The dialectical, auditable AI research desk.像辩论员而非黑盒的 AI 投研台。
Illustrative debate structure · not a live market call
“Buy.”
多头：买入。
The evidence supports improving cash generation and operating leverage. The market may still be under-pricing that change.
“Sell.”
空头：卖出。
The same evidence may already be reflected in valuation. Wait for a better entry or stronger confirmation before taking risk.
Reduce both sidesto one trade.
把两边的话，化成一笔交易。
Specialist agents — fundamentals, sentiment, news, technical, optional macro context, plus bull-and-bear advocates — argue in writing. A manager weighs every line and outputs a single, confidence-calibrated call. The whole transcript is auditable.
Availability follows live data, model and access status
Not a toolbox. One continuous evidence chain.
Monthly Alpha, weekly opportunities and morning picks discover at different horizons. Public Equity explains. Seven agents challenge and decide. The watchlist monitors. Verified records calibrate. Each stage inherits the security, evidence and timestamp boundary from the last.
Discover
Across five days, one month and six months, start from future events, earnings paths and profit transmission.
Research
Use 23 Public Equity workflows for financials, valuation, catalysts and risk.
Decide
Seven agents debate the same evidence and a manager produces one actionable call.
Monitor
Track candidates against catalysts and explicit invalidation conditions.
Review
Validate the process with completed decisions and reproducible backtests.
US, A-share and HK decisions use separate real adapters and market rules. Crypto is available only when CCXT is installed; no market falls back to synthetic prices.
Only decisions actually completed by the backend.
Loading live records…
One ChatGPT prompt cannot do five specialists' jobs.
一条 prompt 干不了五个专家的活。
Ask a single model “should I buy AAPL?” and it hedges every direction. Attention is finite. Conflicting signals get smoothed into a HOLD with low confidence — rarely the optimal trade.
Our pipeline runs five specialist analysts with separate prompts, separate context windows, separate evidence. Each forms an opinion in isolation. Then a bull and a bear persona read all five reports and write opposing pitches. Then a trader synthesises. Then risk approves. Then a manager signs off.
Confidence calibration must be demonstrated out of sample. The track-record page publishes live completed decisions immediately, but leaves performance blank until a reproducible backtest passes its verification checks.
Seven stages. Every one auditable.
七个阶段。每一步都可回溯。
- 01
Fundamentals
SEC EDGAR · akshare
- 02
Technical
OHLCV · Alpha158-lite factors
- 03
Sentiment
Reddit · 东方财富股吧
- 04
News
Public headlines · source-labelled
- 05
Macro
FRED / OpenBB · only when configured
- 06
Bull / Bear debate
Two adversarial personae
- 07
Manager + risk
Synthesis + position size + stop
Three equity routes. No synthetic market fallback.
美股、A 股、港股分别路由；上游缺失就明确缺失。
| Market | Route | Primary sources | Runtime rule |
|---|---|---|---|
US Equity | US adapter | Yahoo Finance · SEC EDGAR | Live, source-dependent |
A-Share | CN adapter | Tencent · Sina · optional akshare | Live, source-dependent |
HK Equity | HK adapter | Yahoo Finance .HK symbols | Live, source-dependent |
Missing data should remain missing.
A decision is useful only when its evidence is traceable.
数据缺失，就应该明确缺失；
证据可追溯，结论才值得讨论。
— Concordal · 2026
Request access to the working beta.
申请体验当前可运行的测试版本。
Payments are not open. Submit your use case and we will confirm the current data, model and access limits before onboarding.