Your saved supplier record can go out of date. I measured how short.

Before relying on a saved supplier record

Check the live record and its query date. The measured change rates below explain why a screenshot has a short shelf life.

1. The record moves — for most companies, within the year

The 12-month rate is consistent across all three provincial excerpts — Shandong 52.2%, Anhui 56.6%, Yunnan 66.7%, so this is not one province’s paperwork habit. These are stable, state-vetted manufacturers with a median age of about 15 years, and even their records move this fast.

2. What moved in the last 12 months

These are not clerical trivia. The registered capital moving in nearly one company in five fits the post-2024 pay-in timetable. The scope moving in one in six changes the first free signal buyers read. And the legal representative moving in one in nine means the name on last year’s licence photo may simply no longer be the registered one. The most frequent recent entries overall were articles-of-association and senior-management filings — routine governance, which is exactly the point: routine governance keeps rewriting the record.

3. What this does to a screenshot

A saved screenshot, an old licence photo or a colleague’s notes describe the company as it was on the capture date. With more than half of even elite companies amending their registration inside 12 months, an undated capture is not a slightly stale record. It is a coin flip on whether something material has moved since. This is the measured version of the rule every check on this site already applies. A change is a dated fact, and only the live record, read with its query date, describes the company you are paying today.

Method: the same census, one more variable

• The frame. Identical to the registration study: all 264 companies on three official provincial-government excerpts of MIIT’s sixth “Little Giant” batch (Shandong 136, Anhui 113, Yunnan 15). A census of the frame: no sampling within it.
• The lookups. Each company’s registered-change log was queried once through a licensed commercial registry-data API on 8 August 2026, from our China-side desk. We recorded the registry’s lifetime change total and, for each company, the dates and category labels of its twenty most recent entries.
• Privacy at collection. Change entries carry before-and-after content that can contain personal names. Those content fields were discarded at collection. The dataset holds change dates and category labels only, consistent with China’s personal-information rules and our editorial policy.
• Field-level counts are lower bounds. The per-field 12-month numbers are computed within each company’s twenty most recent entries. A company that logged more than twenty changes in twelve months would have older entries pushed out of that window. So the true shares can only be equal or higher.

What these numbers do not say

• Change is not distress. Nothing here says these companies are unstable — most recent entries are routine governance filings. The finding is about the record’s freshness. It says nothing about the companies’ health. A change is a dated fact. It is not a supplier verdict.
• Not a picture of marketplace suppliers. This is an elite, state-vetted manufacturer frame. Whether marketplace suppliers’ records move faster or slower is not measured here.
• Not national. Three provinces of thirty-one, as in the registration study. The cross-province consistency is reassuring, not conclusive.
• One day, one source. All lookups are dated 8 August 2026 through one licensed data channel, and category labels follow that channel’s taxonomy.

What a buyer does with this

1. Date every capture. A record without its query date is unusable a year later, and half of these companies changed something within a year.
2. Re-read the live record at each decision gate, shortlist, order, payment, rather than reusing the file from the last one. The official search steps show what to compare.
3. When a supplier sends an old document, ask which version was current at the step that matters, and check the current record yourself — or have it read China-side, with the source and query date named per line.

If the code is already in front of you, that re-read takes a minute. Pull the current registered record, dated today → The free lookup takes the 18-character code directly and matches on the code rather than on a name, which sidesteps the name-matching problem entirely. If that code came off a document nobody has checked, validate its structure offline first →

The records behind these percentages sit in the GSXT (National Enterprise Credit Information Publicity System). That is why any attempt to verify a Chinese company on a months-old capture is really a bet against the 54.9% base rate above.

A recheck cadence that matches the base rate

The census fixes two points on the curve — 54.9% amended within 12 months, 78.4% within 24, and a median of 320 days since the most recent change. What those numbers say, plainly: a saved record older than a year describes a company that, on this frame, had even odds of having amended something since. What they do not say is a magic interval. Any fixed cadence is a rule of thumb built on two measured points. It is not a measurement itself.

• Re-read at every decision gate. Shortlist, order, payment: each gate deserves its own dated read. The fields that moved most often in the last 12 months are exactly the ones buyers rely on. Registered capital (18.9% of companies), business scope (15.5%) and the legal representative (11.7%). The licence photo from last year may name a legal representative who no longer is one.
• Between decisions, keep a calendar. For an ongoing supplier relationship we work on a six-month re-read, half of the measured 12-month window, as a rule of thumb, not as a measured interval. The honest summary of the data is directional: the older the capture, the worse the odds. If your risk tolerance is lower, shorten the interval.
• Re-read immediately on a trigger. New bank details, a changed beneficiary name, a renamed entity or a new signing contact are the four events that move money. A changed beneficiary is the one you can screen in minutes: paste the contract, invoice and beneficiary names into the payment party consistency checker. It runs in your browser and uploads nothing — then confirm the entity behind the winning name.

The re-read itself is the cheap part. Pull the current registered record, dated today →, the free lookup takes the 18-character code directly and matches on the code. It does not match on the name. When the question needs a person reading the official portals, the L3 record check returns each line with its source and query date.

A recheck cadence lowers the chance of relying on a stale record. It does not make any record current, and no interval turns a dated snapshot into a guarantee about tomorrow.

A second group, measured differently, pointing the same way

This study tracks how often records change between two capture dates. On 21–22 August 2026 we approached the same question from the other end on a different group. Instead of watching a set over time, we asked 45 Chinese manufacturers how much change history each one already carries.

Nine established manufacturers in ten already carry a change history. That is consistent with this study’s finding and adds something it could not show: change is not an event that happens to a minority of suppliers between two snapshots. It is the accumulated normal state of almost all of them.

One limit applies to the second group’s counts. Those lists are returned paginated at 20, and several companies hit that figure, so “has change history” is reliable while any specific count is a floor. It also cannot date the changes, which is precisely what this study’s two-capture method does and this one does not.

Citing this study

Quote or reproduce these figures freely, including commercially, provided you keep the query date (8 August 2026) and the frame (264 companies, three provincial excerpts of MIIT’s sixth “Little Giant” batch) and the stated limits travel with them.

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

Machine-readable copy: the published headline aggregates as CSV — aggregate counts and shares only, one row per published measure, with the frame and query date on every row. No per-company records are published.

Currawong, “How often Chinese company records change”, census of registered-change logs for 264 companies on three official provincial excerpts of MIIT’s sixth ‘Little Giant’ batch, queried 8 August 2026. https://currawongweb.com/verify/china-company-record-change-study/ Dataset: https://doi.org/10.5281/zenodo.21858946

@dataset{currawong_verify_china_company_record_change_study_2026,
  author    = {Bao L. Zhou},
  title     = {{Registered-change velocity of 264 'Little Giant' manufacturers}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21858946},
  url       = {https://doi.org/10.5281/zenodo.21858946}
}

If you re-run this on another frame or date and get different numbers, we want to hear it. Corrections that survive checking get published here with attribution, including ones that contradict us.

Browse all measured studies and methods in the research index.

This page reports aggregate change-log statistics. It is not legal advice, and it says nothing about any individual company’s stability, honesty or suitability.