Anyone can give health advice. Almost no one checks, out loud, whether it worked.
Vegetablah makes a promise most tools avoid: every week, for every metric it tracks, your twin writes down what it expects to happen next — and the following week it grades itself in public against what actually happened. Right or wrong, the score stays on the record.
Each week the twin commits to a specific expectation for each metric — a number, not a vibe.
Your real data comes in over the week. Nothing gets quietly revised after the fact.
Next week the prediction is scored against the actual result, and the hit-or-miss is shown to you plainly.
Because a model you can't check is just a confident voice. When predictions are scored where you can see them, two things happen: the twin has to earn your trust instead of assuming it, and you get an honest read on which changes are actually moving your numbers.
The twin says "during this week, your resting heart rate dropped" — never "we caused it to drop." It reports what changed alongside what you did, and lets you draw the conclusion. Association, clearly labeled, beats a causal claim it can't back up.
Soon you'll be able to run consent-based n-of-1 experiments — one change at a time, self-graded the same honest way: "during the experiment, X changed," not "the experiment proved X." Your data, your test, your call on what it means.
Predictions and experiments are for personal insight and goal-setting. Vegetablah is a wellness tool — it doesn't diagnose, treat, or replace your clinician.