Brand resolution reaches general availability
58.4M marks from EUIPO TMview, WIPO GBD and USPTO TSDR are now first-class entities with brd_ IDs, linked to their proprietor. Resolving a trade name no longer silently returns the holding company.
Resources
Read the method.
Then check our work.
Coverage figures, benchmark methodology, ingest latency and incident history, published in full.
Twelve guides and a data dictionary
Every guide ends with a runnable example against the sandbox key, and an edit link into the repo.
Quickstart
Key to first resolved entity in three minutes, in four languages.
5 minMatching guide
How the four scorers work, what each hint buys you, and how to pick a threshold.
18 minBatch & files
CSV in, CSV out. Job lifecycle, partial results, backfilling 40M rows.
12 minWatch & webhooks
Saved queries, signature verification, replay, dead-letter handling.
11 minPoint-in-time reads
as_of semantics, what a register did and did not publish, and the gaps.
9 minWarehouse shares
Snowflake, BigQuery, Databricks. Table layout, incremental models, dbt.
14 minMCP & agents
Wiring the tools, scoping keys, and grading groundedness in your eval.
10 minData dictionary
All 184 attributes: type, source class, coverage, null semantics.
referenceSPOT-Bench v4 methodology
The 250,000-record benchmark: construction, labelling, corruptions, and how to run the harness yourself.
22 minWorked example: sales & GTM targeting
Size a market from registers, resolve the company field at capture, and suppress accounts you already own by ID.
8 minWorked example: market research
Replay the same segmentation at each year end and get formation, dissolution and branch-opening rates that trace to a gazette.
9 minWorked example: KYB & data validation
Existence, status, legal form, VAT and LEI checked against the register of record, with the filing reference attached.
8 minSPOT-Bench v4
The benchmark, including where we lose
250,000 records drawn from real CRM exports, invoice OCR and web forms, labelled by two annotators with a third adjudicating disagreements. Published with the harness.
results · SPOT-Bench v4 · 2026-07n = 250,000
| Input class | Share | Top-1 | Top-5 | Auto-accept precision |
|---|---|---|---|---|
| Clean legal name | 31% | 99.4% | 99.8% | 99.9% |
| Name + city, no legal form | 24% | 98.1% | 99.3% | 99.6% |
| Domain only | 14% | 96.8% | 98.4% | 99.4% |
| Identifier (VAT/LEI/register) | 11% | 99.9% | 99.9% | 100.0% |
| Trade name / brand alias | 9% | 94.2% | 97.6% | 98.8% |
| OCR-corrupted line | 7% | 91.7% | 96.1% | 98.1% |
| Former name, ≥3 years stale | 4% | 88.4% | 94.9% | 97.7% |
| Weighted total | 100% | 97.3% | 98.9% | 99.4% |
Auto-accept precision is measured on the subset scoring ≥ 0.95, the default threshold. Top-1 counts the highest-scoring candidate whatever its confidence.
Run it yourself
The harness, the corruption generators and the label schema are public. The labelled set is available under a research agreement so it does not end up in a training corpus.
Calibration
Predicted confidence against observed precision on the held-out split. On the diagonal the number means what it says.
July run: predicted 0.925 across the 0.90–0.95 band, observed 0.928.
Where it degrades
Former names older than three years are our weakest class at 88.4%. Sole traders in jurisdictions without a public register are excluded from the benchmark entirely — we cannot resolve what nobody publishes, and averaging them in would flatter the number.
How the score is fitted
Per jurisdiction, on a held-out 40,000-record split, refitted monthly. A score is a probability, not a similarity: 0.90 has to be right nine times in ten or the fit is wrong and we say so in the changelog.
192 jurisdictions, three tiers
Financial depth is a separate axis from tier: it tracks what a jurisdiction obliges companies to file, which is why Swiss AGs sit at 22% and stay Tier 1.
Tier 159 jurisdictions
Ingested from the primary register, with the full event history behind every field.
Fields carriedIdentity · identifiers · officers · relations · events · financials where filed
Share of all entities71.4%
IncludesGermany, UK, France, Netherlands, Japan
Tier 276 jurisdictions
Assembled from official sources that publish identity but not a full filing stream.
Fields carriedIdentity · status · identifiers · addresses
Share of all entities26.9%
IncludesUnited States, Poland, Saudi Arabia, Nigeria
Tier 357 jurisdictions
Identity only, from official aggregates and international registers.
Fields carriedIdentity · jurisdiction · LEI where issued
Share of all entities1.7%
IncludesIraq, Angola, Tajikistan, Greenland
| Jurisdiction | Legal entities | Branches | Primary source | Filed financials | Median freshness | Tier |
|---|---|---|---|---|---|---|
| United States | 41,209,774 | 1,884,013 | 50 SoS registries · SEC EDGAR | 27 h | Tier 2 | |
| Germany | 6,412,880 | 318,204 | Handelsregister · Bundesanzeiger | 3.1 h | Tier 1 | |
| United Kingdom | 5,704,119 | 41,806 | Companies House | 1.4 h | Tier 1 | |
| France | 4,988,301 | 602,447 | INPI RNE · INSEE SIRENE | 6.8 h | Tier 1 | |
| Italy | 3,146,902 | 271,509 | Registro Imprese | 9.4 h | Tier 1 | |
| Poland | 2,905,441 | 88,120 | KRS · CEIDG | 11 h | Tier 2 | |
| Netherlands | 2,381,660 | 96,338 | KVK Handelsregister | 5.0 h | Tier 1 | |
| Switzerland | 762,455 | 54,912 | Zefix · SHAB | 2.2 h | Tier 1 |
Eight flagship markets of 192. Beyond the company registers we also ingest EUIPO TMview, WIPO, USPTO, GLEIF, VIES, BRIS, Eurostat LAU, Destatis AGS, INSEE COG, GeoNames, RDAP and Certificate Transparency — 1,340 sources in all.Download full table (CSV) →
1,340 sources, and how often each is read
Grouped by jurisdiction and by class. Cadence is the polling interval, not an average of how quickly a record happens to change.
