A score, and the reasons it came out that way.
Phonetics, spelling distance, shared words and Nice-class relatedness, each carrying the points it contributed. The borderline band gets a second reading before it reaches you. You see every signal, and so does your attorney.
Check your mark freeThe problem
A keyword search finds the wrong things, and misses the rest
Registry search matches strings. The filings that matter sound like your mark, are spelled slightly differently, or sit in a class that overlaps yours — NORTHWYND TALY against NORTHWIND TALLY. Search either floods you with everything containing one of your words, or finds nothing at all.
How it works
How similarity you can audit works
- 1
Candidates, not everything
Phonetic keys and trigram similarity narrow the day’s filings to the ones worth scoring at all.
- 2
Five signals, added up
Double Metaphone and Kölner Phonetik, Levenshtein and Jaro-Winkler, token overlap, and a hand-written class-relatedness table.
- 3
The borderline band gets read again
Scores in the middle go to a second reading that describes how two marks read — and is refused if it tries to say anything more.
Why it matters
What you get
Phonetic, not textual
Marks that sound alike score alike, in English and German phonetics, however they are spelled.
Your own threshold
A sensitivity slider per watch decides what is worth an email. Move it once and every screen and notice follows.
Trained by your corrections
“Not relevant” on a filing is recorded and feeds the labelled set the scorer is tested against.
FAQ
Questions, answered
How many false positives does it produce?
On our own hand-labelled set of 111 mark pairs the false-positive rate is 5.6% at the default sensitivity. That set lives in the repository and the scorer is re-tested against it on every change, so a regression breaks the build rather than your week.
Does an AI decide what I see?
No. The arithmetic decides. A model is asked about the narrow middle band — how similar two marks read — and its answer is discarded if it is malformed or reads as advice. Everything else is scored by code you can inspect the reasons from.
Can I see why something was sent?
Every filing in a notice, a report or the app lists the signals and the points each one contributed. The numbers add up to the score exactly.
Features
More features
The deadline clock
Every filing we show you carries the day its window closes and the days left, computed from the publication date the register itself records.
Learn moreOne email a week
One email per week, soonest closing window first, with only the filings that passed the threshold you set — and nothing at all when there are none.
Learn moreThe register, read every day
A daily pass over the US register that is versioned, replayable and checked for record-count anomalies, so a quiet failure cannot pass for a quiet week.
Learn moreFind out whether a clock is already running.
One mark, its classes, about thirty seconds. No account and no card — and the report gets its own address you can forward to your attorney.