This is what it does
The Brand Voice Lock and the score, running here
Two mechanisms from the product’s own code, in this page: the post-pass that strips your banned phrases out of a draft even if a model put them back, and the 0–100 personalization score computed as arithmetic over the prospect’s own material.
The rewrite is not here. Mirroring a prospect’s voice is a language-model call, so this page does not write your email for you and does not pretend to. Everything below runs on the text in the boxes.
The boxes start with an example so there is something to look at. Change any of them and press the button: nothing you type leaves this page.
What the two checks said
Result
Where these rules come from
Two functions, ported, not invented
- Brand Voice Lock: the six default banned phrases and the post-pass are ported from the product’s
enforceBrandVoice. It is case-insensitive, it reports what it removed, and it collapses the gap the removal leaves. - The score: ported from
calculatePersonalizationScore— 20 points when a hook’s first 20 characters appear, 10 per personality marker up to four, 10 for one of their repeated phrases, and 10 if a hook lands in the opening 120 characters. - What is missing is the product, not the demo: the voice profile itself is extracted by a model, and the rewrite is produced by a model. Neither runs here, and neither is faked.