Measured evidence for supplier decisions, free to cite with its dates.

For editors and researchersPublication-ready assets

Research media kit

Three findings are packaged below as 1200 × 675 graphics, with datasets, DOIs and corrected captions. The graphics and captions retain the sample, date and limits with each finding.

1. English-name fields matched for 47 of 106 inputs

Download PNG Download SVG

Frame: 106 China-listed trailer manufacturers in NHTSA vPIC, searched using the name held in that register. Taken: 12 August 2026; two same-platform runs seven hours apart reportedly returned the same counts.

Limit: Name-field equality does not confirm identity. This dated trailer group is not a population estimate or an upper bound for other suppliers. Interpretation and graphic labels corrected 8 September 2026.

Copy-ready caption:

Currawong recorded matching English-name fields for 47 of 106 inputs (44.3%) from a China-listed trailer manufacturer frame. The fields came from the top results on one business-information platform. Name equality does not confirm identity or predict results for other suppliers. Measured 12 August 2026.

Read the study and method · Download the CSV · DOI 10.5281/zenodo.21907920

2. Import-export wording appeared in 197 of 264 records

Frame labels corrected 8 September 2026. The SVG and PNG now state that the original provincial-list attribution remains unverified. Counts and the 8 August query date are unchanged.

Download PNG Download SVG

Frame: 264 names in the saved study frame: Shandong 136, Anhui 113 and Yunnan 15. The original notes attribute these to provincial ‘Little Giant’ lists; full verification against the originals remains open. These counts describe the saved frame, not all Chinese suppliers. Queried: 8 August 2026.

Limit: Scope wording does not prove current export capacity, required licences, product capability, or shipment history. Absence of the wording is a follow-up signal. It is not a verdict.

Copy-ready caption:

Currawong found import-export wording in 197 of 264 registrations (74.6%) in its saved 264-name frame. The original provincial-list attribution remains unverified. Registered scope wording is not proof of current export capacity. Queried 8 August 2026.

Read the study and method · Download the CSV · DOI 10.5281/zenodo.21858946

3. 145 of 264 records changed within 12 months

Frame labels corrected 8 September 2026. The SVG and PNG now state that the original provincial-list attribution remains unverified. Counts and the 8 August query date are unchanged.

Download PNG Download SVG

Frame: the same saved 264-name frame, using returned change dates and category labels. Its original provincial-list attribution remains unverified. Queried: 8 August 2026. Within the frame, 207 of 264 changed within 24 months and the median lifetime change count was 39.

Limit: A registered change is not itself a risk finding. The field that changed, its before-and-after facts and its relevance to the transaction determine whether it matters.

Copy-ready caption:

Currawong found that 145 of 264 registrations (54.9%) in the saved 264-name frame changed within the prior 12 months. The original provincial-list attribution remains unverified. A registered change is not itself a risk finding; the changed field and transaction context matter. Queried 8 August 2026.

Read the study and method · Download the CSV · DOI 10.5281/zenodo.21858946

What this desk has actually measured, counted on 22 August 2026

The studies below are the written-up ones. Underneath them sits a larger body of measurement that reaches the guides without ever becoming a study of its own. A fill-rate table sits on one page, a probe result on another. This is a count of the whole thing, taken mechanically from the pages themselves.

Thirteen distinct measurement sets, carried by 99 English pages, with query dates spanning 8–22 August 2026. Every table on this site states which set it came from and the date it was queried. A measurement without a date stops being evidence at some point, and nobody can tell when.

Two honest notes on this count. It is a mechanical inventory. We counted the evidence markers embedded in the pages, not the quality of what they show. And several sets are read along different dimensions on different pages: the 45-code lookup frame alone appears on 40 pages, which is one sample answering forty questions, not forty studies.

That distinction matters enough that every page carrying it says so in its own footnote. If you are citing any figure from this site, cite the underlying set and its query date rather than the page you found it on. Method for the largest set: the NHTSA manufacturer census.

Original study92 compared · 94 records

Alibaba vs 1688: Chinese-name fields 0/33 vs 59/59

If a profile does not give you a clear registered identity, ask for the business licence before payment. We compared the fields found in 92 profiles. The source records contain useful clues, but they do not confirm the seller.

Measured 28 August 2026 · Comparison: 59 on 1688 + 33 on Alibaba.com; controls and probes explained below · Prepared by Bao L. Zhou, Currawong’s China-side verification desk.

Platform subject fields: the numbers

The code key was found in all 59 sampled 1688 records. Of these, 43 values passed the format checks, 8 were 15-digit values, and 8 had no string value extracted. This does not establish why a value is missing. On the 33 Alibaba.com records, the searched keys for the Chinese name, company principal and code were marked absent. Other text or page views may still provide identity clues.

