Supplier evidence ratings

What the number is

The Component Atlas, the Service Atlas and the RFQ picker list real companies, taken from public sources. 10xAnd has not contacted them and does not endorse them. Next to each one is an evidence rating: how many of the 4 scored checks found what they looked for in a public source, on a stated date. A further check, “Site names this capability”, is shown with its source as information only and is not counted.

Evidence 2/4 Illustration: 2 checks found, 1 not checked, 1 not found.

Filled segments are checks that passed. A dashed segment is a check that could not be run, and it counts neither way. An empty segment is a check that ran and did not find it. Open any listing to see each check, the page it was found on, and the date.

The checks

Information only

Site names this capability is run and shown on every listing, with what it matched and where, but it does not count toward the score, so the score runs from 0 to 4. A check counts only while at least 90% of the passes in an independent hand sample are confirmed, and this one has not been in one. During development, on 2026-10-20, reviewers confirmed 40 of 46 sampled passes (95% interval 74–94%) by hand. After the fixes, a second sample from outside the first confirmed 31 of 38 sampled passes (95% interval 67–91%). The misses were the family word for a specific product (needle for a Tuohy cannula, a Veress or a fistula needle), terms found in unrelated sentences (packaging; validations), and an ordinary word (resist bending). A “found” here means the category’s terms appear on the company’s own pages, which is weaker than the category name suggests.

The checks are run by a script against the live sources, and every result keeps the address it was found at and the date, so anyone can repeat it. Each one ends in one of three states:

• ✓ Found. The listing links the page or record it was found on and shows the check date.
• ✗ Not found. The check ran and did not find it. The listing says what was looked for and where.
• Not checked. The check could not run: the site blocked automated reading, timed out, disallowed it in robots.txt, showed no text without JavaScript, or no longer belongs to the company, or there was no location to match an FDA record against. It is shown as not checked, never as a fail, and it scores nothing.

A result recorded without its source link or its date is not counted. Within each category, listings are sorted by this number, highest first, then by name.

What the rating does not mean

• It is not a quality score. It says nothing about workmanship, capacity, price, lead time or how a company treats its customers.
• It is not a performance claim about any device, component or service.
• It is not an endorsement or a recommendation. A higher number means more of 4 facts were found in public sources. Nothing else.
• No FDA registration is normal for many suppliers. Makers of components and raw materials are generally exempt from establishment registration under 21 CFR 807.65(a). For them, “not found” is the expected answer and not a finding against the company.
• Stated is not verified. Checks 2 to 4 record what the company says on its own site. We have not seen certificates. Confirm ISO 13485 status with the certificate and the body that issued it.
• Not checked is not a fail. A site that blocks automated reading tells us nothing about the company.

This data

Snapshot of 2026-10-22, method version 1.2, covering 559 companies (n): every company in the source lists below, including suppliers named in the Prototype Atlas. Every check also carries its own date, and a result holds for that date: a site can change after it (in the independent sample, one site’s security certificate expired two days after its check). Results per check, counted per company (for the FDA check, the company’s main listing; each listing’s own result is what its row shows), except the capability check, which is counted per company and category:

Across the 778 listings on the site (n), 778 are matched to this snapshot by website domain and 0 show “Evidence: not yet checked”. Scores among the 778 matched:

Sources read for this snapshot:

• Component atlas listings
• Services atlas providers
• Prototype atlas suppliers
• Company websites : listed URL + homepage + up to 9 same-site pages; robots.txt honoured; User-Agent "10xAndEvidenceBot/1.0 (+https://10xand.com/supplier-ratings)"
• openFDA device registration & listing : source data updated 2026-10-05

Checked by hand

Independent sample

On 2026-10-22, two reviewers re-checked by hand a sample of companies the method had never been developed or tuned on. The 559 companies in the snapshot, sorted by their web addresses written backwards (so the order has nothing to do with the alphabetical start the method was developed on), every second one from the first, until the sample held five FDA passes: 26 companies (positions 1, 3, 5 … 51), 24 of them with at least one scored pass. None of these companies was used to develop or tune the method. Two reviewers checked the sample, each on their own, by opening every cited page or record live between about 01:00 and 02:00 UTC on 2026-10-22 and searching once more for each fail. Each count below is a company-and-check judged once: the two reviewers' readings of the same 65 passes are not counted as 130. They agreed on 64 of the 65 passes; the one they split on (odu-usa.com, ISO 13485) is counted as not confirmed. They checked the snapshot of method 1.1.

