Where the ranking should be overridden

Five situations where the structure report is measuring the right thing and still gives you the wrong answer. Each one is reproduced from the live engine below, so you can check them yourself in the tool rather than taking our word for it.

1. The top-ranked candidate is not the one people use

Run Taylor and Travis through the blender. Tayvis, the blend that is actually in circulation, comes out at position 5. Above it sit Travylor and Taylvis and Travlor and Traylor, because each keeps more total characters than Tayvis does.

Nothing is broken here. Coverage is weighted above balance, so a blend that preserves more letters overall wins on the score even when it splits less evenly. Real people chose Tayvis because it is easy to say, and the engine cannot hear. This is the clearest example on the site of why the report avoids a best-name claim.

2. Very short names

A name of two or three characters has almost nothing to give. Split it and either the blend loses it entirely or it contributes a single letter no reader can trace back.

Ana and Bo return no candidates at all in the current engine, because every split falls below the minimum blend length.

When one name is very short, a portmanteau is usually the wrong format. A paired label such as Ana and Bo, or Ana & Bo, keeps both names intact and readable, and that is what most people settle on anyway.

3. Names that share letters

When two names overlap heavily, the trace becomes ambiguous. A reader cannot tell which name a shared fragment came from, so the coverage figure overstates how much information the blend actually carries.

Ellie and Elliot share their first four characters. The engine returns:

• Elliollie, built from ellio plus llie
• Elliolie, built from ellio plus lie
• Ellioie, built from ellio plus ie
• Ellllie, built from ell plus llie

Several of those are close to one of the source names on their own. The report counts characters correctly and still cannot flag the real problem, which is that the blend does not look like a blend.

4. The recognisable part is not the biggest part

Coverage counts characters. Recognition does not work that way. Jordan splits at 2, 3, 4, 5, so it can contribute Jor, rdan, dan, an or n. The engine builds Aledan from Alex and Jordan with 50% coverage of Jordan, and that fragment is dan. Jor is the part a reader is far more likely to recover, and it is not longer.

There is no fix for this inside the engine. Recognisability depends on which part of a name carries its identity, and that varies by name, by language and by who is reading. The blind name test exists precisely because this cannot be computed.

5. Non-Latin scripts and accented characters

The engine normalises input and works on letters, so accented characters pass through and are counted. What it does not do is understand syllables in scripts that do not mark them with Latin vowels. For names in Arabic, Devanagari, Han, Hangul or Cyrillic script, the vowel-group rule is either wrong or meaningless, and the split points it finds are not syllable boundaries in any real sense.

If you are blending names in those scripts, treat the output as a starting point only. The coverage numbers are still literally true about characters, and they are not telling you what they tell you for a Latin-script name.

6. The blend happens to spell something else

Two names can combine into an existing word, a brand, or a term that means something unfortunate in another language. The engine has no dictionary and no cultural knowledge, so it will rank such a blend on structure alone and tell you nothing is wrong.

This is the failure with the highest cost and the cheapest fix. Search the candidate before you commit to it. Step three of the review guide exists for this and takes about thirty seconds.

How to use this page

If your pair matches one of these shapes, read the ranking as information rather than as a recommendation, and lean harder on the human checks. The methodology page explains what each score measures, and the two tool pages give you ways to test the parts the engine cannot reach.