A code check tests the text. It cannot tell you who sent it. A real code can be copied onto a page. Match it to a dated record, then check the party you plan to pay.

What the badge comparison can tell you

A reasonable objection is that paid or independently assessed suppliers might carry more identity fields. They did not in this frame. Our Alibaba.com sample spans nine combinations of assessor and tier: SGS, TÜV Rheinland and no assessor at all; a “Verified” supplier badge, a “Verified” factory badge, and no badge. All 33 are identical on these fields.

The original notes describe factory data on assessed profiles, such as floor area, export revenue and certification numbers. Those fields do not by themselves identify the contract seller or payee. This comparison does not test the accuracy or scope of the assessment.

What this means at a desk

Use a profile name or code as a clue to check. Our parser read source fields; it did not test whether each field appeared in the visible page. Nor did it verify the company behind the profile.

Our separate English-name lookup study used 106 names from a vehicle regulator list. Its 44.3% matching rate does not estimate outcomes for these platform profiles.

Ask the supplier for the business licence. The registered name and 18-character code are both printed on it, and the check digit can be validated offline before you pay anyone.

Method, and what this study cannot say

There are 92 profiles in the comparison: 59 category profiles on 1688 and 33 on Alibaba.com. The latter include 30 category profiles and 3 earlier probes. The download holds 94 supplier profiles because it also keeps 2 control records from 1688. The 2 excluded shops listed separately could not be fetched by profile identifier; they are not those controls.

The six category searches were phone cases, copper alloy, LED lighting, knitted fabric, stainless-steel kitchenware and plastic packaging. The plan used five Alibaba.com profiles and ten 1688 profiles per category. The retained records and exclusions show what was obtained; the sample is not a survey of either platform.

The saved field labelled legal representative comes from the companyPrincipal key. We did not verify the person’s legal role. Treat this as a company-principal field, not proof of a registered legal representative.

The parser searched for named keys in page source. A matched non-empty string was present. A found key without a matched string value was empty; this includes blank strings and formats the parser did not match. A key not found was absent. This does not establish why a value is missing. Code candidates also faced a check-digit test and a six-digit administrative-division segment test; passing these tests does not verify a company.

One candidate passed the check digit by chance but failed the six-digit administrative-division segment test. That illustrates why a check digit alone is not identity evidence.

What it cannot say

• It is not a count of every supplier. This dated comparison includes category profiles and earlier probes. Platform layouts change; it is not a current platform-wide finding.
• The 1688 profiles were selected for a retrievable factory card. Suppliers without a card were not measured. The results do not set a bound for those suppliers.
• It says nothing about whether the data is correct. The 1688 source block carries provider and pass-date fields. We counted extracted fields, not whether the values are true or current.
• It does not evaluate the badges. We counted fields. Whether an SGS or TÜV assessment is thorough is a different question and not one this method touches.
• It is not reproducible unattended. Both platforms interrupted collection with slider challenges, which a person cleared by hand. We do not solve challenges automatically, and anyone repeating this will need to clear them too.

Citing the platform subject-field study

You may quote or reproduce these figures, including commercially, provided the measurement date, the frame and the stated limits travel with them. One row per profile, with the three-state reading and the check result for every code candidate, is in the result file (JSON).

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

Reading correction, 8 September 2026: the download and archived copies retain the original records and method wording. Use the definitions and sample breakdown above. The earlier wording inferred seller behaviour and platform policy from field states; those inferences are not supported. No profiles were collected again for this correction.

A citation with its scope:

Currawong, “Alibaba and 1688 identity fields: dated source-record comparison”. 92 compared profiles, drawn from 94 retained records across Alibaba.com and 1688, 28 August 2026. https://currawongweb.com/verify/alibaba-vs-1688-supplier-legal-name/ Dataset: https://doi.org/10.5281/zenodo.22136065

@dataset{currawong_platform_subject_fields_2026,
  author    = {Bao L. Zhou},
  title     = {{Subject-identity fields on Chinese supplier profiles: Alibaba.com and 1688 compared (94 profiles, August 2026)}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22136065},
  url       = {https://doi.org/10.5281/zenodo.22136065}
}

This study reports field presence, not a service claim. Currawong verifies public records on request and states sources, query dates and limits with every result; it does not certify suppliers.

Original study106 regulator records · 19 scored fields

Told to check the regulator’s record? Here is what is actually in it.

Use the named contacts as leads to check. We scored 19 fields in 106 vPIC records on 28 August 2026. Field presence does not prove that the details are correct or that a factory can make your product.

