Signa screening is a signal engine that surfaces and ranks potential trademark conflicts to focus attorney review. It is not legal advice and never substitutes for a qualified trademark attorney’s judgment. The screen maintains a deliberate boundary between what it observes from the public register and what only a legal analysis can conclude.
What the screen assesses vs what it never asserts
The screen reports observable signals and the evidence behind them. A human attorney makes the legal determination.
These three distinctions are the honesty backbone of the screen. A signal is not a legal
conclusion.
- Fame signal is not legal fame. A high fame signal reflects observable register and portfolio patterns, including breadth, registration density, and recognized marks in our data. It is not a legal finding of fame or notoriety, which only a tribunal or examiner can determine.
- Portfolio pattern is not a legal family of marks. A detected portfolio or “family” pattern reflects a shared dominant element across marks under continuous common ownership in our data. It is not a legal family-of-marks claim, which requires proof of promotion and public recognition of the family feature.
clear is not registrable. A clear verdict means no conflicts were flagged across the searched and covered scope. It is not an opinion that the mark is registrable or available. Registrability depends on absolute grounds, examiner judgment, unindexed rights, common-law use, and factors outside the register.
| What the screen assesses | What it never asserts |
|---|
| String similarity and shared distinctive tokens | Likelihood of confusion as a legal conclusion |
| Nice class overlap and related classes | Legal fame or notoriety |
| Seniority and priority dates | A legal family of marks |
| Territorial breadth | Registrability or availability |
| Fame and portfolio signals | Freedom to operate |
DuPont factor mapping
Each flagged conflict carries a factor ledger. The ledger maps observable evidence onto the thirteen DuPont factors, the US likelihood-of-confusion factors stated in In re E.I. DuPont DeNemours & Co.
Every one of the 13 factors is represented in the ledger. A factor that cannot be observed from the public register is explicitly marked not_available_from_register instead of being silently omitted. This prevents an unavailable factor from appearing to be a finding of no conflict on that factor.
Each factor entry carries:
- A factor ID
- An assessment from the closed vocabulary
supports_conflict, weighs_against, neutral, not_available_from_register, or not_applicable
- A confidence
- Raw observable evidence, such as exact dates with a date kind of priority, filing, or registration; string edit distance; a shared token with its IDF weight; territory counts; or a fame basis with an as-of date
- Stated limitations
| Factor | Register posture | Evidence available to Signa |
|---|
| 1. Similarity of the marks in appearance, sound, connotation, and commercial impression | Observable | String and phonetic similarity, plus shared distinctive tokens |
| 2. Similarity and nature of the goods or services | Observable | Nice class overlap and related-class expansion |
| 3. Similarity of established trade channels | Partially observable, often not available from the register | Register descriptions may provide limited indications, but established trade channels are often unavailable |
| 4. Conditions of purchase, including impulse vs careful purchasing and sophisticated buyers | Not available from the register | Marked not_available_from_register |
| 5. Fame of the prior mark | Signal only | A fame signal may reflect register and portfolio patterns, but it is not legal fame. Underlying legal fame is not available from the register |
| 6. Number and nature of similar marks on similar goods, including a crowded field | Observable | Crowded-field aggregation over the register |
| 7. Nature and extent of any actual confusion | Not available from the register | Marked not_available_from_register |
| 8. Length of time and conditions of concurrent use without actual confusion | Partially observable | Coexistence and seniority dates may be observed. Actual-confusion history is not available from the register |
| 9. Variety of goods on which a mark is used, including a house mark or family | Signal only | A portfolio pattern may be detected, but it is not a legal family of marks |
| 10. Market interface between applicant and owner, including consent, assignment, and laches | Partially observable | Assignment records may be observable. Consent and laches are not available from the register |
| 11. Extent to which the applicant has a right to exclude others | Not available from the register | Marked not_available_from_register |
| 12. Extent of potential confusion, from de minimis to substantial | Not available from the register | Marked not_available_from_register |
| 13. Any other established probative fact | Observable when present | Captured as available raw observables, with their source and limitations |
Determinism and replay posture
The screening pipeline is deterministic. An offline job produces versioned, content-hashed assets, and the request path performs a pure runtime lookup over those assets. No machine learning model or LLM runs in the request path. The same inputs evaluated against the same asset versions produce the same output.
Every screening response carries a version receipt containing:
- Rules version
- Scorer version
- Feature-semantics version
- Asset versions
- Lane-plan version
- Factor and evidence catalog version
- Per-office corpus watermarks
A deterministic screening_id receipt identifies the decision. Replay means recomputing decision identity from the recorded inputs and the recorded version receipt. It does not mean retrieving a stored snapshot of the response.
This run does not store a snapshot of each response. Replay reproduces a decision by re-running
the recorded inputs against the recorded, content-hashed asset versions. Because corpus watermarks
advance over time, exact byte-for-byte replay is guaranteed only while the same asset and corpus
versions are pinned. The version receipt records exactly which versions were in effect, so any
divergence is attributable.
Professional review remains required
Screening focuses and accelerates professional review. It does not replace it. Consult a qualified trademark attorney for any legal conclusion about registrability, likelihood of confusion, or freedom to operate. 
