A single search result is a candidate. Confirm it with a second identifier.

1. Before treating one result as confirmation

Match the returned entity to the licence or contract with a Chinese registered name or code. Use the official registry workflow. Do not approve the first English-name result without that match.

We used the 106 Chinese manufacturers on NHTSA’s public vPIC trailer/vehicle list. Each name was searched once on 18 August 2026. These rates describe that list, platform and day.

One row was the most common outcome. It still needs to match the party named on your contract.

2. What that one row can be

Across the whole frame the searches returned 84 candidate rows. Four things were in them that a plain list does not distinguish.

• A record in a different jurisdiction — 11 rows, 13.1%. These were registered in Hong Kong while the frame is a list of Chinese manufacturers. Four inputs naming Qingdao and Shandong returned records written in traditional characters and registered in Hong Kong. Ten of the eleven carried no 18-character code at all, and the eleventh carried an eight-digit number. That is not an error in the record. Hong Kong entities are identified by an eight-digit Business Registration Number, adopted as the Unique Business Identifier on 27 December 2023, and not by the mainland’s 18-character Unified Social Credit Code. The check-character test and a GSXT lookup both depend on that mainland code, so neither applies to these Hong Kong entities.
• A record that is already deregistered — 6 rows, 7.1%. They came back beside active records. The status is a field that must be read before a row is used.
• A branch and not a separate company — 12 of the 23 rows beyond the first, 52.2%. Their Chinese registered names begin with the first row’s name and end in 分公司. One input returned three rows that were one company and two of its branches. Each branch has its own registration record and its own code, so each can be “verified” on its own terms. Which of them your contract, invoice and payment instruction name is a separate question, and the search screen does not answer it.
• Several separate companies, every one of them registered: 6 of the 106 inputs, 5.7%. One input returned five at once. All five are registered in Shanghai. All five end in the same eight characters of industry and company form, 机械设备有限公司, and differ in the characters ahead of it. Each carries its own 18-character code and an active status, so each passes the tests this site teaches. A buyer holding only the English name has nothing in the returned rows to choose between them. In one further input, two entities carried the identical Chinese registered name, one in Beijing with an 18-character code and one in Hong Kong without.

A returned row can be valid and still belong to the wrong counterparty. Match the registered name and identifier across the licence, official record and deal documents before you pay.

3. Method, and what this study cannot say

• Frame: all 106 Chinese manufacturers in NHTSA’s vPIC trailer/vehicle set, the same frame as our earlier English-name lookup study, so the two are directly comparable. Every entry was searched — the full list and never a sample.
• Query: one search per input, by the registered English name, run from a mainland egress through a licensed Chinese business-information platform. Rows are recorded in the order the upstream returned them, before any ranking by this site.
• No paid detail record was retrieved. The question was what appears in the search layer before a paid detail record. This is also a limit: the search layer does not return the English name held on each registration record. So this study cannot say how many single rows were the correct company. It shows what one row is insufficient to establish.
• Branch classification is deterministic and never a judgement call: a Chinese branch registration name contains 分公司. The jurisdiction test is the registered province field as returned. The place-name test covers only the 43 single-candidate inputs that carry a recognised province or city word. Of those 43, four (9.3%) returned a jurisdiction that contradicts the place in the input, and all four were Hong Kong.
• A hypothesis that failed, reported as such. This study was designed to test whether search results are crowded by affiliated non-producing entities — investment, holding, supply-chain, consultancy, laboratory. The observation that prompted it came from one large industrial group. On this frame it did not reproduce: zero of the 23 additional rows carried that wording. The finding above is what the data supported instead.
• Retested 18 days later. On 6 September 2026 the same 106 names were searched again, from the same egress through the same platform. All 106 returned an identical candidate list, by name and by the order the rows came back. Nothing moved. The counts above are two measurements that agree, and no longer one day's snapshot.
• Scope limits: one industry (trailers and vehicles), one platform, two dates 18 days apart, one query per input. Return rates differ by sector and by how a name was romanised. Nothing here is a statement about any named company, and nothing here is legal advice.

What a later record check returned

The search study cannot tell how many single rows were the right company. A separate check on 21–22 August 2026 used codes for 45 records from the same list. The table counts records returned by five platform queries. These counts do not establish that a supplier passed due diligence.

A branch record does not prove a production address. Eight of the 45 codes returned branch records. The saved results record presence and counts; they do not compare branch addresses with a factory site.

These later queries used a licensed business-information platform. They are separate from the 18 August search dataset below. A no-record response counts as empty. It is a readable answer, but other sources may hold records.

Citing this study

Archived copies, each with its own DOI, resolving independently of this site: Zenodo · Harvard Dataverse.

A citation with everything it needs:

Currawong, “Is one search result confirmation?”, one search per registered English name for the full 106-company NHTSA vPIC frame from a mainland egress, 18 August 2026. https://currawongweb.com/verify/single-result-confirmation-study/ Dataset: https://doi.org/10.5281/zenodo.22004497

You are welcome to quote or reproduce these figures, including commercially, provided the date I checked it, the frame and the stated limits travel with them. Machine-readable evidence: the summary metrics and the per-input raw records, including every candidate row in the order it was returned. The 6 September 2026 retest is published beside them, as its own metrics and its own raw records, so the two runs can be compared line by line. The August files are unchanged, because their archived copies carry a DOI. Primary sources: NHTSA vPIC (frame; retrieved 9 August 2026), GB 32100-2015 (the 18-character code standard; checked 18 August 2026), Hong Kong FSTB on the eight-digit identifier and the government’s UBI explainer (both checked 18 August 2026), and GSXT, the official registry. More measured data: the research index.