What the record holds, and how old it is

Median record age is 2.1 years. The oldest original submission on the frame dates from 11 December 2004, the newest from 8 July 2026; three records have the same submission and last-update date. This does not show whether their contents changed that day.

Different sources answer different questions

Our separate 264-company registry study found 54.9% amended their registration within twelve months. Here, 73.6% of 106 vPIC records had a stated update date over a year old. These are different groups, fields and dates. The figures do not set a recheck interval or rank the two sources.

Keep the source and date beside each fact. A vPIC contact can help you ask for current details; it does not confirm the Chinese legal entity or the payee. Check those against your order documents. A record date alone does not tell you whether a fact is still correct.

One more thing the frame shows about record quality: the province field is not normalised. Shandong appears as both Shandong and SHANDONG, Henan as both Henan and HENAN. Anyone counting by province from this source without folding case will split one province into two.

Method, and what this study cannot say

The frame is the 106 China-listed trailer manufacturers in our vPIC census of 9 August 2026. On 28 August, we made one request per ID to the public manufacturer-detail endpoint, with no retry and a quarter-second gap. All 106 returned a record. This is the whole fixed frame, not a sample of all Chinese suppliers.

We scored 19 fields. A missing key was absent. A null value, blank string or empty list was empty; other values were present. All scored keys were returned. These states do not explain why a value is missing. The download retains the states and selected values. Full API responses are not retained. Dates come from LastUpdated and SubmittedOn.

All 106 records had a non-empty VehicleTypes list, while PrimaryProduct was empty. A declared vehicle type is a clue to check, not proof of production capability. The 22 non-empty Mfr_CommonName fields were not checked against legal names.

What it cannot say

• A stale record is not evidence of anything wrong. Age measures time since the stated date. It does not explain why that date has not changed.
• We did not verify a single value. Presence of an address or a principal is not proof that either is current or correct.
• One regulator, one industry, one day. Trailer manufacturers in a US database. Other regulators keep different fields to different standards.
• “Last updated” is the regulator’s own field. We report what it says; we cannot see what changed, or whether an update was substantive.
• It says nothing about capability. That the record omits primary product is a fact about the record. It is not evidence that the company makes nothing, or anything in particular.

Citing the regulator-record study

Interpretation corrected 8 September 2026. This page now limits claims to the retained fields and dates. The original downloads are unchanged; no records were fetched again for this correction.

You may quote or reproduce these figures, including commercially, provided the measurement date, the frame and the stated limits travel with them. One row per manufacturer, with the three-state reading for all 19 scored fields and both dates, is in the result file (JSON).

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

A citation with everything it needs:

Currawong, “Told to check the regulator’s record? Here is what is actually in it”, 19 scored fields and record dates for 106 China-listed trailer manufacturers in NHTSA vPIC, 28 August 2026. https://currawongweb.com/research/#vpic-record-age Dataset: https://doi.org/10.5281/zenodo.22144964

@dataset{currawong_vpic_manufacturer_records_2026,
  author    = {Bao L. Zhou},
  title     = {{What a US vehicle regulator holds on 106 China-registered manufacturers, and when it was last updated}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22144964},
  url       = {https://doi.org/10.5281/zenodo.22144964}
}

This study reports what a public regulator record contains and when it was last updated. It is not an assessment of any manufacturer, and Currawong does not certify suppliers.

Original study264 enforcement records · 46 firm names

FDA: 264 records, 80 recall events, 46 Chinese firm names

A recall record should prompt a case-specific review, not an automatic supplier verdict. This study covers every device enforcement record the US regulator returned for a recalling firm located in China: 264 records. We counted what the public corrective-action trail shows.

Firm names and status labels in the saved records

Among records labelled Ongoing, the median time from initiation to 28 August 2026 was 875 days; the longest was 3,600. These are date differences, not measured times to closure. Initiations run from 8 June 2012 to 21 April 2026, with 43.2% of the whole frame initiated during 2024 alone.

Re-run on 7 September 2026 with the same query and the same de-duplication. Fourteen fixed measures matched. The two elapsed-time measures increased by ten days. This aggregate comparison does not prove that every underlying record was unchanged. The reproduction file is fda-china-device-enforcement-2026-09-07.json.

What to check before drawing a conclusion

A recall concerns a product problem and the action taken to address it. It does not settle the quality of a supplier or every product it sells. Nor does absence from this dataset show that a supplier is safe. Match the product and reason to your order, then check the current case information.

