Research
Test sets, methods and measurements we produced because they did not exist. Free to use, including against tools that are not ours — a benchmark that only its author can run is a brochure.
-
A Company Name Without Spaces Misses the Screening List
A company's short name finds its federal screening-list record 98% of the time with spaces and 4% without. Measured on 97 real listed entities.
-
SEC Form 4, 2026 Q2: Issuer-as-Owner Filings, Flag Spellings and Filing Timing
How often a company's EDGAR list holds other issuers' Form 4s, how flags are spelled, and how long after a transaction a Form 4 is filed: 2026 Q2 data.
-
Six Digits Against CBP's Own Rulings: Grading Our HS Classifier on 100 Published Decisions
We gave our own HS classifier the merchandise descriptions from 100 CBP binding rulings. It matched CBP at 6 digits 23.5% of the time. Every miss is published.
-
The COI Extraction Honesty Test
An open, synthetic test set for certificate-of-insurance extraction: does a tool invent a value for a blank cell, and can it tell absent from unreadable?
-
The Headings a Keyword Search of the Tariff Schedule Can Never Find
For 4-10% of products in 140 CBP binding rulings, a keyword search of the tariff schedule never finds the correct heading at any depth. Measured, with the data.
-
The U.S. Tariff Schedule's Three Rate Columns, Measured (HTS Revision 18, 2026)
USITC's tariff schedule counted column by column: lines per Special program symbol and how many at Free, Column 2 against General, the General rate mix.
-
What a Debarred Company's Name Starts With
On the U.S. federal exclusions list, 6% of companies open with the words limited liability and 3.9% with joint stock. Measured on 18,787 records.
What we will and will not publish here
Will: methods anyone can run, synthetic test data with documented ground truth, and our own measured results — including the ones that went against us. Every figure carries the date it was measured, because a number without a date is a claim about now that was true about then.
Will not: studies manufactured to attract links, comparisons that test a competitor's product, or numbers we cannot defend. We sell software in this space, so the only version of research that is worth anything from us is the version a skeptical reader can re-run themselves.
Aervik Labs is operated by EAC Ventures, LLC. See About for who is accountable and how corrections work.