EndoAI
Menu
Download

Home / Pattern Universe

Interactive demonstration · synthetic data

The Pattern Universe.

See how dated observations can be organised into timelines, cycle spirals and co-occurrence maps, and why honest uncertainty and visible gaps matter more than a confident-looking chart.

Synthetic example data only. Every value here comes from a fixed, documented generator. No record belongs to a real person, nothing is loaded from EndoAI’s servers, and nothing you select is stored or sent. This is a demonstration, not a prediction engine or a medical analysis.

Reading patterns honestly

Six words that keep charts truthful.

Observation

Something that was recorded: a symptom logged on a date. It describes what was written down, not everything that happened.

Association

Two things that tend to appear together in the record, such as fatigue and pelvic pain on the same days. Association can be a coincidence or have a shared cause.

Uncertainty

How much a number could change with more data. With few recorded days, intervals are wide, and that is the honest answer.

Missing data

Days without a record. They are shown as gaps and never filled in. If recording is patchy, a pattern may be hidden or exaggerated.

Causation

One thing making another happen. A chart of records cannot establish causation, and nothing on this page claims it.

Clinical interpretation

What a pattern might mean for your health. That needs a healthcare professional who can consider your full history, an examination and tests.

Method

How every number is calculated

  • Middle value and spread. The median and interquartile range of pelvic-pain entries on recorded days in the window (Hyndman–Fan type 7 quantiles). Missing days are excluded, never estimated.
  • Trend. The Theil–Sen estimator: the median of the slopes between every pair of recorded days. It resists unusual values. The 95% interval is Sen’s distribution-free interval using a normal approximation, so it is approximate for small or tied samples. We show it as change per 30 days.
  • How often. For an entry recorded on k of n recorded days, a Beta(1 + k, 1 + n − k) distribution (a uniform prior) gives the 95% equal-tailed credible interval.
  • Shared days. A plain count of recorded days on which both entries appear. The layout uses spring and repulsion forces purely to arrange the picture.
  • Cycle spiral. Cycle starts are taken from recorded period days only (the first recorded period day after five or more days without one). Each ring is one recorded cycle; a full turn would be 45 days, so shorter cycles end earlier around the ring. No cycle length is assumed.
  • Data. Three synthetic timelines of 120 days from a seeded generator (version 1.1.0), described in full in the site’s source code and checked by automated tests.

EndoAI’s Insights describe your own records with rule-based trends; the Pattern Discovery Lab, in development, uses these families of method. They are explained further in the science section. None of them diagnoses, predicts or evaluates treatment.