it, take it live — at your pace.
Pick a typical trading bot. It bought QQQ at 14:00 ET on 07 May. Why? "Signal triggered." That's not a trade you can defend — to yourself, your spouse, or the IRS.
QuorumTrading writes a paragraph for every decision and stamps it with three independent votes. Open any council tick and you get the chart, the trail, the reasoning — the same panels you'd see if it were running your money.
| When | Action | Votes (T · F · S) | Price |
|---|---|---|---|
| 13:54 ET | HOLD | ✓Tech✕Fund✓Sent | $451.80 |
| 10:00 ET | BUY | ✓Tech✓Fund✓Sent | $942.10 |
| 07 May | BUY | ✓Tech✓Fund✓Sent | $445.30 |
Technical reads price action and momentum. Fundamental reads filings and ratios. Sentiment reads news, filings, and social. Each runs on its own model with its own context window — no agent sees another's reasoning until after the orchestrator counts the vote.
The cards on the right show what we actually measure: how often each agent voted with the final consensus, and how often the orchestrator chose to override the majority. We don't track "accuracy" — there's no labeled ground truth for an agent's vote being right.
A QuorumTrading strategy is a markdown document with a short YAML frontmatter. The frontmatter holds the machine-readable rules — evaluation cadence, tickers, consensus method. The body is plain English: entry signals, exits, risk caps, what each agent should pay attention to.
Three agents read both before they vote. Fork from the public library or start with a blank page. Revisions are stamped, diff-able, and one click to roll back.
--- version: v3.2 consensus_method: MAJORITY consensus_threshold: 0.65 evaluation_interval_seconds: 3600 stock_tickers: [QQQ] market_hours: timezone: "America/New_York" evaluation_times: ["10:00", "14:00"] ---
A trend-following QQQ strategy. Enter on confirmed golden crosses, exit on death crosses or trailing-stop hits. The council weighs technical signals against earnings windows and macro sentiment.
Replay 1–10 years of historical bar data. The council votes on each tick as if it were live. You get the equity curve, the decision trail, every hold and every override.
Every backtest renders against benchmarks. Below: Tech ETF — MA vs. QQQ buy & hold and the public Megacaps Rotor strategy on the same 5-year window. Click any tick in-app to restart the council from that point with a tweaked rule.
Three states, one toggle each. The council doesn't know the difference; only the broker does. Cash floor, max position, max drawdown enforced in code at every state.
Same data feeds, same agents. Orders never leave the platform. Debug a strategy quickly without spinning up a broker account.
mockReal fills against real market data with paper capital. Most users live here 4–12 weeks before promoting.
paperOne toggle, one confirmation. DCA top-ups optional. Pause from the dashboard any time — the trail keeps writing.
live