Intent Translator

Preamble

Premium can mean ten different things. Intent Translator makes me choose which one before the model turns a vague desire into the nearest stereotype.

---

Desire Needs A Fork In The Road

Creative intent often arrives as a blur: premium, strange, calm, editorial, alive, less SaaS, more human. Those words can help a conversation. They collapse when they become the whole prompt.

The translator has to ask what I mean by the word in this context. “Premium” can mean restraint, polish, trust, scarcity, price, typography, silence, or inherited prestige. Each route produces a different artifact, so the system should make the fork visible before generation begins.

The How Is Three Readings

The first pass should return options with consequences:

premium as restraint -> fewer elements, quiet type, no gloss
premium as trust -> conservative hierarchy, evidence, calm spacing
premium as luxury -> scarcity cues, material detail, sharper exclusion

Then it turns the chosen reading into a direction object:

intent -> canonical terms -> reference anchors -> constraints -> anti-patterns -> generation notes

The translator is doing more than prompt cleanup. It is asking whether the word carried my taste or the model’s stereotype of my taste.

The First Test

The first test is adversarial. Give the translator one vague phrase and make it return three different readings with consequences.

When the options give me no traction to choose, the intent is still unearned. If the model chooses for me, the system has converted uncertainty into false conviction.

The unresolved pressure is delegation. The translator should help me decide what I mean. It should not let me outsource the meaning because choosing was hard.