How we track, rank and test

1. What gets listed

A project enters the registry when it is clearly part of the AI-for-investing stack and is either open source with a public repository, or a product whose claims can be checked against public documentation. We currently track 78 projects, 71 of them with a GitHub repository. We prefer projects that are used by other projects in the registry (a data library most agents import, a broker most frameworks integrate) over projects that merely mention AI.

We do not list a project because its README promises returns, and we do not de-list a project because it has few stars. Stale projects stay listed with a maintenance note when they are still a reference point. Anyone can propose an addition or a correction through the submission form.

2. GitHub metrics

A sync job calls the GitHub REST API once a day for every repository in the registry and stores stars, forks, open issues, language, license, topics, creation date, last push and the latest tagged release. Those fields are shown unmodified. The timestamp of the last sync is printed in the footer and next to every live table.

Star deltas and momentum

• Stars · 7d / 30d is the difference between today's star count and the count 7 or 30 days earlier.
• The preferred source is our own daily snapshot, which the sync job appends to a history file in the repository. A window is only reported from snapshots once the oldest snapshot is at least that many days old.
• Until our history covers a window, we use OSSInsight's public star-history endpoint, which reconstructs daily cumulative stars from GitHub's public event stream. Its counts can differ slightly from GitHub's live number; we only use it for differences, never for the headline star count.
• 30d momentum is the 30-day gain divided by the star count 30 days ago, shown as a percentage. It favours small, fast-growing projects by design; sort by stars if you want size.
• A dash means no source covers the window for that repository yet. We never estimate or extrapolate.

Radar statistics

• Releases · 30d counts repositories whose latest tagged GitHub release was published in the last 30 days.
• Pushed · 7d counts repositories whose default branch received a push in the last 7 days.
• Stars tracked is the sum of current stars across all tracked repositories.

3. Rankings

The leaderboard and radar are sorted by the objective columns above and nothing else. There is no editorial weighting, no “InvestingStack score” yet, and no sponsored positions. If we introduce a composite score, the formula and its inputs will be published here before it appears on any page.

4. Descriptions

Taglines and descriptions are written by us after reading the repository, its documentation and, where one exists, the paper. We describe what a project is and does, not what it claims to earn. Performance claims from READMEs are quoted only inside research articles, verbatim, with a note on whether a period and benchmark were stated. The maintainers' own one-line description is shown separately on each project page so readers can compare.

5. Experiments

Experiments are pre-registered: the hypothesis, the data, the procedure and the success criteria are published before the run starts, and the page is marked pre-registered. While data is collected it is marked running. Results are published with the raw inputs and outputs committed to the public data folder so that anyone can recompute them. We do not change a design after the run starts; if we have to, the change and its reason are logged in the article.

We do not run simulated or real portfolios and we publish no return figures of our own. Experiments score systems on what can be verified from their code, documentation and published artefacts, not on market outcomes.

6. Limits

• Stars measure attention, not quality or returns. Star counts can be gamed; we have no detection for that beyond common sense.
• GitHub metadata lags by up to a day and the OSSInsight series by a further day or two.
• The registry is curated by hand and will miss things. Tell us.