Name and pronoun hold: the only identity test that matters

Why a tag is not an identity system

inRole’s trans category is a useful example of catalog logic: a wall of tagged characters plus tools to make your own. That tells you the marketplace organizes people-shaped bots the way it organizes any other genre. It does not tell you whether “she/her” survives a topic change, who assigned the tag, or whether the creator wrote a backstory that actually holds.

Talkie’s ranking card for this query opened on a sexualized shopping scene. SpicyChat’s ranking card opened on a childhood-friend frame with NSFW tags. Those are creator choices, not identity systems. If you wanted a continuing conversation and got a costume, the mismatch is the result page, not your request.

The hold test, in four moves

Set name, pronouns, and one fact. Ask for an echo. Change the subject for four turns. Come back to a body or clothing question that would expose a slip — “what did I ask you to call me?” — without repeating the pronouns yourself.

Pass: the bot uses the name and pronouns without a prompt. Soft fail: it hedges (“whatever you prefer”) after you already chose. Hard fail: it reverts to the catalog gender, deadnames you, or describes the wrong body.

Do not coach the model

If you correct the bot every other line, you are writing the scene, not testing it. One correction is allowed. A second correction is a fail. Paid plans that advertise “better memory” are not a substitute for a free-surface hold. Memory that only works after you pay is a sales story until the unpaid chat already proves it.