Methodology and limitations

What is measured

Every company-registry search run through LemReveal's free tools is logged as an aggregate record: the country searched, whether a record came back, how long it took, and when. No query text, no IP address and nothing identifying a searcher is stored, so this dataset cannot be traced to any individual and contains no personal data.

Version 2026-09-12 covers 9,459 searches across 54 countries between 2026-08-26 and 2026-09-12. A country needs at least 20 recorded searches in the window to appear; below that the percentage is noise and we leave it out rather than publish it.

Capability data

Alongside the live measurement, every register we connect to is automatically re-probed once a day with a known, real company for that country. The probe records whether the register returned a usable record, returned nothing, or failed outright, plus the exact error. Those verdicts drive both the capability columns here and the public registry status pages.

Registry Reliability Observatory

The Global Company Registry Reliability Observatory uses those scheduled known-company probes, not the real-user search observations used by the Registry Access Index. For each registry and UTC day, it uses the final completed probe recorded that day. A usable record is a success; a degraded answer or failed answer is an incident day. A missing probe day is shown as missing rather than treated as downtime.

A route answered only by a global fallback is a coverage gap, not a successful national-register probe and not a failure attributed to the national register. Coverage gaps and internal scheduling exclusions are omitted from success-rate denominators. The current schedule is daily; where a registry has fewer than 25 valid observations in 30 days or 75 in 90 days, we say insufficient observation history rather than rank it.

One patient retry may be made when LemReveal's own throttle answers. That retry is folded into the end-to-end probe duration; retries are not stored as separate public days. Duration can include connector and fallback processing, so it is never described as direct registry response time.

The Observatory is versioned by its UTC data cutoff. The append-only history supports dated API views, and the JSON and CSV include the version, cutoff, outcome taxonomy and sufficient fields to reproduce the published success-rate denominators. If a seed-company identifier changes or a classification needs correction, we preserve the prior recorded observation and document the corrected interpretation in a later dated dataset rather than silently rewriting a headline.

Recomputable is not the same as independently repeatable. The published JSON and CSV let a reader recompute the displayed totals and rates from the released observations. They do not expose every seed-company identifier, connector implementation detail or operational note required to repeat each probe independently. Those details are withheld to protect the tested routes and prevent the monitoring setup from becoming an abuse guide.

The unit of analysis is a tested company-record route. A row can be a national register route, a specialist official source, or an aggregated route. It identifies the route tested by LemReveal; it should not be read as a claim that every row is the complete company- registration system for its country.

How the score is calculated

The score is out of 100 and deliberately simple, so anyone can recompute it from the published CSV and disagree with the weights in the open.

• Measured search success rate — up to 45 points. Share of real user searches that returned a record, scaled to 45 points. This is the only component based on live measurement rather than capability.
• Free public search that works — up to 20 points. 20 points where we have a working free search route into the register, 10 where the only route is an aggregator or open-data mirror, 0 where nothing free answers.
• Official source — up to 15 points. 15 points where the answer comes from the government register or its official open data, 7 for an official bulk mirror, 0 for third-party aggregation.
• Directors and officers published — up to 10 points. 10 points where the free route exposes officer or director names.
• Filings or documents — up to 5 points. 5 points where accounts or filed documents are reachable free of charge.
• Change monitoring possible — up to 5 points. 5 points where the register can be re-checked programmatically for changes.

Bands: 60 and above is open, 30 to 59 is partially open, below 30 is effectively closed.

Limitations — read these before quoting the numbers

• The sample is our traffic, not the world's. These are searches from people who came to LemReveal. A country with heavy demand here is not necessarily a country with heavy demand globally.
• A failed search is not always the register's fault. Misspelled names, companies that never existed, and entities dissolved decades ago all count as empty results. The failure rate is an upper bound on registry inaccessibility, not a precise measure of it.
• Access is measured through free public routes only. Several registers that score poorly here are perfectly usable if you pay, register an account, or visit in person. The index measures free public access, which is the thing that is unevenly distributed.
• The window is short. Version 2026-09-12 covers under three weeks. A register that happened to be down during the window is penalised for it. Later versions will cover longer periods and we will publish each version separately rather than silently restating old numbers.
• Response times include our own fallback chain. A slow mean response usually means several sources were tried, not that the register itself is slow.

Corrections

If you operate a register listed here and the capability data is wrong, tell us and we will correct it in the next version with a note recording the change. We would rather be corrected than be quoted incorrectly.

Licence

The index, the per-country pages and both downloads are published under CC BY 4.0. Reuse it commercially, republish it, build on it — the only condition is crediting LemReveal with a link.