InnerMatch now handles continuous numeric data natively. Assessment scores, mastery percentages, biomarkers, confidence values — any measurement that varies smoothly is now a first-class input, with no API changes required.
June 14, 2026 · Behrang Mehrparvar
We are releasing InnerMatch version 1.1 today. The headline capability is continuous vector support — a meaningful expansion of what kinds of data InnerMatch can accurately match on. If you have ever wanted to feed the engine assessment scores, skill proficiency levels, probabilities, or any other smoothly varying measurement, this release is for you.
InnerMatch has always excelled at matching on discrete features: skills you either have or do not have, preferences expressed as categories, ratings on a fixed scale. These are the natural language of job descriptions, product catalogues, and user profiles.
But many real-world datasets do not speak in binary. They speak in degrees. A candidate's Python proficiency might be expressed as a test score. A learner's mastery of a topic might be a percentage. A product's relevance signal might be a probability. In each case, what matters is not just whether two values are similar, but how similar they are — and the penalty for being far apart should be proportional to that distance.
Previous versions of InnerMatch did not make this distinction: a score of 0.85 and a score of 0.10 were treated in the same structural way as a "yes" and a "no." This worked well for binary and categorical data, but produced degraded ranking quality when the inputs were genuinely continuous. Version 1.1 fixes this.
The engine now detects automatically whether the data you send is discrete or continuous. You do not need to set a flag, change your API calls, or restructure your feature vectors. If your data contains fractional values, InnerMatch switches to a proximity-aware comparison for that entire request — rewarding closeness, and penalising features that are far apart.
The practical result is that matches are now ranked by how well the values align, not just by whether they overlap. Two candidates with nearly identical assessment profiles will score highly against each other. Two who share a category label but differ sharply on every measured dimension will not.
This update is most directly valuable for teams working in:
Existing integrations built on discrete features are completely unaffected. The automatic detection means InnerMatch selects the right approach for each request independently — binary data continues to use the same engine it always has.
For existing customers, this update is invisible from an API perspective. There are no new endpoints, no new parameters, and no migration steps. If you start sending continuous data, the engine handles it correctly. If you continue sending discrete data, nothing changes. The upgrade is backwards-compatible in every respect.
New customers whose use cases naturally involve scored or measured features can now integrate InnerMatch directly, without needing to discretise or bin their data as a preprocessing step.
InnerMatch 1.1 is live now on the AWS Marketplace. All subscribers are automatically on the updated version. The interactive demo at synaptosearch.com/innermatch_demo has also been updated and will reflect the new behaviour when continuous-valued feature vectors are submitted.
If you have a use case involving continuous measurements and would like to discuss whether InnerMatch 1.1 is a fit, reach out — we are happy to walk through it.