The 23 firm names appearing more than once account for 241 records (91.3%). But only 9 firm names have more than one event ID. The name with 88 records has one event ID. Record counts alone cannot establish repeated incidents.

openFDA warns against using these reports to track a recall’s lifecycle: status is not updated after publication. The 219 Ongoing labels are stored values. They do not show that 219 cases remain open today or how long correction took.

Method, and what this study cannot say

We fetched all rows for country:"China" from openFDA’s device enforcement endpoint. The row count matched its reported total. No sampling or retry. Country means where the recalling firm is located, not where goods were made or sold. We parsed the YYYYMMDD dates and grouped firm names after trimming spaces and ignoring case.

What it cannot say

• It is not a supplier ranking, and we publish none. The record file carries the firm names the regulator publishes; this page reports distributions only.
• Absence proves nothing. Not appearing in this dated device-recall frame is not a clean bill of health.
• Devices only. One regulator, medical devices only. Other sectors have different records and different reporting cultures.
• Ongoing is a filing state, not a safety state. We report the saved label; it cannot establish the current case status or whether the issue was fixed.
• Names are not verified entities. Two spellings can split one firm; a shared name can combine firms. We did not resolve these cases.

Citing the enforcement-record study

You may quote or reproduce these figures, including commercially, provided the measurement date, the frame and the stated limits travel with them. The CSV contains aggregate measures. The JSON contains one row per enforcement record, including its event ID, classification, stored status, dates and stated reason. Download the result file (JSON).

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

Interpretation corrected 8 September 2026. We withdraw the claims about time still open and repeated incidents. The saved JSON files and external archives keep their old labels. Read them with this correction. Raw data and the 28 August query date are unchanged. See the openFDA field reference.

A citation with everything it needs:

Currawong, “When it goes wrong in the US market, what does the record show?”, all 264 openFDA device enforcement records for recalling firms located in China, 28 August 2026. https://currawongweb.com/research/#fda-enforcement Dataset: https://doi.org/10.5281/zenodo.22144974

@dataset{currawong_fda_china_device_enforcement_2026,
  author    = {Bao L. Zhou},
  title     = {{US device enforcement records for recalling firms located in China: concentration, status and age}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22144974},
  url       = {https://doi.org/10.5281/zenodo.22144974}
}

This study reports the shape of a public enforcement record. It is not an assessment of any manufacturer, and Currawong does not certify suppliers.

Original study46 query inputs · three fields

Capital, status and scope: what the returned fields show

A large capital figure, active status and manufacturing wording do not clear a supplier. We checked what a commercial platform returned for 46 selected entries. The results show field values, not bank balances or factory capacity.

Capital fields do not show a current cash balance

A capital field is a recorded amount. It does not show how much cash is available today or verify a contribution. Company type and dates matter when reading the legal rules; see the study’s official sources.

Subscribed capital on this frame was numeric and labelled CNY in 44 of 46 entries. It ranged from ¥100,000 to ¥1,023,855,833, with a median of ¥9,000,000. Two entries had no parsed capital value. Lower paid-in values and empty fields are separate findings.

What differs between the two groups

The 17 entries with an empty paid-in field had an average founding year of 2017. Their median subscribed value was ¥3,000,000, using the 15 numeric CNY amounts. The 29 entries with a paid-in value averaged 2011, with a median subscribed value of ¥20,000,000. These differences do not explain why a field was empty.

Our earlier study used a different group of 264 manufacturers. Its results should be read with its own frame and method. This comparison does not show that age, size or a legal deadline caused the difference in returned fields.

Status is a state, not a grade, and it is written three ways

The platform returned 存续 for 26 entries, 开业 for 19 and 仍注册 for one. These are the stored labels. We did not confirm each entity’s identity or current status against an official source for this study.

The records contain 11 company-type strings. Folding full-width and half-width brackets leaves ten. The set includes a branch and other entity types; one legal rule cannot be applied to the whole set without checking which entity each record describes.

Read the status together with its source and date. A platform label does not establish solvency, trading history or delivery capacity.

Scope wording does not verify factory capacity

The stored flags show both production and trade terms for 39 of 46 entries. Six matched trade terms alone; one matched neither. We have no verified factory/trader labels against which to test these groups. The counts do not show how well scope classifies a supplier.

The stored scope lengths have a median of 193 characters across all 46 entries, including one zero-length value; the longest is 591. A licensed-project or general-project phrase matched in 30 of 46 entries (65.2%). No time series or full scope text was retained, so this does not measure a trend in wording.

Method, and what this study cannot say

The frame came from earlier searches with one candidate and a non-empty code field. There were 45 18-character strings and one 8-character string. The filter did not validate codes or confirm identities. Two commercial-platform interfaces were queried once per input: base information and shareholder contributions. No retry.

