Couple Name Lab

Engine regression checks

Engine version 1.0 · Research Desk · Last updated August 2026

Eight fixed name pairs run through the blending engine, with the top three candidates and their real scores recorded for each. The table below is generated by calling the engine during the site build, so it always shows what the tool actually does today rather than what it did when somebody last updated a page by hand.

The point of this page is boring on purpose. The engine is deterministic, so these results should not change unless the scoring changes. If they do change, that is either a fix we made deliberately or a regression we need to know about, and either way this page is where it becomes visible.

All rows use the couple context. Context affects only the ideal-length penalty, so the candidate set is the same across contexts and only the ordering shifts slightly.

Taylor + Travis

CandidateBuilt fromTaylorTravisBalanceJoin
Travylortrav + ylor67%67%evenvy
Taylvistayl + vis67%50%closelv
Travlortrav + lor50%67%closevl

Alex + Jordan

CandidateBuilt fromAlexJordanBalanceJoin
Alerdanale + rdan75%67%evener
Jordalexjorda + lex75%83%evenal
Jordlexjord + lex75%67%evendl

Sofia + Mateo

CandidateBuilt fromSofiaMateoBalanceJoin
Matfiamat + fia60%60%eventf
Softeosof + teo60%60%evenft
Mafiama + fia60%40%closeaf

Percy + Annabeth

CandidateBuilt fromPercyAnnabethBalanceJoin
Annarcyanna + rcy60%50%evenar
Perbethper + beth60%50%evenrb
Perabethper + abeth60%63%evenra

Brad + Angelina

CandidateBuilt fromBradAngelinaBalanceJoin
Bralinabra + lina75%50%closeal
Braelinabra + elina75%63%closeae
Bragelinabra + gelina75%75%evenag

Ben + Jennifer

CandidateBuilt fromBenJenniferBalanceJoin
Beniferbe + nifer67%63%evenen
Benniferbe + nnifer67%75%evenen
Beiferbe + ifer67%50%closeei

Kim + Kanye

CandidateBuilt fromKimKanyeBalanceJoin
Kinyeki + nye67%60%evenin
Kiyeki + ye67%40%uneveniy
Kanmkan + m33%60%unevennm

Ana + Bo

No candidates. At least one of these names is too short to split without losing it entirely, which is covered in the limits report.

What this page is not

It is not evidence that the top-ranked candidate is a good name. It shows that the engine is consistent, which is a much smaller claim. Two of the rows above make the difference obvious: for Taylor and Travis, the engine ranks Traylor and Travlor above Tayvis, and the world uses Tayvis. Consistency and correctness are different properties, and only one of them can be tested by software.

The human review protocol is where the other question gets asked, and it needs people rather than a build step.