Methodology
We relay and check work that is already published. We do not run our own trading tests, and we never present an AI result as a promise.
1. We start from published work
Our raw material is research and data that others have published: academic papers (arXiv, SSRN, NBER, conference proceedings), public AI leaderboards and benchmarks, central bank and regulator reports, company filings on SEC EDGAR, and official disclosure systems such as the House Clerk and Senate eFD for congressional trades.
2. We open the primary source
A news story, a post or an aggregator is a lead, not a source. Before we publish, we open the original document: the paper, the leaderboard page, the filing. Every article ends with a source box that lists each primary source and its status.
- Verified We opened it and checked the figures we use against it.
- Linked Provided by the authors (code, data). Linked for you, not re-checked by us.
- Unconfirmed Reported but not confirmed by a primary source. We say so in the text.
3. Every number is checked
Each figure in an article is checked against the primary source, with the date we checked it. If a number cannot be verified, we do not publish it. When a source contradicts itself, we say so or leave the number out.
4. What it shows, what it does not
Every article includes a "What it shows / What it doesn't" box. Benchmarks are not trading results, backtests are not live results, paper trading is not real money, and a study run by the company that sells the product is not an independent audit. We spell those limits out.
5. Who Bought What
For trades by members of Congress, insiders and funds, we use the official filings (STOCK Act reports, SEC Form 4, Form 13F). Third-party trackers are used only to spot a filing; we confirm on the official record. We report facts and dates. We never claim or imply that a trade was illegal or based on inside information.
6. Market data
Live quotes on this site come from TradingView widgets and belong to TradingView and its data providers. We do not republish market data feeds.
7. AI in our workflow
We use AI tools to find, summarize and draft. AI output is never published unchecked. Each article carries a note on how AI was used.
8. No naked percentages
A percentage without context can say almost anything. So every percentage or score we cite comes with a context badge that gives four things: the sample size (n=), the period it covers, the source, and the margin of error or confidence interval when the source publishes one. For example:
The top model scores 65.40% n=450 test questions, 3 runs · Oct 12, 2026 leaderboard · Vals AI · ±0.32 SEM
- If the source does not give a figure, the badge says not disclosed. We never estimate or invent it.
- When a source publishes two different numbers for the same thing, we show both and say so, or we leave the number out.
- Differences smaller than the published margins are described as unsettled, not as a win.
- Our site generator refuses to pass a page that cites a percentage without its badge.
The rule also applies to our directories of AI tools and AI models, where every price carries the date we checked it.
9. Corrections
If we get something wrong, we fix it and note the change and the date at the bottom of the article. Send corrections to wallstreetdebug@gmail.com.