The parser recorded the platform’s no-result code as no-record and a success code as present. For shareholder entries it read one page, requested at a size of 50. Empty paid-in values stayed separate from numeric zero. Original responses were not retained, so the stored fields cannot establish complete official coverage or why a value was missing.

What it cannot say

• An empty field is not proof of non-payment. This is a platform observation, not a finding about what a company paid or published.
• Nothing here evaluates a supplier. These are field-level properties of a record, not judgements about any company.
• One platform, one day. The data comes from a licensed Chinese business-information platform, not from the government portal directly, and platform coverage is not identical to official coverage.
• A selected lookup frame. The inputs came from a trailer-manufacturer lookup, not a representative sample of Chinese suppliers.
• Keywords do not prove an activity. The stored flags cannot confirm a factory, a product licence or work done for an order.

Citing the registration-field study

Correction, 8 September 2026: we corrected the input description, median calculation and interpretation. The original downloads and archive title remain unchanged. Their capital median of 10 million CNY and scope-length median of 194 use the upper middle value; the usual medians are 9 million and 193. No queries were rerun.

You may quote or reproduce these figures, including commercially, provided the measurement date, the frame and the stated limits travel with them. The JSON contains one row per input; the CSV contains summary metrics. These include the parsed capital figures, status, company type and the scope-wording flags. Download the result file (JSON). Natural-person names are not stored: the legal representative and shareholder fields appear as presence and counts only.

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

A citation with everything it needs:

Currawong, “China scope records: 0 of 46 mention manufacturing without trade wording”, returned fields for 46 selected commercial-platform query inputs, 28 August 2026. https://currawongweb.com/verify/registered-capital-business-scope-china/ Dataset: https://doi.org/10.5281/zenodo.22145056

@dataset{currawong_registration_field_meaning_2026,
  author    = {Bao L. Zhou},
  title     = {{What Chinese registry fields guarantee: paid-in capital, registration status and business scope across 46 manufacturers}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22145056},
  url       = {https://doi.org/10.5281/zenodo.22145056}
}

This study reports properties of registry fields. It is not an assessment of any company, and Currawong does not certify suppliers.

Original studysix record types · 46 query inputs

An empty risk screen does not clear a supplier.

Check what each result says and which source returned it. On 28 August 2026 we queried six record types for 46 entries from an earlier manufacturer lookup. This measures platform responses. It does not measure supplier safety or the best order of checks.

What the platform returned

Twenty-eight entries (60.9%) were classified as no-record across all six types. Of the other 18, 14 had one type present and 4 had two. Among entries with hearing notices, the stored count had a median of 3 and a maximum of 52. The original responses were not retained, so the file cannot show whether each count was a total or a returned page count.

How to use these results

Read the result with its source and date. A platform returning no record does not establish that the official source has none. A returned record needs its details checked; its presence alone does not settle the question about your supplier.

Frequency does not set priority. Hearing notices were the most common result in this file. We did not measure whether they helped buyers more than other records. Choose checks for the question you need to resolve, rather than for how often they returned data in this group.

These counts do not predict recovery. We stored no amounts, case outcomes or transaction results. The data cannot tell you whether a creditor recovered money, whether a company can pay, or what would happen in your dispute.

Method, and what this study cannot say

Six company-level interfaces on a commercial business-information platform were queried once per entry. The 46 entries came from earlier searches that returned one candidate with a non-empty code field.

45 query strings were 18 characters long; one was an 8-character string. That entry returned no-record for all six types. Length alone does not verify the other entries. We retain the original denominator of 46 rather than silently changing the historical results.

Only counts and presence were stored. No case numbers, parties, amounts or individual names were recorded, and no person-level interface was queried. All 276 stored results were classified as present or no-record; none was marked unreadable.

The parser treated the platform's no-result code and a successful null result as no-record. It used a total when supplied; otherwise it counted returned array entries. An empty top-level array would be labelled present with a count of zero, although that combination does not occur in this file. Without the original responses, we cannot recheck which response shape produced each stored result.

The original write-up also cited two positive-control queries. Their responses are not included in this dataset, so those counts cannot be independently checked from this download. A successful control would not prove that every other result was complete.

What it cannot say

• Zero of 46 does not mean zero in the population. This selected group does not establish a population rate. Do not read it as "this never happens".
• The platform is not the official register. A platform holding no record is not proof the government register holds none.
• Absence never clears a supplier. The returned data do not establish supplier safety.
• One industry, one day. The frame came from a trailer-manufacturer lookup. It is not a representative sample of Chinese suppliers.
• A hearing notice is not a verdict. It records that a case was scheduled, not who was right, and not whether the company was plaintiff or defendant.

