Help

Describe the fund you need in plain words. The app turns your question into stage-1 filters (fees, risk, currency, asset class and so on) and up to 8 criteria (themes such as "Dividend / income", or statements a fund document could confirm). You can edit all of them.

It is a research aid: scores are AI estimates from fund documents, not ratings, advice or a suitability check.

Filters apply instantly in your browser. When 250 or fewer funds are left, AI rank reads each fund's KID excerpt and top holdings and gives an AI-estimated likelihood (0–100) for every criterion. Weights and limits then re-sort the matrix and the list instantly, without another run. No fund is perfect: the list shows each fund's weakest criterion.

About 13,166 share classes in 4,387 funds, from the funds' own Key Information Documents (KIDs) plus issuer data. The data is a snapshot taken on the export date and refreshed by a new release, not live.

• Currency, ongoing charge and the risk indicator: almost every class (99.7% or more).
• Asset class, region and index-tracking: every class. Distribution (accumulating or distributing): 96%.
• SFDR article: 47% of classes. Fund size (AUM): about 35%.
• Returns and volatility: about 11,500 classes with a price history, in each class's own currency, to its latest price date. They are price returns, not total returns, so distributing classes look worse than they are.
• Prospectuses, SEC filings and other secondary sources are not in the catalogue yet.

There are no P/E ratios, price-to-book, earnings, ROE or any other issuer fundamentals, and no analyst ratings. If your question asks for them, a "Not in data" note appears under the search box and those words are left out of the ranking rather than guessed.

An actual dividend yield is known for only about 783 iShares classes. For everything else, "high dividends" is approximated by the "Dividend / income" theme, plus the Distributing filter when it is set. Add the Income objective filter to tighten it. The note under the search box says when this proxy is in use.

Both run from 1 to 7, but the SRI (PRIIPs KIDs) and the SRRI (older UCITS KIIDs) are calculated differently, so a 4 on one is not a 4 on the other. The app always shows which scale a number uses. On the matrix, circles are SRI, squares are SRRI and crosses are funds whose scale is unknown. You can filter on the scale itself.

Each fund has two separate indicators.

Tier: Verified ●

Every filter could be checked against known data.

Tier: Unverified ○

The fund passed, but some data was missing (for example no SFDR article), or AI rank could not tell on a criterion. Missing data never removes a fund; it moves it below the verified ones.

Scoring: Not scored —

Not AI-ranked yet. The score is a precomputed "Theme score": an AI reading of the KID, raised by holdings exposure where holdings are trusted (never for synthetic funds). It is never called a "Match".

Scoring: Scoring ◐

AI rank is working on it. Dots move as results arrive.

Scoring: Scored ●

Every active criterion has an AI rank answer: a "Match" score.

Scoring: Cannot tell ■

AI rank couldn't decide at least one criterion from the documents. That criterion is left out of the score, and the fund is Unverified. It never counts against a limit, so check these funds by hand.

• Up to 200 questions interpreted per rolling 24 hours.
• Up to 30 AI rank runs and 5,000 funds ranked per rolling 24 hours. Your current usage is on the Settings page.
• AI rank takes at most 250 funds and 8 criteria at a time. A run stops after about 25 seconds; if it didn't finish, run it again to rank the rest. Funds already ranked with the same criteria are reused and don't count again.

/ search · j / k next and previous fund · Enter open the fund · r AI rank · 1–5 matrix axes · Esc close.

Sign-up is limited to @patchworks.ai email addresses. The address is not verified by email yet, so treat this as an internal tool. Your searches and AI rank results are private to your account. A shared link carries the filters and criteria, but not your saved search.

Everything in the app is also available over a JSON API. Interactive reference docs are at /api/v1/docs; requests use your signed-in session.