Your context layer.
Your devices are good at recording what happened. They can tell you how long you slept, what your heart rate was, how far you ran. What they cannot tell you is how any of it felt. That gap is where the most useful questions live, and it is what the context layer is built to fill.
The check-in
Five points, once a day: sleep quality, energy, mood, stress, and soreness, each rated on a scale of one to ten. It takes under 15 seconds. The individual ratings matter less than the combination. A day where sleep scores well but energy scores poorly is more interesting than either number alone. A week where soreness stays high while training load drops is worth looking at. The check-in gives you a daily subjective read that no device can produce.
Notes
Add a note after a hard session, a difficult week, a change in medication, a run that felt worse than the numbers suggested. Even a sentence is enough. Anything that might explain a discrepancy between how you felt and what your devices recorded is worth capturing. Life happens around the data. Notes are how you capture it.
Some examples of what is worth recording:
You do not need to write much. You need to write consistently.
The cross-check
This is where the context layer earns its place. Your HRV is low but you feel great. Your sleep data looks fine but you rated it a four. You ran a personal best on a day your recovery metrics said you should have been flat. These discrepancies are not noise. They are the signal. Cross-checking how you felt against what your devices recorded is where the most interesting questions come from, and it is the kind of question an AI tool can actually do something useful with.
What changes over time
A single check-in tells you very little. Two weeks of consistent check-ins starts to tell a story. Three months reveals patterns that no single appointment or weekly summary would ever surface: how your mood tracks with training load, how your energy shifts across a cycle, what recovery actually looks like for you specifically rather than for the average person in a research study.
The context layer is not useful immediately. It becomes useful over time, and then it becomes something you would not want to compile without.