W-Enterprises/Trademark clearance
Clearance Studio / screening and generation
A conflict is where every search stops. It is where this one starts.
A clearance search tells your client their mark is ninety percent similar to an earlier registration, and ends. What the client wanted was to file. This screens a proposed mark on four axes, and when it finds a conflict it generates alternative marks, re-screens each one against every registration scored, and returns the ones that clear while still reading as the same brand.
Checking the demo
The row a fused average throws away
This is a real result from the demo corpus, not an illustration. The applicant's mark and this registration share an identical wordmark and look nothing alike. Appearance scores zero. The fused average comes to 0.31, which sits comfortably inside the clear band.
Every system that reports one blended similarity number files this away as clear. An examiner refuses it.
Goods overlap gates how hard a breach can push. These two sit in different Nice classes, so the mark is held at review rather than refusal. Same wordmark inside one class and it is a refusal.
What the applicant actually filed
The same mark, screened against the whole corpus, hits a registration it genuinely collides with. Appearance and Wordmark both breach their ceilings at once.
Then it does the half nobody built
Moving away from the mark that blocked you is the easy half, and a designer does it by eye. Not landing on a different registration nobody had looked at is the half that cannot be done by eye. Every option below was generated, then re-screened against all sixty registrations, and each change is named rather than redrawn.
Every figure on this page is from one recorded run:
the demo corpus of 60 synthetic marks, 19 August 2026,
config config.yaml sha256 1818031a7679,
source sha256 90ca5f30ac5d. That second hash covers
every module in the package, reporting code included, so it moves more often
than the figures do. The run is pinned to
data/landing-pin.json in the source tree and this page is
generated from it, so every number here can be checked against the artefact
that produced it rather than taken on trust. The scores are deterministic and
reproduce exactly. The timing is not: it is the median of
7 runs on one laptop, spanning
0.4 to 0.8 s,
and will differ on your hardware.
The options that did not clear are the useful ones. AQUNIA kept more of the original brand than anything else here, 0.56, and that is exactly why it failed: staying close enough to be recognisable left it close enough to collide. Re-screening put it at review against PALOMAR on appearance and device — a registration nobody in the matter had looked at. 4 of 5 were held back rather than presented.
Of the 1 that cleared, AQUIQ does both at once, clearing furthest at 0.45 and keeping the most brand at 0.44, which is the outcome to want and not the one to expect. Retention is printed beside every score rather than averaged into it, because one number cannot carry both.
Every change is named
A clearance option is only useful if you can defend it. Each one is a diff on a specification, so the reasoning survives into an opposition and reproduces six months later.
| Axis | Change applied to AQUIQ | Effect |
|---|---|---|
| Orthography | Stem AQU retained, new termination -IQ | — |
| Device | Figurative element replaced with a diamond (angular family, was round) | — |
| Palette | Livery moved to magenta, 46 dE from the cited mark's palette | — |
| Re-screen | Scored against all 60 registrations | 0.45 |
What this is, and what it is not
It is a screening aid that produces a reproducible number and a written reason a paralegal can put in front of a client. It is not a register search, and it does not sign an opinion. The public demo runs against a corpus of sixty synthetically generated marks, clearly labelled as such on every report, so no fabricated application number can ever be mistaken for a real prior right.
The wordmark axis compares Latin script. Goods comparison is Nice class overlap, not the specification text. Register status, filing dates and opposition history are not modelled. Every report states these limits on its own face rather than in a footnote.
Where it earns its place
- IP firms and trademark attorneys who currently deliver a refusal and a dead end. This turns the worst moment in the matter into a deliverable the client pays for.
- Branding and design agencies who find out about a conflict after the identity is signed off, and absorb the rework.
- In-house brand teams running a shortlist of candidate marks before committing budget to one.
- Filing platforms wanting to offer alternatives rather than only bad news, through the API rather than the interface.
Common questions
Does this replace a clearance search?
No, and it does not claim to. It produces a reproducible score and a written reason; a qualified attorney still signs the opinion. The demo screens against a synthetic corpus, not a register, and says so on the report itself.
How is it different from the search platforms we already pay for?
They tell you a mark conflicts and stop. Name generators invent words and check them. Neither closes the loop for device and figurative marks, which is this: a blocked logo in, modified artwork out, re-screened against everything scored, with each change enumerated.
Why not one similarity score?
Because the mark at the top of this page fuses to 0.31 and an examiner would refuse it in the same class. One blended number hides the single decisive axis. Each axis carries its own ceiling instead, and colour has none, because nobody is refused a mark for using blue.
What happens to artwork we upload?
Nothing is kept. Each request runs in its own temporary directory which is deleted before the response is sent, and the report comes back with its images inlined so there is nothing left to leak. There is an automated test asserting exactly that.
Can we run it against our own data?
That is the real engagement. The demo corpus exists so the mechanism can be inspected without anyone's confidential filings in it. Loading a firm's own corpus, and the registry coverage question that comes with it, is scoped per firm.
Is it reproducible?
Yes, deliberately, because a report you cannot reproduce is a report you cannot defend. Same mark, same numbers. Options come from a fixed seed. Every constant that decides an outcome lives in one file, fingerprinted into the report so a score quoted today can be re-derived later.
Request access to the live demo
The demo is opened per person rather than left standing open. Tell me who you are and I will send an access code, usually the same day. There is no charge and nothing to install.
Nothing you upload to the demo is stored. Each request runs in its own temporary directory which is deleted before the response is sent, and the report comes back with its images inlined, so there is nothing left on disk. The demo screens against a synthetic corpus, never a real register.
Screen a mark yourself
The demo takes a logo, a wordmark and a Nice class, and gives you the conflicts in about a second. Generating alternatives takes roughly half a minute. Nothing installed, nothing retained, and access is opened on request rather than left standing open.