How we find the picks
Every week the USPTO records a new batch of patents that expired because their maintenance fees went unpaid, while others simply reach the end of their 20-year term. This page explains how we turn both into one ranked list.
1. Pull the lapses
The USPTO publishes a file of every maintenance-fee event since 1981. A patent lapses when its 3½-, 7½- or 11½-year fee goes unpaid through the six-month grace period that follows, and the office records the expiration (as code EXP.) about five weeks after that deadline passes. The batch dated Oct 5, 2026 had 3,852 of them.
2. Add the patents that reach full term
We also list patents that simply ran their course. A utility patent lasts 20 years from its earliest US filing date, counting the parent application for continuations and divisionals, plus any extra days the USPTO granted for its own delays. We compute that date for every patent still in force in the categories we follow and add the ones that expire each week. These owners paid every maintenance fee, which is about the strongest evidence there is that a product earned money. Patents with a terminal disclaimer, which ties their expiry to another patent, are left out because their date can't be worked out from the patent alone.
Swipe sideways to see the whole figure.
Swipe sideways to see the whole figure.
3. Drop anything that came back
Owners can ask the USPTO to accept a late payment and reinstate the patent. Of the patents that lapsed since 2018, about 1.5% were eventually revived, roughly a third of those within 90 days and most within a year. We remove any patent that shows a later payment, a reinstatement or a pending petition, and we check again every week.
4. Keep the consumer categories
Each patent carries Cooperative Patent Classification codes, and for the main list we keep the ones in categories people shop for, such as kitchen, home, pets, toys, sports, outdoors, tools, apparel, bags, baby and garden. That leaves about 5% of the batch. Chemistry and pharmaceuticals are left out entirely, while software, medical devices and industrial equipment go to their own tracks, described below.
5. Read them
A language model screens each remaining patent against a strict rubric, asking whether it is something an ordinary person would buy and whether a small shop could make it, and it drops the industrial parts and medical devices that the classification codes let through. The patents that pass get a written brief covering what the product is, what the main claim actually required, how new the idea was, the likely materials and manufacturing route, and the words a shopper would search for. Every brief links to the original patent so you can check it against the source.
6. Check demand
For the strongest candidates we search the web for similar products on sale now, note typical prices and a few example listings, and rate how much evidence there is that people buy them. A product with many sellers and reviews scores higher than one that nobody seems to make.
7. Look for strings attached
Patents often come in families, and if a lapsed patent has a parent, continuation or divisional that is still in force, the product may still be covered. We flag every related patent that we can't confirm has expired.
Tracks
Consumer products are the core list, but lapsed patents cover every field, so each patent is sorted into one track by its classification and readers choose the tracks they want to follow:
| Track | What's in it |
|---|---|
| Consumer products | Kitchen, home, pets, toys, outdoors, tools, apparel: things a small shop can make and sell. |
| Gadgets & electronics | Audio, chargers, smart-home devices, wearables, accessories with a circuit board. |
| Cybersecurity & privacy | Authentication, encryption, access control, fraud and threat detection. |
| Software & AI | Machine learning, search, user interfaces, developer tools, data processing. |
| Fintech & commerce | Payments, e-commerce, marketplaces, point of sale, logistics software. |
| Medical & health devices | Braces, monitors, mobility aids, therapy and clinic equipment. Usually regulated. |
| Auto, bikes & mobility | Vehicle parts and accessories, bicycles, scooters, trailers, EV hardware. |
| Energy & batteries | Batteries, solar, power electronics, heating and cooling. |
| Manufacturing & industrial | Machine tools, fixtures, robotics, molding, material handling, fasteners. |
| Food, farm & garden tech | Food processing, farm equipment, irrigation, livestock and greenhouse gear. |
Software, security and fintech picks are judged on how a small team would build the idea today and whether anyone already sells something like it, instead of on materials and tooling.
The opportunity score
Each pick gets a score out of 100, built from five factors that are each measured from 0 to 1 and then combined with fixed weights. All five appear on every patent's page, so you can see why something ranked where it did and disagree with us if you know the market better.
Swipe sideways to see the whole figure.
Swipe sideways to see the whole figure.
| Factor | Weight | How it's measured |
|---|---|---|
| Demand | 30% | A web search for products like it that are on sale now, rated from 1 to 5 on how much evidence there is that people buy them. If the search finds only similar products rather than this design, the rating counts for 80%. When we haven't run a search, the figure is our own estimate of how familiar the product is, discounted, and it's marked "est." |
| Makeable | 20% | How hard it would be to produce, judged from the part count, the tooling it needs and whether it relies on electronics. Simple molded, stamped or sewn goods score highest. When the patent covers a material, a chemical composition or a manufacturing process rather than a product's design, makeable is capped at 2 out of 5, because copying it takes formulation and process work rather than a mold. |
| Novelty | 20% | Mostly measured from the patent record. Just under half is the originality index used in economics research on patents (Hall, Jaffe and Trajtenberg, 2001), which looks at the earlier patents the examiner and applicant cited and asks how many different technical fields they come from: an invention that draws on several distant fields tends to be a more original combination than one that only builds on its own field. About a third is claim breadth, the length of the patent's first claim, because every word in a claim narrows what it covers; a first claim of 60 words or fewer scores fully and the score falls steadily to its floor at 260 words. The remaining fifth is our reading of how new the idea was compared with what was common when it was filed. |
| Track record | 15% | How long the owner kept paying maintenance fees. A patent dropped after the owner renewed it once (at the 8-year mark) scores highest, because renewing suggests the product earned money while the design is still fairly current. One dropped at the first fee often covered something that never sold. |
| Clear of risks | 15% | Points come off for related patents (continuations or divisionals) that may still be in force, judged by their age because the patent record keeps calling a patent "patented" after its term ends, for products that need safety certification, and for consumer products that depend on electronics. |
The weights are a judgment call, and we'll adjust them as we learn which picks readers actually act on.
What we can't tell you
- Whether a design patent covers the product's appearance. Design patents have no maintenance fees, so they never lapse this way and never appear in our data.
- Whether the product's name, look or packaging is a trademark.
- Whether some other company's patent covers the same thing.
- Whether the owner will revive the patent later.
That's why every pick is a lead rather than a clearance, and why it's worth having a patent attorney run a freedom-to-operate check before you spend money on tooling.