• Website answers. 26 companies, re-checked by hand on 2026-10-22, sample not used to develop the method: 24 of 24 passes confirmed (95% interval 86–100%). Results of “not found” checked: 2, of which 0 should have been found.
• Site says it serves medical devices. 26 companies, re-checked by hand on 2026-10-22, sample not used to develop the method: 20 of 20 passes confirmed (95% interval 84–100%). Results of “not found” checked: 4, of which 1 should have been found. Not checked (site not readable): 2. In 3 of the 20 passes the result was right but the excerpt shown was the wrong sentence.
• Site names this capability (information only). Not in the independent sample.
• Site states ISO 13485 certification. 26 companies, re-checked by hand on 2026-10-22, sample not used to develop the method: 15 of 16 passes confirmed (95% interval 72–99%). Results of “not found” checked: 8, of which 1 should have been found. Not checked (site not readable): 2. In 2 of the 16 passes the result was right but the excerpt shown was the wrong sentence. Not confirmed: a group's certifications page on a subsidiary's site whose linked certificates name only the group's other plants (odu-usa.com).
• FDA establishment registration. 26 companies, re-checked by hand on 2026-10-22, sample not used to develop the method: 5 of 5 passes confirmed (95% interval 57–100%). Results of “not found” checked: 21, of which 1 should have been found.

A check counts in the score only while at least 90% of the passes in its independent sample are confirmed (the point estimate; the 95% interval is shown beside it, and a small sample has a wide one). Website answers: 100% (24 of 24); Site says it serves medical devices: 100% (20 of 20); Site states ISO 13485 certification: 93.7% (15 of 16); FDA establishment registration: 100% (5 of 5). All 4 scored checks clear it. “Site names this capability” has no independent sample and stays information only. Across the scored checks, 64 of 65 passes confirmed (95% interval 92–100%), from 26 companies.

The results the reviewers judged wrong, or split on:

• odu-usa.com, ISO 13485 (found): not confirmed (one reviewer of two). The group's certifications page lists ISO 13485, but the certificates it links cover only ODU GmbH & Co. KG (Mühldorf, Germany) and ODU Shanghai, not the listed Camarillo, CA company. Now: unchanged: the method counts a group statement on the company's own site.
• biofortuna.com, medical (not found): false fail. “Full validation of final IVD kits” on /our-services/assay-manufacturing/, a page the checker read; IVD kits are in vitro diagnostic devices, which the method already counted when spelled out. Now: found (method 1.2).
• kyocera.com, ISO 13485 (not found): false fail (both reviewers). americas.kyocera.com publishes Kyocera International, Inc.'s (San Diego) ISO 13485:2016 certificate, DQS 31629302 MP2016, valid to 2028-11-20. The listed www.kyocera.com redirects to global.kyocera.com, whose 10 pages read include no quality page. Now: unchanged: open (the listing's website points at the group portal).
• multisorb.com, FDA (not found): false fail for one reviewer; right under the method as written for the other. openFDA registration 3015775984, Multisorb Technologies India Pvt Ltd, Hyderabad, active, owner/operator Multisorb Technologies, Inc. The site names its “India Site” but not the town, and the method does not read the owner/operator. Now: unchanged: open (method rule).

In 5 of the 65 passes the result was right but the stored excerpt was the wrong sentence: alleima.com, cncalfa.pl and 7pcb.com (medical: what a material is used for, a truism) and sanmina.com and saltmedtec.com (ISO 13485: a customer's need on a case-study page; a related-post card that has since rotated off the cited page).

Changed after the independent sample

The sample was judged on method version 1.1. These fixes came after it, so the figures above do not test them; the snapshot on this page is method version 1.2:

• Medical: “IVD” now counts, as “in vitro diagnostic” already did. One result changed: biofortuna.com, not found to found (the sampled false fail).
• Medical excerpts: of two statements that rank the same, one about something other than the company (what a material is used for, “they” as its subject, “as we all know”) no longer wins unless it names the company. With the IVD change, 16 of 401 medical excerpts moved, among them the three the reviewers flagged. No result changed.
• ISO 13485: on a news, blog or case-study page the sentence itself, not the line it sits on, must be the company's own statement; and the excerpt comes from the company's own pages before a related-post card or a story. 9 of 212 ISO excerpts moved, among them the two the reviewers flagged. No result changed.
• All of this was re-run from the pages saved on 2026-10-20, so every other result is the one the reviewers sampled. Not fixed: kyocera.com (listing website), odu-usa.com (group certificates), multisorb.com (FDA owner/operator).

