Review 03 · Markets & Capital
Prior Art.
A keyless patent desk that does not try to out-search the big engines. It helps you search them properly and writes down what you missed. On test day, two of its feeds were down.
- Tested 27 September 2026, 20:10–20:21 UTC
- Where the work happens through a narrow relay on the tool’s own site
- Against Google Patents, The Lens, PATENTSCOPE
1What it is, and who it is for
Prior Art is a desk for a first prior-art pass. It writes any patent number correctly for each engine; compiles one search into the dialects of Google Patents, Espacenet, USPTO PPS, The Lens and PATENTSCOPE, flagging each one's traps; carries the whole CPC scheme; shows a weekly gazette of new US and European grants; searches PubChem's patent collection; and keeps a log that ends with what the search did not cover.
It is for inventors, founders, students and professionals who need several free engines without learning each one's syntax.
2How it works
Most of the work happens in your browser. patents/assets/js/patents.js wires tested modules together: pn.js normalises numbers; pq.js compiles the neutral query into five dialects using only operators its own recon verified for each engine; cpc.js reads a vendored 23 MB pack of CPC release 2026.08; and log.js builds what this search missed only from what the log shows was not done (pq.js 1–5, 104–186; log.js 46–70).
The in-page search calls PubChem's SDQ backend and its PUG-View record API straight from the browser; the source file itself calls SDQ 'not a documented API' (pubchem.js 9–11). The weekly gazette comes through gazette.php, a keyless relay on the same site that reads SureChEMBL at EMBL-EBI for the US issue and the EPO publication server's weekly list for Europe. It caches every answer, serves a stale copy labelled as stale for up to eight days, backs off a failing upstream for five minutes, and spaces its calls 1.1 seconds apart (gazette.php 2–50). A keyed relay for EPO OPS and the USPTO portal sits dormant, answering 501.
3The test
We loaded the desk, pulled the newest US and European gazettes through its relay, opened a grant, and repeated the in-page PubChem search.
Loaded the page.
HTTP 200, 55,535 bytes. The CPC pack is 23 MB in 681 subclass shards.
Newest US issue through gazette.php.
HTTP 200: the issue of 22 September 2026, 6,126 grants, sourced 'SureChEMBL (EMBL-EBI), from the USPTO issue', served from a copy fetched at 20:11:48.
Newest EP issue.
HTTP 502 in 0.28 s: 'The upstream did not answer, and this relay has no recent copy.' The EPO server itself answered 429 that minute.
One page of the US issue, then grant US-12740716-B1.
One page: 20 of the 6,126 grants (307 pages). The record carried title, applicant, four inventors, filing and grant dates, 8 CPC and 6 IPC codes and the abstract.
The in-page PubChem search for 'graphene', with and without a priority year.
HTTP 404, a 373,109-byte 'Not Found' page, for every query and for the bare endpoint. A PUG-View record read still answered 200.
4Against the field
Each rival was read on its own pages on the test day. Pencil marks what it does better; red, where Prior Art goes further.
Google Patents
patents.google.com · checked 27 September 2026, 20:22 UTC, from its help pages
Full-text search across many offices, from the US, Europe, Japan, China and Korea to WIPO, plus non-patent literature from Google Scholar, Google Books and the Prior Art Archive.
In pencil
- It searches the full text itself and returns ranked results; Prior Art only writes the query and opens it there.
- It includes non-patent literature, which the desk's own log says it does not cover.
In red
- It writes the same search for Google and four other engines at once, warning about Google's traps: wildcards that expand to only 25 forms, and automatic stemming with invisible synonyms.pq.js 124, 139, 155
The Lens
lens.org · checked 27 September 2026, 20:22 UTC
An open platform with patents from over 100 countries and over 200 million scholarly records, linking patents to the works they cite.
In pencil
- Patents and scholarship searched together and linked through citations, with shareable collections and a documented API.
In red
- The whole CPC scheme on the page, with a words-to-codes lookup that shows what each suggestion matched.index.html 655–662; cpc.js 1–12
PATENTSCOPE
wipo.int · checked 27 September 2026, 20:22 UTC
WIPO's database: international PCT applications in full text on the day they publish, national and regional collections, non-patent literature, and search in several languages.
In pencil
- PCT applications in full text on publication day.
- Searching in other languages; the desk's log admits it works with English terms only.
In red
- It knows PATENTSCOPE's own rules while compiling: a leading * or ? is removed, with the reason given.pq.js 142
5Where it falls short
- An undocumented dependency. The in-page search relies on an endpoint its own code calls undocumented; at 20:21 UTC it answered 404 to everything.
- Half a gazette. The European issue failed with a 502 while the EPO was rate-limiting, and the relay had no cached copy to fall back on.
- No full text of its own. It writes queries for the big engines; it does not search them.
- Thin rows. PubChem's patent rows have no claims text or filing and publication dates to filter on.
6The verdict
Prior Art is a patent searcher's tutor and notebook rather than a search engine, and a careful one; on test day its in-page search and its European gazette were both down.
- In red ink
- One search compiled for five engines with each trap flagged, 254,314 CPC codes on the site, and a log that names its own gaps.
- In pencil
- For reading full text and ranked results: Google Patents, The Lens or PATENTSCOPE.