DEHandelsregister · Unternehmensregister · Bundesanzeigerevery 15 min
CHZefix · SHAB/FOSC · BFS commune registerevery 15 min
GBCompanies House · FCA FS Registerevery 10 min
FRINPI RNE · INSEE SIRENE · BODACChourly
NLKVK Handelsregisterhourly
ITRegistro Imprese · InfoCamereevery 4 h
ESRegistro Mercantil · BORMEevery 6 h
ATFirmenbuch · Ediktsdateihourly
NORDICSBolagsverket · Brønnøysund · CVR · PRHevery 2 h
USSEC EDGAR · 50 Secretary-of-State registriesdaily 04:00 UTC
EUBRIS · VIES VAT · EBA credit institutionsdaily
GLOBALGLEIF LEI · ISO 3166 / ISO 20275daily
BRANDSEUIPO TMview · WIPO GBD · USPTO TSDRweekly
PLACESEurostat NUTS/LAU · Destatis AGS · INSEE COG · GeoNamesquarterly
WEBRDAP / WHOIS · certificate transparency · DNScontinuous
A feed that falls behind its cadence raises an alert against the expected volume for that source on that day, so a court holiday reads as a court holiday rather than an outage. Live lag per jurisdiction is on the status panel, and every ingest change is in the changelog.
Changelog
Breaking changes ship behind a dated version and the previous one runs for a year. Coverage and performance changes ship continuously.
Italy moves to Tier 1
Registro Imprese ingest now includes deposited accounts. Financial coverage for IT rises from 34% to 79%; median freshness drops from 31 h to 9.4 h.
p95 on resolve down 26%
Blocking moved to a learned key set. p95 falls from 152 ms to 112 ms with no measurable change in top-1 accuracy (97.28% → 97.31% on SPOT-Bench v4).
Saved-query enter and exit events
Watch subscriptions can now be defined by a saved search rather than an ID list, so a target universe maintains itself. Backfill of the first 90 days is included.
API version 2026-06-01
Confidence is now calibrated per jurisdiction rather than globally; alternatives is capped at 10 and sorted; the deprecated score field is removed. Prior version supported until 2027-06-01.
Postmortem: 41-minute elevated error rate in eu-central-1
A bad index build on the blocking store returned 503s for 8.2% of resolve calls between 09:14 and 09:55 UTC. Root cause, timeline and the four changes we made are on the status page.
Engineering blog
Long posts about ingest internals, written by whoever did the work.
“The same company” is a modelling problem, not a string problem
Why edit distance plateaus around 78% on real CRM data, and what a register key does that no amount of fuzzy matching can.
SPOT-Bench v4: how we measure match quality without grading our own homework
250,000 labelled records, five input corruptions, blind adjudication, and the parts where we still lose.
Ingesting the Handelsregister: 1.2M announcements a year at 3.1-hour median latency
Polling strategy, the XML that lies, and how a court's holiday schedule became a monitoring alert.
Bitemporal entities, or: what did we know and when did we know it
Separating when the world changed from when we found out, and why audit workflows need both.
Grounding agents with IDs instead of names
A 1,200-question eval showing where name-matched RAG breaks and what ID-keyed retrieval fixes.
Status
Availability is probed from four networks we do not run. Incidents get a public postmortem within five business days.
status.spotit.ai 8 of 9 operational
API · eu-central-199.994%
API · eu-west-199.981%
API · us-east-199.97%
API · ap-southeast-199.94%
Batch jobs99.972%
Watch delivery99.91%
Warehouse shares99.989%
DE ingest · Handelsregister2.4 h behind
FR ingest · INPI RNE14 h behind
INPI RNE bulk feed is running 14 h behind following an upstream schema change. Incident inc_2026_0904 · updated 40 min ago.
Security & compliance
We hold registry facts, your query logs and your entity mappings — no payment data, no end-customer PII, no credentials to your systems.
controls
| SOC 2 Type II | Audited annually; report on request |
| ISO 27001 | Certified · scope covers ingest, API and support |
| GDPR | DPA with SCCs, sub-processor list, 30-day deletion |
| EU residency | EU-only processing and log retention, no egress on failover |
| Penetration testing | Independent test twice a year, summary on request |
| Access | SSO/SAML, SCIM, scoped keys, full audit log |
on request
| SOC 2 Type II report | Under NDA, current period |
| Security questionnaire | CAIQ and SIG Lite pre-filled |
| Sub-processor list | With 30 days notice of changes |
All of it from trust.spotit.ai.
A key, a curl, a resolved entity.
That is the whole quickstart. If it does not work inside five minutes, tell us and we will fix the docs rather than explain the error.
$ export SPOTIT_KEY=spk_test_8f2c... $ curl -s https://api.spotit.ai/v1/resolve \ -H "Authorization: Bearer $SPOTIT_KEY" \ -H "Content-Type: application/json" \ -d '{"q":"nordwind logistik hamburg"}' \ | jq '{id: .match.entity_id, conf: .match.confidence, name: .entity.name}' { "id": "ent_01JR8K3F5T2QW9", "conf": 0.987, "name": "Nordwind Logistik GmbH" }