Development samples

These samples were used to develop the method, so they are its optimistic end, and they do not decide what is scored. On 2026-10-20, reviewers opened the source of a sample of ratings in a browser and judged each one by hand. The sample came from the start of the alphabetical company list (3dsystems.com to bayareacircuits.com, 64 companies), and each check was judged on its own. They found 28 wrong results. The causes were fixed (below), the checks were re-run (method version 1.1), and the sampled ratings were scored again against the new results:

• Website answers. Development sample of 37 ratings, 2026-10-20: 29 of 29 passes confirmed after the fixes (95% interval 88–100%). Before the fixes: 29 of 32 (95% interval 76–97%). Second development sample, 2026-10-21: 23 of 25 (95% interval 75–98%); after fixing what it found, 23 of 23 (95% interval 86–100%).
• Site says it serves medical devices. Development sample of 43 ratings, 2026-10-20: 39 of 39 passes confirmed after the fixes (95% interval 91–100%). Before the fixes: 38 of 38 (95% interval 91–100%). Second development sample, 2026-10-21: 25 of 25 (95% interval 87–100%).
• Site names this capability (information only). Development sample of 50 ratings, 2026-10-20: 41 of 41 passes confirmed after the fixes (95% interval 91–100%). Before the fixes: 40 of 46 (95% interval 74–94%). Second development sample, 2026-10-21: 31 of 38 (95% interval 67–91%).
• Site states ISO 13485 certification. Development sample of 47 ratings, 2026-10-20: 21 of 21 passes confirmed after the fixes (95% interval 85–100%). Before the fixes: 19 of 19 (95% interval 83–100%). Second development sample, 2026-10-21: 24 of 25 (95% interval 80–99%); after fixing what it found, 24 of 24 (95% interval 86–100%).
• FDA establishment registration. Development sample of 54 ratings, 2026-10-20: 16 of 16 passes confirmed after the fixes (95% interval 81–100%). Before the fixes: 13 of 13 (95% interval 77–100%). Second development sample, 2026-10-21: 20 of 20 (95% interval 84–100%).

These figures are for the same sample the errors were found in, so they are the optimistic end. A second sample of passes, drawn at random from companies outside that range, was read against the saved copy of each cited page on 2026-10-21 (by the build agent, not by the hand reviewers), before the last fixes.

What the development samples changed (method version 1.1):

• Website answers: an expired or mismatched certificate on the listed https:// address is now a fail even when plain http:// answers; a bot-detection “404 predicted at edge” is not checked rather than dead; and a site must name the listed company itself: the domain name in an e-mail address does not, and a short or common name (an acronym, “Design Science”) counts only when the site also states what the company is listed for.
• ISO 13485: the certification word must belong to 13485 in the same sentence and line, not to ISO 9001 or to “FDA registered”; a question, a consulting service list, “a requirement for” certification and an edition withdrawn in 2019 do not count, nor a bare mention on a site whose own pages name only a withdrawn edition; “not currently certified” or “Stage 1 audit planned” anywhere overrides; up to three quality pages are read, certificate-of-analysis pages are skipped, a news story is not a certificate, and when a quality page is linked but could not be read, “not mentioned” is not checked.
• Medical: “healthcare products” and “medical grade” no longer count on their own; a “Medical” market heading counts on a markets page when it names what it covers; excerpts come from the company’s own statement rather than a testimonial, a news teaser or a certificate’s scope.
• Capability: each category’s terms now name the capability itself (a “Get a Quote” button is not a partner network; “instruments” is not a ratchet; PCB assembly needs “assembly”), are read in the page body only, and the cited page carries every matched term when one page does. It is still shown as information only, because a second sample confirmed 31 of 38 passes.
• FDA: each listing is checked on its own name and place (a sister division’s registration no longer counts), only active registrations count, a parent or sibling registrant does not match, legal forms are dropped only from the end of a name (“APS Materials”), and a registration elsewhere matches when the company’s own site names its town with the country, or names the establishment itself (“Alleima Karlsruhe”); a miss cites one search URL that reproduces every name searched.
• Reading: the sitemap is always searched, company pages under /blog/ or /news/ are read when they are about medical, ISO 13485 or the listed capability, a non-Latin homepage is read through its English version, and up to nine pages besides the homepage are read.

Corrections

If a check about your company is wrong, or a page has moved, email [email protected] with the subject Evidence correction: and the company name. Include the listing and the public page that shows the correct fact. We re-run the check against that page. A result changes only when a public source supports the change, which is what keeps every rating repeatable by anyone.

Browse the listings in the Component Atlas and the Service Atlas.