zen.mzarras.com

Live blood glucose, last 48 hours, with the zen-predict forward projection

  • Glucose · 48h
  • Zen · +2h
  • Target 70–180
  • Low / high
mg/dl

Loading…

Glucose data table
Blood glucose readings over the charted window, sampled hourly.
Timemg/dLRange

Live forecast

Predicted blood glucose in mg/dL from the most recent cycle, zen-predict versus oref.
Horizon Zen oref Difference
Loading…

Scoreboard

Mean absolute error in mg/dL by horizon: zen-predict versus the naive persistence baseline, plus head-to-head win rate against oref.
Horizon Zen MAE Naive MAE Edge Beats oref
Loading…

Lower is better — mean absolute error, in mg/dL, scored on data the model had never seen when it made the call.

What is this

The software that doses my insulin treats everyone roughly the same. Zen Predict is a model that learned me — my meals, my workouts, my patterns — and on days of data it had never seen, it predicted my blood sugar meaningfully better than the naive guess that it stays where it is. This page is that model running forward in public: every prediction is published before the outcome is known, then scored when the outcome arrives.

It doses nothing. My pump and its closed loop keep making every insulin decision; the model only predicts. Nothing here is medical advice, and nothing here is a treatment recommendation — it is one person’s forecast being graded in the open.

One fairness note. My pump’s projections are cautious planning sketches — some deliberately pessimistic, because that is the safe way to dose — not its best honest guess. So the claim I stand behind is the one against the naive baseline (“persistence”: your glucose in two hours is exactly what it is right now), which is embarrassingly hard to beat. The win over oref’s projections comes with that footnote attached.