A pattern score is not a person's feelings.
Method disclosure updated September 7, 2026
What the free preview calculates
The preview uses fixed rules, not a model prediction of attraction. It starts at 50 and adjusts for message-count ratio, average word-count ratio, questions, selected emoji or casual wording, and selected hedging phrases. The result is rounded and limited to 8–94. Those limits are interface choices, not a validated accuracy range.
The score has not been calibrated against known relationship outcomes. A score of 72 does not mean a 72% chance of interest. It is not a diagnosis, compatibility test, safety assessment or instruction to continue a relationship.
What can go wrong
- Speaker labels matter. Use “Me:” for your messages and one distinct label for the other person. Unlabeled input can be divided into two halves as a fallback; that is not a reliable speaker attribution.
- Word length is estimated using spaces, so it is not comparable across all languages. Some phrase rules are English-specific. Sarcasm, culture, disability and communication style can change the meaning of a pattern.
- The excerpt can omit earlier messages or offline behavior. The rules do not infer timestamps, motives, manipulation or future actions.
Coverage is not confidence in intent
“High” coverage requires at least 80% labeled lines, 12 messages and 4 messages from each side. “Medium” requires 60%, 6 and 2 respectively. Other samples are “Low.” These are input-coverage tiers only, not statistically validated confidence or a guarantee the interpretation is correct.
Where AI is used
The full report's flags and reply suggestions are generated separately by an AI provider, Moonshot AI (Kimi), reached at its api.moonshot.cn endpoint. Long conversations may be shortened by retaining the start and end, omitting the middle. The model can make mistakes; check its wording against the actual conversation. Fictional examples are demonstrations, not customer results or model-accuracy tests.
Use the output as a prompt for reflection and respectful communication. No score should override consent or personal safety.
Processing and control
The free preview uses deterministic code; the full report sends conversation text and context to an AI provider. A purchased report may remain in this browser for restoration. “No account database” does not mean no third-party processing or instant deletion everywhere. Read the privacy policy before submitting. Only use your own communication or material you have permission to analyze.
Changes and feedback
This revision documents the existing score and coverage rules and replaces intent claims with descriptions of the observed excerpt. It does not introduce new validation evidence. Publisher: ReadBetween. For product or report issues, see support; never post private chat excerpts publicly.