A comprehensive analysis of 48 days of Garmin health and fitness data. This page explores daily recovery metrics, workout patterns, and the statistical relationships that emerge across sleep, stress, and training load.
Track how resting heart rate, HRV, sleep, and other recovery signals evolved across the archive period.
Overview of workout types, durations, and performance across the 48-day period.
Beyond daily numbers — what relationships and structures emerge from 48 days of continuous monitoring?
Pearson correlation measures whether two metrics tend to move together (+1), in opposite directions (−1), or have no linear relationship (0).
See it in the data: the two most correlated metrics plotted day-by-day. When they track each other closely, the correlation is working. When they diverge, something changed.
An AR(1) coefficient measures how much today's value carries into tomorrow. Near 1.0 = very sticky; near 0 = each day is independent.
Daily change magnitude: how much does each metric actually jump day-to-day? High persistence = small jumps. Low persistence = large swings.
Each day's body battery is classified into a state: Low (<30), Moderate (30–54), Good (55–74), High (≥75). The chart below shows how recovery state shifted day-to-day — and the transition matrix shows the probability of moving between states.
Averages hide variability. The chart shows each metric's full range (whiskers) and interquartile spread (solid bar), with CV measuring relative volatility.
The actual daily values: see where each metric lives day-to-day. The horizontal lines show the median — when data hugs the line, it's stable. When it scatters, it's volatile.
Comparing next-day metrics on workout days vs rest days to estimate the short-term recovery impact of training.
Day-by-day reality check: the timeline below highlights workout days (green) vs rest days. Each dot is the next-day metric value — so you can see the recovery pattern with your own eyes, not just averages.
Synthesized observations from the 5-agent analysis pipeline, computed from the archived data.
Query the archived data. The AI agent analyzes your question in 3 diagnostic layers — this can take 30–90 seconds.