A red team for your trade ideas

Most trading tools answer “what happened in similar setups?” ThesisBreaker answers a different question: which of your assumptions is false right now?

Every trade idea rests on a chain of beliefs — about fundamentals, price structure, timing, liquidity, macro and sentiment. A trade usually fails at its weakest link, and that link is rarely the one the trader was thinking about. ThesisBreaker is a pre-trade red team: it names those links and tries to break them before the market does.

It is built for tokenized U.S. stocks on Bitget (rTokens), which trade 24/7 even when the U.S. cash market is closed, and for crypto. That off-hours gap creates assumptions — about tracking, liquidity and gaps — that are easy to miss.

1. You describe a trade idea in plain English.

2. ThesisBreaker identifies the ticker and fetches live price evidence from Bitget: last price, 24h change and range, volume, spread, and the last 72 hourly candles.

3. Qwen extracts 4–8 explicit, testable assumptions. Each gets evidence for, evidence against, a status (holds / weak / broken) and a 0–100 fragility score. Every model answer is validated against a strict schema before it is shown or saved.

4. A verdict — PASS, REVISE or BLOCK — names the single weakest assumption and what would have to be true for the trade to work. Simple rules can escalate the model's verdict when the scores cross a threshold, but never soften it.

5. The result is saved to your history. ThesisBreaker never places orders.

Market data: Bitget public spot REST API v2 (tickers, candles, symbol list). Exchange symbols are verified against Bitget's live symbol list — they are never guessed.

Model: Qwen (qwen3.8-max by default) via Bitget's OpenAI-compatible hackathon endpoint, called only from the server. Results for the same idea are cached for 24 hours to save credits, and token usage is logged.

Offline mode: if the model is not configured or fails, a deterministic keyword-and-price heuristic produces a clearly labeled “offline heuristic” analysis instead.

Coming in Phase 2: fundamentals, estimates and news via the Bitget Agent Hub MCP, and optional paper orders through a Bitget Agentic Account — always with a human confirmation.

ThesisBreaker is a research tool, not financial advice. It does not recommend buying or selling any asset, and its verdicts describe the assumptions in an idea — not future prices.

Model output can be wrong, incomplete or out of date. Background knowledge in the analysis is marked “[General knowledge]” and may be stale; only items marked “[Bitget market data]” come from live data. Tokenized stocks and crypto are volatile and can lose value quickly.

AI analyzes; you decide. You are responsible for every trading decision you make.