Citing the base-rate study

Correction, 8 September 2026: we corrected the query-input description and removed conclusions about check priority and recovery. No queries were rerun. The original downloads and archived title below remain unchanged; read them with the corrected method and limits above.

You may quote or reproduce these figures, including commercially, provided the measurement date, the frame and the stated limits travel with them. One row per queried entry, with the six per-signal states and counts, is in the result file (JSON). No case details or individual names appear in either file.

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

A citation with everything it needs:

Currawong, “An empty risk screen does not clear a supplier”, six commercial-platform record types across 46 selected query inputs, 28 August 2026. https://currawongweb.com/research/#risk-signal-base-rate Dataset: https://doi.org/10.5281/zenodo.22145063

@dataset{currawong_public_risk_signal_base_rate_2026,
  author    = {Bao L. Zhou},
  title     = {{Base rate of six public risk signals across 46 Chinese manufacturers}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22145063},
  url       = {https://doi.org/10.5281/zenodo.22145063}
}

This study reports how often six platform record types were returned for the selected query inputs. It is not an assessment of any company, and Currawong does not certify suppliers.

Original measurement106 companies

Can you find a Chinese supplier from its English name?

Use an English name to find candidates, then check the Chinese registered name and business licence. On 12 August 2026, we compared search results for 106 China-listed trailer makers from NHTSA vPIC. A matching English-name field was a lead to check, not proof of identity.

Measured 12 August 2026 · Frame: 106 China-listed trailer manufacturers in NHTSA vPIC · Prepared by Bao L. Zhou, Currawong’s China-side verification desk.

English-name lookup: the numbers

We searched all 106 names in our 9 August 2026 list of China-based trailer manufacturers in NHTSA vPIC. The wider list had 2,589 China-based makers. This search test used the 106 trailer makers. It does not cover all Chinese suppliers.

The match links your input to text in a search result. The public file does not show a separate identity check for each result. An empty field or a name that differs also does not prove that it is the wrong firm.

If a result lacks a code, ask for the Chinese registered name and business licence. You may still be able to check the firm using that name.

The brand-name trap

Twenty-two entries carried both a full name and a brand or short name. Searching the same group both ways gave these return counts:

Brand inputs returned more results in this group. The first write-up gave examples of wrong firms. The public CSV does not list those results or count how many were wrong. It cannot show how often a brand search found the wrong firm.

Short names can refer to more than one firm. Check who each result refers to. This test did not measure why brand searches returned more results.

More returned results do not establish more correct matches. Keep search return counts separate from identity checks.

What this means at a desk

• Start with a candidate. The 47 matching fields in this group were leads for further checks. They were not 47 verified supplier identities.
• Check the first result before using it. Search rank does not establish identity.
• Ask for the business licence (营业执照). It carries the registered Chinese name and 18-character code, turning a search problem into a lookup.
• A refusal is evidence too. A supplier unwilling to provide a business-licence photo has not proved fraud. It has blocked the shortest route to confirming the entity, and should never be silently treated as a clean result.
• Compare the transaction parties. Match the registered name and code against the licence, dated records and order documents. An English-name match does not establish who is contacting you.

Method, and what this study cannot say

Each name from vPIC was submitted to a licensed Chinese business-information platform from a mainland network egress. This reports the dated workflow used for the study; it does not establish access from other locations today.

For searches with candidates, we retrieved the top result from the same platform and compared its English-name field with the input. Normalisation used uppercase text with non-alphanumeric characters removed. This tests text equality, not whether two records refer to the same legal entity.

The study reports two runs seven hours apart with the same totals. Both used the same platform. This is not a check against a separate source. Equal totals do not prove that each result stayed the same or was correct.

What it cannot say

• It is one dated group. The 106 trailer makers do not stand for all Chinese exporters. This test gives no upper bound for other sectors.
• The paired comparison has 22 entries. It counts search returns. The public file does not say how many firms were checked against a separate source.
• Other name sources were not measured. These results do not predict the success rate for Alibaba storefront names or names from other sectors.
• A matching English name confirms nothing else. Status, scope, export capability and counterparty identity are separate checks.
• One platform, one date. Coverage differs between providers and changes over time. The measurement date travels with the numbers.

English-name lookup questions

Citing the English-name study

Wording corrected 8 September 2026. The values and query date are unchanged. Older archives say “confirmable”. This means the name text matched; it does not mean the firm’s identity was proved. The public summary does not include each query response.

You may quote or reproduce these figures, including commercially, provided the measurement date, frame and limits travel with them. Machine-readable metrics with numerators and denominators are in the result file. The frame comes from our census of Chinese manufacturers in NHTSA vPIC, which readers can re-derive from the vPIC API.

