Measure what your mystery gives away.

Compare teaser, clue, and reveal copy with the real premise. Local sentence embeddings expose semantic disclosure order; they do not grade creative quality or predict hype.

Ready. The quantized model runs in this browser; no campaign copy is sent to an inference service.

Meaning moves in vectors.

The method is inspectable. The judgment remains yours.

1. Establish the destination

The premise is the semantic reference. Every campaign stage is embedded with the same pinned model and normalized to unit length.

2. Compare direction

Cosine similarity measures whether each stage points toward the same meaning, even when its surface wording changes.

3. Inspect the lifts

A positive clue lift means the clue discloses more of the premise than the teaser. A positive reveal lift means the reveal advances again. Neither result proves the campaign is good.

Why use embeddings instead of word overlap?

Embeddings compare meaning across paraphrases. The tool also shows lexical overlap as a transparent secondary signal, because model similarity and shared words answer different questions.

What counts as an ordering failure?

If clue similarity does not exceed teaser similarity, the middle stage has not advanced semantic disclosure. If reveal similarity does not exceed clue similarity, the sequence stalls or retreats before closure.

What should a reviewer do next?

Revise the first non-positive stage, rerun the audit, and inspect the raw evidence. Then validate the creative with audience research; this tool does not replace testing.

"Sparse" describes length. "Mysterious" describes information. They are not the same.
Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.