Investment Lab is the public side of a personal investment discipline system. Most investing tools tell you what to buy. This one checks a decision against written rules and shows the evidence, and it never recommends a trade.
What you can try
Pre-trade gate. Pick a fictional portfolio, or type in your own holdings (they stay in your browser; the server does not save them), propose a trade, and see which position limits, memo requirements and checklist items it passes or fails. The self-check questions at the end (for example "not driven by a short-term price rise") re-run the gate as you tick them. You can also pick a well-known institution: its ten largest holdings from its latest 13F are scaled to a $10,000 sample portfolio so you can see how your gate treats a portfolio like theirs.
Financial snapshot. Ten large US companies are tested one by one, and any other US-listed company registered with the SEC works too. The page shows eight quarters of revenue, margins, cash flow, capital expenditure and segment results. Values derived from reported figures, such as a fourth quarter computed from the full year minus nine months, are marked with their formula, and each figure links to the SEC filing it came from. A company page also has:
- "What happened last quarter": an AI reading in which every point cites its source and its exact wording is checked against the filings, plus a plain-language sentence that is not allowed to add reasons or judgements the original does not contain.
- An estimate of the next earnings date, and a small macro strip (Treasury yields, the fed funds rate, inflation, unemployment), labelled as background only.
- Insider trades (Form 4): open-market buys and sells by officers and directors over the last 12 months, grouped by person and charted by month. Grants, option exercises and tax withholding are left out, and the page says it is not a signal.
- Well-known institutions' holdings (13F): whether each institution on a fixed list held the company at their latest quarter-end, how large the position was, and the change from the quarter before.
Reference portfolios. The 13F profiles of eleven well-known institutions (Berkshire, ARK, H&H, Himalaya, Pershing Square, Oaktree, Bridgewater, Duquesne, Baupost, Renaissance, Soros), plus any other 13F filer by CIK: largest holdings, concentration, new and closed positions since the previous quarter, and a draft rule file whose numbers are copied mechanically from the filing, which you can adjust and carry into the gate. The AI structure reading works like this: code computes every number, date and company name and fills them in, the model writes only text without numbers, and a reading that fails the checks is rejected.
AI skeptic. Write a one-line thesis about a company and a model argues the other side using the same SEC data. It never says buy or sell, and its answer is shown only if every number matches the filings. Example theses are hidden until you ask for them, so they don't steer your own idea. There is also a memo flow: write your reasons, answer three counter-arguments, state invalidation conditions and a review date, ask the AI whether your answers are clear, then record your own decision; the gate reads that memo.
The Lab is available in English and Chinese.
What it is not
The portfolios are fictional and the rules are a demo rule set. Nothing here is investment advice, and the system has no code for placing orders. Insider trades, institutional holdings and macro data are background only: they do not feed the gate or the rules, and the AI never sees them. The AI skeptic and the quarterly reading run on claude-haiku-4-5. The skeptic passed 34 adversarial cases (requests for advice, prompt injection, made-up numbers, off-topic input) with no advice shown, and the results are published on its page. The AI reading of 13F structure has no test set yet, and the page says so.
The full case study is under Projects: Investment Discipline System.