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

A citation with everything it needs:

Currawong, “Can you find a Chinese supplier from its English name?”. 106 China-listed trailer manufacturers from NHTSA vPIC searched by the name held in that register from a mainland egress, two same-platform runs, 12 August 2026. https://currawongweb.com/verify/find-chinese-company-by-english-name/ Dataset: https://doi.org/10.5281/zenodo.21907920

@dataset{currawong_english_name_lookup_2026,
  author    = {Bao L. Zhou},
  title     = {{English-name lookup rates for Chinese manufacturers (NHTSA vPIC frame, August 2026)}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21907920},
  url       = {https://doi.org/10.5281/zenodo.21907920}
}

This study reports rates, not a service claim. Currawong verifies public records on request and states sources, query dates and limits with every result; it does not certify suppliers.

Engineering practice, not a buyer decision98 strings · 15 languages

One model translated. Another one checked.

A second model flagged 11 of 98 strings after they passed the text checks. We then edited eight strings and kept three with notes on the choice. The flags do not tell us the true error rate.

Reviewed 12 August 2026 · Frame: 98 interface strings across 15 languages, two rounds · Prepared by Bao L. Zhou, Currawong’s China-side verification desk.

Machine-translation review: the numbers

Flags occurred in six of the 14 non-English languages: Arabic, Vietnamese, Hungarian, Russian, Polish and Portuguese. Eight languages had no flags. English was also included in the 15-language frame. These small counts do not rank language quality or establish that unflagged text was correct.

The document-term flag

One reviewed string asked a buyer to request a photo of the Chinese business licence (营业执照) when a search returned nothing. The licence was the intended document.

The reviewer interpreted the Arabic wording as “work permit”. The dated summary classified this as the one high-severity flag.

If the reviewer was right, a buyer could ask for the wrong document. We did not test that with buyers or measure harm. The text passed the character and length checks. Those checks did not settle what the term meant.

The write-up also raised doubts about Hungarian and Polish terms. The public file gives totals. It does not show the strings or a native speaker’s verdict. Readers cannot check each language claim from this file alone.

Why the automated gates missed all eleven

The recorded workflow used four checks before model review:

• Completeness: every string present in every language, so no reader silently falls back to English.
• Script and diacritics: check the expected characters and accents.
• Forbidden wordings: a per-language list of claims we must never make. Examples: calling an unconfirmed candidate a verified supplier, or calling an outage a company that does not exist.
• Length: apply a per-string budget. Rendered layout still needs a browser check.

The original write-up records two failure probes: stripped Turkish diacritics and an injected forbidden phrase caused their respective checks to fail. These probes test those rules, not translation accuracy.

The rules did not flag these 11 strings. That does not mean code can never find the same issues. A new rule could check a known term. That rule would still need to be tested.

Keep text checks and meaning review separate. A string can pass the configured rules and still need its meaning checked.

Method, and what this study cannot say

One model translated the text. A different model received each string and its intended English meaning, then back-translated and flagged concerns. It did not see the first model’s reasoning. We classified the flags and recorded corrections or accepted drift.

A different model gives a second pass. Both models may share the same mistakes. The reviewer saw the intended meaning, so this was not a blind test or a check against a trusted translation.

The 12 August record says we changed eight strings and kept three with notes. For one status term, the suggested new word had not been checked. These counts describe that day’s work. They do not show what the live site says today.

What it cannot say

• It is not a benchmark. 98 short, domain-specific interface strings, one translating model, one reviewing model, one review date. It measures a workflow, not machine translation as a field.
• Neither model is a native speaker. The reviewer flagged what it could see; it also stated plainly that it could not find authoritative sources for some phrases it was judging. Its flags are findings to investigate, not verdicts.
• 11.2% is not a verified error rate. The reviewer could miss errors or flag words that were fine. There was no trusted reference for every string. The count gives neither a lower nor an upper bound on the true error rate.
• The per-language split proves nothing. Six languages with flags and eight without, at one to three flags each, is not evidence about those languages.
• This covers our own text. The review did not include native-speaker validation. The editorial policy explains the machine-assisted language boundary.

Citing the machine-translation review

Wording corrected 8 September 2026. Counts and the review date are unchanged. “Independent” in the archive title is the name of the old deposit. It does not prove that the models made errors on their own. The public CSV gives totals; it does not list each string.

You may quote or reproduce these figures, including commercially, provided the review date, the frame and the stated limits travel with them. Machine-readable metrics with numerators and denominators are in the result file.

