Bypassify
Transparency · Risk Meter v1.0.0

What the Risk Meter actually measures

The Risk Meter is an internal heuristic built by Bypassify. It reads patterns in your own text — cadence, repetition, structure, vocabulary — and reports how machine-like those patterns look. It is deterministic: the same text always produces the same score.

No writing tool can guarantee a specific result from external AI detection systems. The Risk Meter is an internal heuristic that measures patterns in your text — treat it as an indicator, not a guarantee.

Structure and cadence

  • Sentence uniformity — how alike your sentences are in length.
  • Sentence-length variance — the spread between your shortest and longest sentences.
  • Rhythm uniformity — whether cadence changes as the piece moves along.
  • Paragraph uniformity — whether every block is the same size.

Language patterns

  • Repeated sentence openers — how often sentences start the same way.
  • Vocabulary repetition — words and phrases recurring more than natural writing repeats them.
  • Formulaic transitions — stock connectives that cluster in generated prose.
  • Punctuation variety — reliance on a single punctuation rhythm.
  • Lexical richness — vocabulary breadth relative to document length.

How the signals are weighted

Risk Meter Calibration v1 — effective 2026-07-29. Initial deterministic weighting derived from the BCE analysis layer. No external detector data has been used to fit these weights. Signals that a document is too short to support are switched off, and the remaining weights are re-normalised so short text is never penalised for missing evidence.

sentence uniformity
18%
length variance
15%
rhythm uniformity
12%
opener repetition
10%
vocabulary repetition
12%
transition repetition
15%
structural consistency
8%
punctuation uniformity
5%
lexical richness
5%

What confidence means

Confidence describes how much internal evidence we had — sample length, how many signals were active, and whether those signals agreed. High confidence means the reading is well supported by your text. It is never a probability of passing or failing any external tool.

What we don't do

  • We do not send your text to any third-party AI detection service.
  • We do not store detector results, because we never request any.
  • We do not use an external language model to produce this score.
  • We do not promise or guarantee any specific outcome elsewhere.

Try it on your own text

This runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.