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

A citation with everything it needs:

Currawong, “One model translated, another checked: 11 of 98 flagged”, second-model review of 98 interface strings across 15 languages in two rounds, 12 August 2026. https://currawongweb.com/research/#machine-translation-review Dataset: https://doi.org/10.5281/zenodo.21908377

@dataset{currawong_machine_translation_review_2026,
  author    = {Bao L. Zhou},
  title     = {{Independent back-translation review of machine-translated interface text (98 strings, 15 languages, August 2026)}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21908377},
  url       = {https://doi.org/10.5281/zenodo.21908377}
}

This study reports what one review found in our own interface text. It is not an assessment of any named translation product, and it is not advice on whether to use machine translation for regulated content.

Original study50,248 observations · 621 probes · 32 countries

Told to check the official registry: can a buyer outside China reach it at all?

Buyers are routinely told to confirm a Chinese supplier against the official registry before paying. That advice quietly assumes the portals answer from wherever the buyer happens to be. We measured whether they answer at the network layer: eight official portals and two controls, from 621 RIPE Atlas probes in 32 countries outside mainland China, continuously for fourteen days.

What answered, and what did not

DNS and TLS put the official portals within a few points of two hosts nobody claims are hard to reach. The ping column is the one that looks alarming, and it is the one that misleads.

Three hosts answered no ping at all: 0 of 4,974 attempts. They are both court portals and the national standards portal. Over the same fourteen days, from the same probes, those three kept completing the TLS handshake, at the per-host rates in the table above. The absent ping is an ICMP policy on the host or its path. It is not evidence that anyone cannot reach them.

This matters beyond these three hosts. A reachability check built on ping alone would have reported three of eight official portals as unreachable while their TLS listeners were answering throughout. Anyone quoting a ping-based figure about Chinese government hosts should say which layer it came from.

Method, and what this study cannot say

Thirty-one periodic RIPE Atlas measurements ran over ten hosts, eight official portals and two controls. They used the ping, TLS certificate and DNS measurement types, from 26 August 2026 at 11:29 UTC to 9 September 2026 at 05:42 UTC. 50,248 result rows from 621 probes in 32 countries, none of them in mainland China. Every row carries its measurement id, probe id, probe country and timestamp. The measurement definitions stay public, so anyone can re-read them from the RIPE Atlas API and recompute every figure above.

Success is defined per measurement type and stated in the file: at least one packet received for ping, a completed handshake for TLS, at least one answer for DNS. For DNS rows the queried hostname is joined in from the measurement definition, because a RIPE Atlas DNS result does not repeat it. No content was fetched from any portal. These are ping, TLS-handshake and DNS measurements only. Nothing here was obtained by scraping, proxying, or working around an access control.

What it cannot say

• Reachable is not usable. This says nothing about whether a page renders, whether a search returns a result, or whether a CAPTCHA or a login stops a person. Our separate browser-level measurement of the same portals found most front pages refused on its test date. That is a different layer and a different method. The two figures must never be quoted as one.
• It does not establish intent. A failed observation may reflect routing, filtering, server-side rate limiting, conditions at the probe, or an ordinary outage. We report what the probes saw.
• The probes are not a sample of anything. RIPE Atlas probes are largely volunteer-hosted on residential connections. Placement is not random, and no country-level statistic should be read out of this file, however tempting the 32-country spread looks.
• It expires. Measurements stopped on 9 September 2026. Reachability changes; every row carries its own timestamp for that reason.
• Absent ICMP is not absent service, as three of these eight portals demonstrate. Nor is the reverse true: a completed handshake does not mean the service behind it will answer a query.

Citing the network-reachability study

You may quote or reproduce these figures, including commercially, provided the measurement window, the layer measured and the stated limits travel with them. Every observation, one row each, is in the CSV (JSON, which also carries the per-host summary the tables above are computed from).

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

A citation with everything it needs:

Currawong, “Told to check the official registry: can a buyer outside China reach it at all?”, 26 August to 9 September 2026. Network-layer reachability of eight Chinese official portals, measured from 621 RIPE Atlas probes in 32 countries. https://currawongweb.com/research/#atlas-network-reachability

@dataset{currawong_china_portal_network_reachability_2026,
  author    = {Bao L. Zhou},
  title     = {{Network-layer reachability of eight Chinese official due-diligence portals from outside mainland China (RIPE Atlas, 26 August to 9 September 2026)}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22681948},
  url       = {https://doi.org/10.5281/zenodo.22681948}
}

Measurements were funded by RIPE Atlas research credits under ticket #1094691; publishing the resulting dataset was a condition of that grant. This study reports network-layer observations. It is not an assessment of any portal operator, and it does not show that any portal is usable or unusable for a given task.